Method, device and system for predicting nitrogen oxides in catalytic cracking regeneration flue gas

Abstract

The application provides a catalytic cracking regeneration flue gas nitrogen oxide prediction method, device and system, and belongs to the chemical industry field.The prediction method comprises the following steps: obtaining data and performing pretreatment to obtain a processed sample set; analyzing key variables closely related to the regeneration flue gas nitrogen oxide, and constructing a catalytic cracking regeneration flue gas nitrogen oxide initial prediction model according to the sample set; utilizing a particle swarm optimization algorithm based on flight information to optimize the catalytic cracking regeneration flue gas nitrogen oxide initial prediction model, obtaining a catalytic cracking regeneration flue gas nitrogen oxide target prediction model, using the target prediction model to predict the pretreated to-be-tested data, and obtaining a nitrogen oxide prediction value.Through the establishment of the nitrogen oxide prediction model based on the fuzzy neural network, the particle swarm optimization algorithm based on the flight information is used to optimize the initial prediction model, the prediction accuracy of the model is improved, and finally, the model can quickly perform prediction, which is beneficial to improving the nitrogen oxide prediction accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the chemical industry, and in particular to a catalytic cracking regeneration flue gas nitrogen oxide prediction method, a catalytic cracking regeneration flue gas nitrogen oxide prediction device, and a catalytic cracking regeneration flue gas nitrogen oxide prediction system. BACKGROUND

[0002] The catalytic cracking device regeneration flue gas is the largest air pollution source of the catalytic cracking device, mainly containing harmful substances such as carbon monoxide (CO), sulfur dioxide (SO2), nitrogen oxides (NOx) and dust, especially NOx. x ) and dust, especially NOx x When the flue gas is discharged into the air, it will cause serious pollution to the atmosphere. Due to the large fluctuation range of the pollutant concentration of the flue gas at the outlet of the catalytic cracking regenerator with the working condition adjustment, the concentration value of the nitrogen oxides in the outlet regeneration flue gas cannot be accurately and timely predicted at present. At the same time, the desulfurization and denitrification device of the catalytic cracking is affected by the fluctuation of the inlet flue gas, and the desulfurization and denitrification agent cannot be accurately added, causing the response lag of the parameter adjustment of the treatment facility, and the flue gas discharge is difficult to be in a stable state, and the waste gas discharge has the risk of exceeding the standard.

[0003] According to the different generation mechanisms, NO x can be divided into thermal type, fuel type and fast type. The generation mechanism of the thermal type is that the nitrogen (N2) of the air is oxidized to form NO x at high temperature. The generation of the thermal type is directly related to the combustion temperature, oxygen concentration and the residence time of the flue gas in the high temperature zone. Generally, when the combustion temperature is lower than 1500 degrees Celsius, the generation amount of the thermal type NO x is very small. The NO x oxidized and generated from the nitrogen compounds in the fuel in the combustion is called fuel type NO x . Since the carbon hydrogen compounds in the fuel volatiles are decomposed to generate CH free radicals at high temperature, and react with N2 in the air to generate HCN and N, and further generate NO x with O2 at an extremely fast speed, it is called fast type. The temperature of the catalyst regeneration in the catalytic cracking device regenerator is generally about 700 degrees Celsius, and the proportion of the thermal type NO x is low, especially in the completely combusted regenerator. The NO x in the catalytic cracking flue gas mainly comes from the nitrogen in the coke, that is, the fuel type NO x . Studies have shown that in the catalytic cracking device, 30% to 50% of the nitrogen in the raw material is converted to the coke attached to the catalyst, and in the catalyst regeneration process, most of the nitrogen in the coke generates N2, and 10% to 30% of the nitrogen in the coke generates NO x and is discharged into the regeneration flue gas. The final NO x generation amount has a certain relationship with the nitrogen content of the raw material and the operation conditions of the regenerator.

[0004] Because the pollutant concentration in the flue gas outlet of the catalytic cracking regenerator flues significantly with varying operating conditions, the desulfurization and denitrification units of the catalytic cracking are affected by fluctuations in the inlet flue gas, making it difficult to maintain stable flue gas emissions. To identify the influencing factors of abnormal fluctuations in the flue gas concentration at the catalytic cracking outlet, Li Xiaoyan et al. analyzed the NO₂ concentration in the flue gas from the dual-stage desulfurization and denitrification process, starting with daily operation and the mechanisms of action of denitrification agents and combustion improvers. x Five reasons for NO exceeding the standard were identified, and solutions were proposed to address these reasons. Experimental results show that the proposed solutions can effectively control NO in flue gas. x The emission concentration of NO in the catalytic cracking regeneration flue gas at equilibrium. x The concentration of NO, and its influence x To determine the equilibrium concentration, Wang Longyan et al. used the atomic coefficient matrix method to determine the independent reactions of the complex regenerator system and performed thermodynamic calculations to analyze the effects of regenerator operating temperature, pressure, and other conditions on NO in the regenerated flue gas. x The effect of equilibrium concentration. Experimental results show the influence of NO... x The main factors affecting equilibrium concentration are regenerator operating temperature, pressure, initial concentration of O2 in the regenerator system, and initial concentration of CO.

[0005] Currently, the amount of pollutant data in regenerated flue gas is enormous, and research on prediction methods for nitrogen oxides in regenerated flue gas is not yet mature. Improving the accuracy of prediction remains a challenge that needs to be addressed. Summary of the Invention

[0006] The purpose of this invention is to provide a method, apparatus, and system for predicting nitrogen oxides in catalytic cracking regenerated flue gas, so as to at least solve the problem of low accuracy in predicting nitrogen oxides in regenerated flue gas.

[0007] To achieve the above objectives, a first aspect of the present invention provides a method for predicting nitrogen oxides in flue gas from catalytic cracking regeneration, the method comprising:

[0008] Preprocess the variable data to obtain the processed sample set;

[0009] An initial prediction model for nitrogen oxides in catalytic cracking regenerated flue gas was constructed based on the sample set.

[0010] The initial prediction model of nitrogen oxides in the catalytic cracking regeneration flue gas was optimized using a particle swarm optimization algorithm based on flight information to obtain the target prediction model of nitrogen oxides in the catalytic cracking regeneration flue gas.

[0011] The target prediction model is used to predict the preprocessed test data to obtain the predicted value of nitrogen oxides.

[0012] Optionally, the preprocessing of the variable data to obtain the processed sample set includes:

[0013] acquiring all variable data;

[0014] cleaning the variable data;

[0015] normalizing the cleaned variable data to obtain normalized data;

[0016] performing dimensionality reduction on the normalized data to obtain an input variable data set;

[0017] selecting a preset number of data sets from the input variable data set as a sample set.

[0018] Further, the cleaning of the variable data comprises:

[0019] According to the threshold range of each variable data, the abnormal values in each variable data are removed;

[0020] Using linear interpolation algorithm to fill in the missing data in each variable data, to obtain the cleaned variable data.

[0021] Optionally, the dimensionality reduction on the normalized data to obtain an input variable data set comprises:

[0022] Using principal component analysis to perform dimensionality reduction on the normalized data, to obtain a variable set whose correlation coefficient and contribution rate are greater than a threshold value, as the input variable data set.

[0023] Optionally, the variable data includes: regenerative flue gas import and export pollutants and reactor data during catalytic cracking device operation, regenerator data, desulfurization and denitrification facility data, product distribution data, and treatment facility raw material data; the reactor data, regenerator data, desulfurization and denitrification facility data, product distribution data, and treatment facility raw material data all include: raw material nitrogen content, regenerator oxygen content, regenerator dense phase storage capacity, regenerator main air volume, riser oil slurry feed rate, upper riser temperature, outlet flue gas temperature, total feed rate, soot concentration, upper riser temperature, regenerator bottom dense phase temperature, and regenerator dilute phase section pressure.

[0024] Optionally, the construction of a catalytic cracking regenerative flue gas nitrogen oxide initial prediction model according to the sample set comprises:

[0025] Constructing a catalytic cracking regenerative flue gas nitrogen oxide basic model based on a fuzzy neural network:

[0026]

[0027] In the formula, y(t) is the actual output of nitrogen oxide at time t, w(t) = [w1(t), w2(t), …, w P(t)] is the fuzzy neural network output weight, w l (t) is the weight of the connection between the output neuron and the lth rule layer neuron at time t, P is the total number of neurons, v l (t) is the output of the lth rule layer neuron at time t, v l The calculation formula of (t) is as follows:

[0028]

[0029]

[0030] In the formula, c j (t) = [c 1j (t), c 2j (t), …, c mj (t)] is the center of the jth radial basis function neuron at time t, σ j (t) = [σ 1j (t), σ 2j (t), …, σ mj (t)] is the width of the jth radial basis function neuron at time t, is the output value of the jth radial basis function neuron at time t, x(t) = [x1(t), x2(t), …, x m (t)] is the input of the nitrogen oxide prediction model at time t;

[0031] The error function expression is defined as:

[0032]

[0033] Wherein, z = 1, 2, …, Z, Z is the number of test samples, y d is the expected output of nitrogen oxide, y is the actual output of nitrogen oxide, e is the error of the nitrogen oxide prediction model;

[0034] The sample set is input into the catalytic cracking regeneration flue gas nitrogen oxide basic model for training, to obtain a catalytic cracking regeneration flue gas nitrogen oxide initial prediction model.

[0035] Optionally, the catalytic cracking regeneration flue gas nitrogen oxide initial prediction model is optimized by using a particle swarm optimization algorithm based on flight information, to obtain a catalytic cracking regeneration flue gas nitrogen oxide target prediction model, which comprises:

[0036] A particle represents a neural network, the population size of the particle is n, and the position expression of the particle is:

[0037] x i = {(w i，1 , σ i，1 , c i，1), (w i,2 , σ i,2 , c i,2 )...(w i,j , σ i,j , c i,j )};

[0038] where w i,1 is the weight of the first neuron in the fuzzy neural network represented by the i-th particle, σ i,1 is the width of the first neuron in the fuzzy neural network represented by the i-th particle, and c i,1 is the center of the first neuron in the fuzzy neural network represented by the i-th particle.

[0039] The positions, velocities and inertia weights of the particles in the particle swarm are initialized, and the population size and the maximum iteration number of the function are defined.

[0040] According to the error function, the error function value of each particle is calculated, and the error function value of each particle is compared with the current global optimal position g best (k), if the error function value of the particle is better, the g best (k) is updated, and the error function value of each particle is compared with the current historical optimal position p i (k), if the error function value of the particle is better, the p i (k) is updated, and SP(k) is calculated; wherein SP is an index representing the diversity of particles, and the expression is:

[0041]

[0042] SP(k+1) is the diversity of the particle swarm, d i (k+1) is the minimum Euclidean distance of the i-th particle and other particles in the k+1 iteration, is the average of all d i (k+1);

[0043] The position x i (k) and velocity v i (k) of the particles in the particle swarm are updated according to the velocity update formula and the position update formula, and the inertia weight ω i (k) is calculated.

[0044] The position update formula of the i-th particle is:

[0045] x i (k+1) = x i (k) + v i (k+1);

[0046] The velocity update formula of the i-th particle is:

[0047] v i (k+1) = ω i (k) v i (k) + ε1R1(p i (k) - x i (k)) + ε2R2(g best (k) - x i (k));

[0048] where ω(k) is the inertia weight, ε1 and ε2 are learning factors, R1 and R2 are random values between 0 and 1, p i (k) is the history optimal position of the particle at the kth iteration, g best (k) is the global optimal position found by the whole population at the kth iteration, x i (k) is the position of the i-th particle at the kth iteration, v i (k) is the velocity of the i-th particle at the kth iteration.

[0049] The error function value of the current position of the particle is calculated, the history optimal position p i (k) and the global optimal position g d (k) are updated, the value of the inertia weight ω of the current iteration is calculated according to the global optimal position g best (k) and the position x i (k) of the particle, and the inertia weight ω is updated according to the following formula:

[0050]

[0051] where g best (k+1) is the global optimal position of the i-th particle at the k+1th iteration, p i (k+1) is the history optimal position of the i-th particle at the k+1th iteration, f(p i (k+1)) is the fitness value of the i-th particle at the k+1th iteration, x i (k+1) is the position of the i-th particle at the k+1th iteration.

[0052] The algorithm stops when the number of iterations reaches the maximum number of iterations, and the optimal parameter value of the initial prediction model of the nitrogen oxides in the flue gas of the catalytic cracking regeneration is obtained.

[0053] Optionally, after the optimal parameter value of the initial prediction model of the nitrogen oxides in the flue gas of the catalytic cracking regeneration is obtained, the performance of the catalytic cracking regeneration flue gas nitrogen oxides prediction model using the optimal parameter is evaluated according to the root mean square error and the accuracy, and it is judged whether the prediction error and the accuracy are within a preset range.

[0054] If yes, a catalytic cracking regeneration flue gas nitrogen oxide prediction model using the optimal parameters is used as a catalytic cracking regeneration flue gas nitrogen oxide target prediction model;

[0055] The RMSE expression is:

[0056]

[0057] The calculation formula of the prediction accuracy is:

[0058]

[0059] Wherein, Z is the number of test samples, y d is the expected output of nitrogen oxide, and y is the actual output of nitrogen oxide.

[0060] The second aspect of the present application provides a catalytic cracking regeneration flue gas nitrogen oxide prediction device, comprising:

[0061] a controller configured to

[0062] preprocess variable data to obtain a processed sample set;

[0063] construct a catalytic cracking regeneration flue gas nitrogen oxide initial prediction model according to the sample set;

[0064] optimize the catalytic cracking regeneration flue gas nitrogen oxide initial prediction model using a particle swarm optimization algorithm based on flight information to obtain a catalytic cracking regeneration flue gas nitrogen oxide target prediction model;

[0065] use the target prediction model to predict preprocessed test data to obtain a nitrogen oxide prediction value.

[0066] Optionally, the preprocessing of the variable data to obtain the processed sample set comprises:

[0067] obtaining all variable data;

[0068] cleaning the variable data;

[0069] normalizing the cleaned variable data to obtain normalized data;

[0070] dimensionally reducing the normalized data to obtain an input variable data set;

[0071] selecting a preset number of data sets from the input variable data set as a sample set.

[0072] Optionally, the cleaning of the variable data comprises:

[0073] According to the threshold range of each variable data, the abnormal values in each variable data are removed.

[0074] The missing data in each variable data is filled in using a linear interpolation algorithm to obtain cleaned variable data.

[0075] Optionally, the dimensionality reduction processing of the normalized data to obtain the input variable data set includes:

[0076] Principal component analysis was used to reduce the dimensionality of the normalized data, and the variables with correlation coefficients and contribution rates greater than a threshold were used to form the input variable data set.

[0077] Optionally, the step of constructing an initial prediction model for nitrogen oxides in catalytic cracking regeneration flue gas based on the sample set includes:

[0078] Constructing a basic model for nitrogen oxide emissions from catalytic cracking regeneration flue gas based on fuzzy neural networks:

[0079]

[0080] In the formula, y(t) is the actual output of nitrogen oxides at time t, and w(t) = [w1(t), w2(t), ..., w P [(t)] represents the output weights of the fuzzy neural network, w l (t) represents the weights connecting the output neuron and the l-th regular layer neuron at time t, where P is the total number of neurons, and v l (t) is the output of the l-th regular layer neuron at time t, v l The formula for calculating (t) is as follows:

[0081]

[0082]

[0083] In the formula, c j (t)=[c 1j (t),c 2j (t),…,c mj [(t)] is the center of the j-th radial basis function neuron at time t, σ j (t)=[σ 1j (t),σ 2j (t),…,σ mj [(t)] is the width of the j-th radial basis function neuron at time t. It is the output value of the j-th radial basis function neuron at time t, x(t)=[x1(t),x2(t),…,x m [(t)] is the input to the nitrogen oxide prediction model at time t;

[0084] Define the expression for the error function:

[0085]

[0086] wherein z = 1, 2, …, Z, Z is the number of test samples, y d is the expected output of nitrogen oxides, y is the actual output of nitrogen oxides, and e is the error of the nitrogen oxides prediction model;

[0087] inputting the sample set into the catalytic cracking regeneration flue gas nitrogen oxides basic model to train, to obtain a catalytic cracking regeneration flue gas nitrogen oxides initial prediction model.

[0088] Optionally, the catalytic cracking regeneration flue gas nitrogen oxides initial prediction model is optimized by using a particle swarm optimization algorithm based on flight information to obtain a catalytic cracking regeneration flue gas nitrogen oxides target prediction model, including:

[0089] A particle represents a neural network, the population size of the particle is n, and the position of the particle is expressed as:

[0090] x i = {(w i，1 ,σ i，1 ,c i，1 ), (w i,2 ,σ i,2 ,c i,2 )...(w i,j ,σ i,j ,c i,j )};

[0091] In the formula, w i,1 is the weight of the first neuron in the fuzzy neural network represented by the i th particle, σ i,1 is the width of the first neuron in the fuzzy neural network represented by the i th particle, and c i,1 is the center of the first neuron in the fuzzy neural network represented by the i th particle.

[0092] The position, speed, and inertia weight of each particle in the particle swarm are initialized, and the population size and the maximum number of iterations of the function are defined;

[0093] According to the error function, the error function value of each particle is calculated, and the error function value of each particle is compared with the current global optimal position g best (k), if the error function value of the particle is better, the g best (k) is updated, and the error function value of each particle is compared with the current historical optimal position p i (k), if the error function value of the particle is better, the p i (k) is updated, and SP(k) is calculated; wherein SP is an index representing the diversity of particles, and the expression is:

[0094]

[0095] SP(k+1) is the diversity of the particle swarm, d i (k+1) is the minimum Euclidean distance of the i-th particle to other particles in the k+1-th iteration, is the average value of all d i (k+1);

[0096] The position x i (k) and the velocity v i (k) of the particle in the particle swarm are updated according to the position update formula and the velocity update formula, and the inertia weight ω i (k) is calculated.

[0097] The position update formula of the i-th particle is:

[0098] x i (k+1) = x i (k) + v i (k+1);

[0099] The velocity update formula of the i-th particle is:

[0100] v i (k+1) = ω i (k) v i (k) + ε1R1(p i (k) - x i (k)) + ε2R2(g best (k) - x i (k));

[0101] In the formula, ω(k) is the inertia weight, ε1 and ε2 are learning factors, R1 and R2 are random values between 0 and 1, p i (k) is the historical optimal position of the particle in the k-th iteration, g best (k) is the global optimal position found by the entire population in the k-th iteration, x i (k) is the position of the i-th particle in the k-th iteration, v i (k) is the velocity of the i-th particle in the k-th iteration.

[0102] The error function value of the current position of the particle is calculated, the historical optimal position p i (k) and the global optimal position g d (k) are updated, the value of the inertia weight in the current iteration is calculated according to the global optimal position g best (k) and the position x i (k) of the particle, and the inertia weight update formula is:

[0103]

[0104] wherein g best (k+1) is the global optimal position of the i-th particle in the k+1 iteration, p i (k+1) is the historical optimal position of the i-th particle in the k+1 iteration, f(p i (k+1)) is the fitness value of the i-th particle in the k+1 iteration, x i (k+1) is the position of the i-th particle in the k+1 iteration.

[0105] The algorithm stops when the number of iterations reaches the maximum number of iterations, and the optimal parameter value of the initial prediction model of the nitrogen oxides in the FCC regenerator flue gas is obtained.

[0106] Optionally, after obtaining the optimal parameter value of the initial prediction model of the nitrogen oxides in the FCC regenerator flue gas, the performance of the prediction model of the nitrogen oxides in the FCC regenerator flue gas using the optimal parameter is evaluated according to the root mean square error and the accuracy, and it is determined whether the prediction error and the accuracy are within a preset range.

[0107] If yes, the prediction model of the nitrogen oxides in the FCC regenerator flue gas using the optimal parameter is used as the target prediction model of the nitrogen oxides in the FCC regenerator flue gas.

[0108] The expression of the RMSE is:

[0109]

[0110] The calculation formula of the prediction accuracy is:

[0111]

[0112] wherein Z is the number of test samples, y d is the expected output of nitrogen oxides, and y is the actual output of nitrogen oxides.

[0113] The third aspect of the present application provides a prediction system of nitrogen oxides in the FCC regenerator flue gas, characterized in that the system comprises:

[0114] a data processing module for preprocessing variable data to obtain a processed sample set;

[0115] an initial prediction model construction module for constructing an initial prediction model of nitrogen oxides in the FCC regenerator flue gas according to the sample set;

[0116] an initial prediction model optimization module for optimizing the initial prediction model of nitrogen oxides in the FCC regenerator flue gas by using a particle swarm optimization algorithm based on flight information to obtain a target prediction model of nitrogen oxides in the FCC regenerator flue gas;

[0117] A data prediction module is configured to use the target prediction model to predict the preprocessed data to be measured to obtain a nitrogen oxide prediction value.

[0118] In another aspect, the application provides a machine readable storage medium having instructions stored thereon for causing a machine to perform the catalytic cracking regeneration flue gas nitrogen oxide prediction method.

[0119] Through the above technical solution, the data set with higher correlation with the regeneration flue gas nitrogen oxide is obtained as the training sample set and the test sample set through the preprocessing of the data, the fuzzy neural network model is used as the initial prediction model, the particle swarm optimization algorithm based on flight information is used to optimize the initial prediction model, and finally the target prediction model capable of quickly predicting and having higher accuracy is obtained, which is beneficial to improving the nitrogen oxide prediction accuracy.

[0120] Other features and advantages of the embodiments of the application will be described in detail in the following detailed description. BRIEF DESCRIPTION OF DRAWINGS

[0121] The accompanying drawings are included to provide a further understanding of the embodiments of the application, and constitute a part of the specification, and are used together with the following detailed description to explain the embodiments of the application, but do not constitute a limitation on the embodiments of the application. In the drawings:

[0122] Figure 1 is a catalytic cracking regeneration flue gas nitrogen oxide prediction method flowchart provided by an embodiment of the application;

[0123] Figure 2 is a FIPSO-fuzzy neural network structure diagram provided by an embodiment of the application;

[0124] Figure 3 is a catalytic cracking regeneration flue gas nitrogen oxide prediction system block diagram provided by an embodiment of the application. DETAILED DESCRIPTION

[0125] The specific embodiments of the application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the application, and are not used to limit the application.

[0126] Figure 1 is a catalytic cracking regeneration flue gas nitrogen oxide prediction method flowchart provided by an embodiment of the application. As shown in Figure 1 , the method comprises:

[0127] Step 1: Preprocess the variable data to obtain a processed sample set, specifically comprising:

[0128] 1) Obtain all variable data, including: regenerator inlet and outlet pollutants and reactor data during the operation of the catalytic cracking unit, regenerator data, desulfurization and denitrification facility data, product distribution data, and treatment facility feedstock data; the reactor data, regenerator data, desulfurization and denitrification facility data, product distribution data, and treatment facility feedstock data all include: feedstock nitrogen content, regenerator oxygen content, regenerator dense phase inventory, regenerator main air volume, riser slurry feed volume, riser upper temperature, outlet flue gas temperature, total feed volume, soot concentration, riser upper temperature, regenerator bottom dense phase temperature, and regenerator dilute phase section pressure.

[0129] In the present embodiment, the variable data is generally obtained through online instruments or laboratory analysis, and the obtained variable data needs to be arranged according to a time scale and stored in a database.

[0130] 2) Clean the variable data, which in the present embodiment includes two parts: abnormal value deletion and missing data supplementation. Abnormal value deletion needs to delete abnormal values in each variable data according to the threshold range of each variable data; missing data supplementation needs to supplement missing data in each variable data using a linear interpolation algorithm to obtain cleaned variable data.

[0131] 3) Normalize the cleaned variable data to obtain normalized data, which can eliminate the influence of differences in dimensions and orders of magnitude on the model training process.

[0132] 4) Reduce the dimension of the normalized data to obtain an input variable data set. In the present embodiment, principal component analysis is used to reduce the dimension of the normalized data to obtain an input variable data set composed of variables with a correlation coefficient and a contribution rate greater than a threshold value. Through principal component analysis, the most relevant data to nitrogen oxides can be determined from numerous variable data, reducing the amount of data to be calculated in the prediction process.

[0133] 5) Select a preset number of data sets from the input variable data set as a sample set. In the present embodiment, the sample set includes a training sample set and a test sample set, which are randomly divided according to a set data ratio.

[0134] Step two: Construct a catalytic cracking regenerator flue gas nitrogen oxide initial prediction model according to the sample set, including:

[0135] A fuzzy neural network (FNN) based basic model of nitrogen oxides in FCC regenerator flue gas is constructed. The topology of the basic model is divided into four layers: input layer, radial basis function (RBF) layer, normalization layer and output layer. The connection mode of the topology is k-Q-Q-1, the connection weight between the input layer and the RBF layer is 1, and the connection weight between the normalization layer and the output layer is randomly assigned in the interval [-1, 1]. The expected output of the fuzzy neural network is represented as y d , and the actual output is represented as y.

[0136] The mathematical expression of the basic model of nitrogen oxides in FCC regenerator flue gas is:

[0137] In the formula, y(t) is the actual output of nitrogen oxides at time t. w(t) = [w1(t), w2(t), …, w P (t)] is the output weight of the fuzzy neural network, w l (t) is the weight connecting the output neuron and the lth rule layer neuron at time t, P is the total number of neurons, v l (t) is the output of the lth rule layer neuron at time t, v l (t) is calculated as follows:

[0138]

[0139]

[0140] In the formula, c j (t) = [c 1j (t), c 2j (t), …, c mj (t)] is the center of the jth RBF neuron at time t, σ j (t) = [σ 1j (t), σ 2j (t), …, σ mj (t)] is the width of the jth RBF neuron at time t, v is the output value of the jth RBF neuron at time t, x(t) = [x1(t), x2(t), …, x m (t)] is the input of the nitrogen oxides prediction model at time t;

[0141] The error function expression is defined as:

[0142]

[0143] wherein z = 1, 2, …, Z, Z is the number of test samples, y d is the expected output of nitrogen oxides, y is the actual output of nitrogen oxides, and e is the error of the nitrogen oxides prediction model.

[0144] inputting the sample set into the catalytic cracking regeneration flue gas nitrogen oxide basic model to train, and obtaining a catalytic cracking regeneration flue gas nitrogen oxide initial prediction model.

[0145] Step three: optimizing the catalytic cracking regeneration flue gas nitrogen oxide initial prediction model by using a flight information based particle swarm optimization algorithm (FIPSO) to obtain a catalytic cracking regeneration flue gas nitrogen oxide target prediction model, including:

[0146] 1) defining a particle to represent a neural network, as shown in the following formula: Figure 2

[0147] i i，1 i，1 i，1 i,2 i,2 i,2 i,j i,j i,j

[0148] i,1 i,1 i,1

[0149] 2) initializing the position, speed and inertia weight of each particle in the particle swarm, defining the population size and the maximum iteration number of the function, and in practice, the initial position of each particle can be set as the current historical optimal position, and the optimal value in the particle swarm can be set as the global optimal position;

[0150] 3) according to the error function, calculating the error function value of each particle, comparing the error function value of each particle with the current global optimal position g best (k), if the error function value of the particle is better, updating g best (k), and at the same time, comparing the error function value of each particle with the current historical optimal position p i (k), if the error function value of the particle is better, updating p i (k), and calculating SP(k); wherein, SP is an index representing the diversity of particles, and the expression is as follows:

[0151]

[0152] ​​​​​​​​​​​​​​​​SP(k+1) is the diversity of the particle swarm, d i (k+1) is the minimum Euclidean distance between the ith particle and other particles in the k+1th iteration, is the average value of all d i (k+1). When the diversity of the particle swarm increases, it means that the particle swarm is more dispersed, and the local search ability of the particle needs to be improved. The corresponding measure is to reduce the inertia weight ω. Conversely, when the diversity of the particle swarm decreases, it means that the distribution of the particle swarm is more concentrated, and it is necessary to jump out of the local optimal solution to avoid premature convergence. Therefore, the global search ability of the particle needs to be improved, and the corresponding measure is to increase the inertia weight ω. In the case of constant population diversity, in order to prevent the particle from falling into the local optimum, the global search ability of the particle needs to be enhanced. Therefore, the value of the inertia weight needs to be increased.

[0153] 4) Update the position x i (k) and the velocity v i (k) of the particle in the particle swarm according to the velocity update formula and the position update formula, and calculate the inertia weight ω i (k);

[0154] The position update formula of the ith particle is:

[0155] x i (k+1) = x i (k) + v i (k+1);

[0156] The velocity update formula of the ith particle is:

[0157] v i (k+1) = ω i (k) v i (k) + ε1R1(p i (k) - x i (k)) + ε2R2(g best (k) - x i (k));

[0158] In the formula, ω(k) is the inertia weight, ε1 and ε2 are learning factors, R1 and R2 are random values between 0 and 1, p i (k) is the historical optimal position of the particle in the kth iteration, g best (k) is the global optimal position found by the entire population in the kth iteration, x i (k) is the position of the ith particle in the kth iteration, v i (k) is the velocity of the ith particle in the kth iteration;

[0159] 5) Calculate the error function value of the current position of the particle, update the historical optimal position p i(k) and the global optimal position g d (k), according to the global optimal position g best (k) and the position x of the particle i (k) to calculate the value of the inertia weight of the current iteration, and the inertia weight ω update formula is:

[0160]

[0161] Wherein, g best (k+1) is the global optimal position of the i-th particle in the k+1 iteration, p i (k+1) is the historical optimal position of the i-th particle in the k+1 iteration, f(p i (k+1)) is the fitness value of the i-th particle in the k+1 iteration, x i (k+1) is the position of the i-th particle in the k+1 iteration.

[0162] 6) When the number of iterations reaches the maximum number of iterations, the algorithm stops, and the optimal parameter value of the initial prediction model of the nitrogen oxide of the flue gas of the catalytic cracking regeneration is obtained.

[0163] After obtaining the optimal parameter value of the initial prediction model of the nitrogen oxide of the flue gas of the catalytic cracking regeneration, the performance of the catalytic cracking regeneration flue gas nitrogen oxide prediction model using the optimal parameter is evaluated according to the root mean square error and the accuracy, and it is judged whether the prediction error and the accuracy are within the preset range;

[0164] If yes, the catalytic cracking regeneration flue gas nitrogen oxide prediction model using the optimal parameter is used as the target prediction model of the nitrogen oxide of the flue gas of the catalytic cracking regeneration;

[0165] The RMSE expression is:

[0166]

[0167] The calculation formula of the prediction accuracy is:

[0168]

[0169] Wherein, Z is the number of test samples, y d is the expected output of nitrogen oxide, and y is the actual output of nitrogen oxide.

[0170] Step four: using the target prediction model to predict the preprocessed data to be tested, and obtaining the predicted value of nitrogen oxide.

[0171] The above method can be used to master the concentration of nitrogen oxide at the outlet of the regenerator in real time, guide the adjustment of the parameters of the subsequent desulfurization and denitrification facilities and the addition of additives, and play a key guiding role in the regulation and control of the desulfurization and denitrification facilities.

[0172] Example One

[0173] (1) Collecting data and cleaning

[0174] In this embodiment, the pollutants at the inlet and outlet of the regenerated flue gas and the reactor data, regenerator data, desulfurization and denitrification facility data, product distribution data, and raw material data of the treatment facility during the operation of the catalytic cracking device are obtained through online instruments or laboratory analysis, which includes the following 74 parameters, specifically: raw material nitrogen content, regenerator oxygen content, regenerator dense phase storage capacity, regenerator main air volume, riser oil slurry feed volume, riser upper temperature, outlet flue gas temperature, total feed volume, dust concentration, riser upper temperature, regenerator bottom dense phase temperature, regenerator dilute phase section pressure, etc. The data of all the collected variables are subjected to data preprocessing operations to eliminate abnormal data; then the data are subjected to normalization processing to eliminate the influence of the differences in dimensions and orders of magnitude on the model training process.

[0175] The collected data are subjected to dimension reduction using principal component analysis, and based on the correlation coefficient and contribution rate of the variables, it is found that the input variables are 6: raw material nitrogen content, regenerator oxygen content, regenerator dense phase storage capacity, riser oil slurry feed volume, riser upper temperature, and total feed volume. The output variable to be collected is the nitrogen oxide concentration value. 1000 groups of data are selected from all the collected data and divided into two parts: 600 groups are used as training samples, and 400 groups are used as test samples.

[0176] (2) Establishing a catalytic cracking regenerated flue gas nitrogen oxide initial prediction model

[0177] A fuzzy neural network is used to design a catalytic cracking regenerated flue gas nitrogen oxide prediction model, which has a topological structure divided into four layers: an input layer, an RBF layer, a normalization layer, and an output layer; the topological structure is connected in the manner of 6-9-9-1, the connection weight value between the input layer and the RBF layer is 1, the connection weight value between the normalization layer and the output layer is randomly assigned, the assignment interval is [-1, 1], and the expected output of the fuzzy neural network is represented as y d , and the actual output is represented as y; the expression of the catalytic cracking regenerated flue gas nitrogen oxide initial prediction model based on the fuzzy neural network is:

[0178] In the formula, y(t) is the actual output of the nitrogen oxide at time t. w(t) = [w1(t), w2(t), …, w P (t)] is the FNN output weight value, w l (t) is the weight value connecting the output neuron and the lth rule layer neuron at time t, and P is the total number of neurons.

[0179] The model includes:

[0180] ① Input layer: the layer is composed of 6 neurons, and its output is,

[0181] x i = u i (i = 1, 2, …, 6)

[0182] x = [x1, x2, …, x6],

[0183] wherein the number of input layer neurons is 6, u i is the input value of the i-th input neuron, and x is the input vector.

[0184] ② RBF layer:

[0185]

[0186] The number of RBF layer neurons is 9, c j = [c 1j , c 2j , …, c kj ], σ = [σ 1j , σ 2j , …, σ kj ] are the center and width of the j-th neuron respectively, and φ j is the output value of the j-th neuron.

[0187] ③ Normalization layer: the normalization layer has the same number of neurons as the RBF layer

[0188]

[0189] y = Wv,

[0190] W = [w 1 , w 2 , …, w M ] T ,

[0191] ④ Output layer: the output of the output layer is the actual output of the catalytic cracking regeneration flue gas nitrogen oxide prediction model:

[0192]

[0193] y m is the output of the m-th neuron of the output layer; w is the weight matrix, w m = [w1 m , w2 m , …, w q m ] is the weight of the m-th neuron between the output layer and the rule layer; and v is the output of the rule layer.

[0194] The error function expression is defined as:

[0195]

[0196] wherein y d is the desired output of nitrogen oxides, y is the actual output of nitrogen oxides, and e is the error of the nitrogen oxides prediction model.

[0197] (3) optimizing the initial prediction model of nitrogen oxides in catalytic cracking regeneration flue gas by using a particle swarm optimization algorithm based on flight information

[0198] In this embodiment, first, the population size is set to 20, the maximum number of iterations of all functions is set to 5000, and each algorithm is independently run 30 times on all test functions. The positions, velocities and inertia weights of each particle in the particle swarm are randomly initialized. The initial position of each particle is set as the current historical optimal position, and the optimal value in the particle swarm is set as the global optimal position.

[0199] Then, the error function value of each particle is compared with the current global optimal position g best (k), and if better, the g best (k) is updated, and the error function value of each particle is compared with the current historical optimal position p i (k), and if better, the p i (k) is updated, and SP(k) is calculated; wherein SP is an index representing the diversity of particles, and the expression is:

[0200]

[0201] SP(k+1) is the diversity of the particle swarm, d i (k+1) is the minimum Euclidean distance of the i-th particle from other particles in the k+1 iteration, is the average value of all d i (k+1), and the error function of each particle is calculated.

[0202] The position x i (k) and velocity v i (k) of each particle in the particle swarm are updated according to the velocity update formula and the position update formula, and the inertia weight ω i (k) is calculated.

[0203] The position update formula of the i-th particle is:

[0204] x i (k+1) = x i (k) + v i (k+1)

[0205] The velocity update formula of the i-th particle is:

[0206] v i (k+1) = ω i (k)v i (k) + ε1R1(p i (k) - x i (k) + ε2R2(g best (k) - x i (k)

[0207] where ω(k) is the inertia weight, ε1 and ε2 are learning factors, R1 and R2 are random values between 0 and 1, p i (k) is the history optimal position of the particle at the kth iteration, g best (k) is the global optimal position found by the whole population at the kth iteration, x i (k) is the position of the i-th particle at the kth iteration, v i (k) is the velocity of the i-th particle at the kth iteration.

[0208] The error function value of the current position of the particle is calculated, the history optimal position p i (k) and the global optimal position g d (k) are updated, the value of the inertia weight ω of the current iteration is calculated according to the global optimal position g best (k) and the position x i (k) of the particle, and the inertia weight ω is updated.

[0209]

[0210] where g best (k+1) is the global optimal position of the i-th particle at the k+1th iteration, p i (k+1) is the history optimal position of the i-th particle at the k+1th iteration, f(p i (k+1)) is the fitness value of the i-th particle at the k+1th iteration, x i (k+1) is the position of the i-th particle at the k+1th iteration.

[0211] The algorithm stops when the number of iterations reaches the maximum number of iterations, and the optimal parameter value of the initial prediction model of the nitrogen oxide in the flue gas of the catalytic cracking regeneration is obtained.

[0212] After obtaining the optimal parameter value of the initial prediction model of the nitrogen oxide in the flue gas of the catalytic cracking regeneration, the performance of the catalytic cracking regeneration flue gas nitrogen oxide prediction model using the optimal parameter is evaluated according to the root mean square error and the accuracy, and in this embodiment, the root mean square error and the accuracy evaluation of the prediction model are shown in Table 1. As can be seen from Table 1, the error and accuracy of the nitrogen oxide prediction are within a reasonable range.

[0213] Table 1 prediction results of different neural networks

[0214] Prediction method RMSE Prediction accuracy FIPSO-fuzzy neural network 1.222 90.2％ Existing prediction methods 1.622 84.5％

[0215] Example two

[0216] (1) Obtain data and organize and clean up

[0217] In this embodiment, the pollutants at the inlet and outlet of the regenerated flue gas and the reactor data, regenerator data, desulfurization and denitrification facility data, product distribution data, and raw material data of the treatment facility during the operation of the catalytic cracking device are obtained through online instruments or laboratory analysis, which contains the following 60 parameters, specifically: raw material nitrogen content, regenerator oxygen content, regenerator dense phase storage capacity, regenerator main air volume, riser oil slurry feed volume, upper riser temperature, outlet flue gas temperature, total feed volume, dust concentration, upper riser temperature, regenerator bottom dense phase temperature, regenerator dilute phase section pressure, etc. The data of all the obtained variables are subjected to data preprocessing operation to eliminate abnormal data; and the data are subjected to normalization processing to eliminate the influence of the differences in dimensions and orders of magnitude on the model training process.

[0218] The principal component analysis method is used to reduce the dimension of the collected data, and based on the correlation coefficient and contribution rate of the variables, the input variables are 5: raw material nitrogen content, regenerator oxygen content, regenerator dense phase storage capacity, upper riser temperature, and total feed volume. The output variable to be collected is the nitrogen oxide concentration value. 1200 groups of data are selected from all the collected data and divided into two parts: 600 groups as training samples and 600 groups as test samples.

[0219] (2) Establishing an initial prediction model for nitrogen oxides in the regenerated flue gas of catalytic cracking

[0220] A fuzzy neural network is used to design a prediction model for nitrogen oxides in the regenerated flue gas of catalytic cracking, and the topological structure is divided into four layers: input layer, RBF layer, normalization layer, and output layer; the connection mode of the topological structure is 5-7-7-1, the connection weight value between the input layer and the RBF layer is 1, the connection weight value between the normalization layer and the output layer is randomly assigned, the assignment interval is [-1, 1], and the expected output of the fuzzy neural network is represented as y d , and the actual output is represented as y.

[0221] (3) Optimizing the initial prediction model for nitrogen oxides in the regenerated flue gas of catalytic cracking using a particle swarm optimization algorithm based on flight information

[0222] In the embodiment, first, the population size is set to 20, the maximum number of iterations of all functions is set to 5000, each algorithm is independently run 30 times on all test functions. The position, speed and inertia weight of each particle in the particle swarm are randomly initialized. And the initial position of each particle is set as the current historical optimal position, and the optimal value in the particle swarm is set as the global optimal position.

[0223] Then according to the error function value of each particle is compared with the current global optimal position g best (k) best (k), and the error function value of each particle is compared with the current historical optimal position p i (k) i (k), SP(k) is calculated.

[0224] According to the speed update formula and the position update formula, the position x i (k) and the speed v i (k) of the particles in the particle swarm are updated, and the inertia weight ω i (k) is calculated.

[0225] The error function value of the current position of the particle is calculated, the historical optimal position p i (k) and the global optimal position g d (k) are updated, and the value of the inertia weight of the current iteration is calculated according to the global optimal position g best (k) and the position x i (k) of the particle.

[0226] When the number of iterations reaches the maximum number of iterations, the algorithm stops, and the optimal parameter value of the initial prediction model of the nitrogen oxide in the flue gas of the catalytic cracking regeneration is obtained.

[0227] After obtaining the optimal parameter value of the initial prediction model of the nitrogen oxide in the flue gas of the catalytic cracking regeneration, the performance of the catalytic cracking regeneration flue gas nitrogen oxide prediction model using the optimal parameter is evaluated according to the root mean square error and the accuracy, and in the embodiment, the root mean square error and the accuracy evaluation of the prediction model are shown in Table 2. As can be seen from Table 2, the error and accuracy of the nitrogen oxide prediction are within a reasonable range.

[0228] Table 2 Prediction results of different methods

[0229] Prediction method RMSE Prediction accuracy FIPSO-fuzzy neural network 1.213 88.2％ Existing prediction methods 1.312 85.2％

[0230] Example Three

[0231] (1) Obtain data and organize and clean up

[0232] In this embodiment, pollutant data from the regenerated flue gas inlet and outlet, as well as reactor data, regenerator data, desulfurization and denitrification facility data, product distribution data, and raw material data from the treatment facilities during the operation of the catalytic cracking unit, are acquired through online instruments or laboratory analysis. This includes 60 parameters: raw material nitrogen content, regenerator oxygen content, regenerator dense phase volume, regenerated main air volume, riser slurry feed rate, riser top temperature, outlet flue gas temperature, total feed rate, dust concentration, riser top temperature, regenerator bottom dense phase temperature, and regenerator dilute phase section pressure. All acquired variable data undergoes preprocessing to remove outliers; then, the data is normalized to eliminate the impact of differences in dimensions and orders of magnitude on model training.

[0233] Principal component analysis was used to reduce the dimensionality of the collected data. Based on the correlation coefficients and contribution rates of the variables, five input variables were identified: raw material nitrogen content, regenerator oxygen content, regenerator dense phase content, riser top temperature, and total feed rate. The output variable to be collected is the nitrogen oxide concentration. 1200 data sets were selected from all collected data and divided into two parts: 600 sets as training samples and 600 sets as test samples.

[0234] (2) Establish an initial prediction model for nitrogen oxides in catalytic cracking regeneration flue gas.

[0235] A fuzzy neural network was used to design a prediction model for nitrogen oxides in catalytic cracking regeneration flue gas. The model's topology consists of four layers: an input layer, an RBF layer, a normalization layer, and an output layer. The topology follows a 5-7-7-1 connection pattern. The connection weight between the input layer and the RBF layer is 1, while the connection weight between the normalization layer and the output layer is randomly assigned within the range [-1, 1]. The expected output of the fuzzy neural network is represented by y. d The actual output is represented as y.

[0236] (3) The initial prediction model for nitrogen oxides in the catalytic cracking regeneration flue gas was optimized using a particle swarm optimization algorithm based on flight information.

[0237] In this embodiment, firstly, the population size is set to 20, the maximum number of iterations for all functions is set to 5000, and each algorithm is run independently 30 times on all test functions. The position, velocity, and inertia weight of each particle in the particle swarm are randomly initialized. The initial position of each particle is set to the current historical best position, and the best value in the particle swarm is set as the global best position.

[0238] Then, based on the error function value of each particle and the current global optimal position g... best (k) is compared; if it is better, then g is updated. best (k), and at the same time, the error function value of each particle is compared with the current historical best position p.i (k) if better, update p i (k), calculate SP(k).

[0239] According to the speed update formula and the position update formula, the position x of the particle in the particle swarm is updated i (k) and the speed v i (k), the inertia weight ω i (k) is calculated.

[0240] The error function value of the current position of the particle is calculated, and the historical optimal position p i (k) and the global optimal position g d (k) is updated according to the global optimal position g best (k) and the position x of the particle i (k) to calculate the value of the inertia weight of the current iteration.

[0241] When the number of iterations reaches the maximum number of iterations, the algorithm stops, and the optimal parameter value of the initial prediction model of the nitrogen oxide in the catalytic cracking regeneration flue gas is obtained.

[0242] After obtaining the optimal parameter value of the initial prediction model of the nitrogen oxide in the catalytic cracking regeneration flue gas, the performance of the catalytic cracking regeneration flue gas nitrogen oxide prediction model using the optimal parameter is evaluated according to the root mean square error and the accuracy, and in this embodiment, the root mean square error and the accuracy evaluation of the prediction model are shown in Table 3. As can be seen from Table 3, the error and accuracy of the nitrogen oxide prediction are within a reasonable range.

[0243] Table 3 prediction results of different methods

[0244] Prediction method RMSE Prediction accuracy FIPSO-fuzzy neural network 1.223 89.3％ Existing prediction methods 1.421 86.2％

[0245] The second aspect of the present application provides a catalytic cracking regeneration flue gas nitrogen oxide prediction device, comprising:

[0246] A controller is used for

[0247] Pretreating the variable data to obtain a training sample set and a test sample set;

[0248] According to the training sample set, a catalytic cracking regeneration flue gas nitrogen oxide initial prediction model is constructed;

[0249] The particle swarm optimization algorithm based on flight information is used to optimize the catalytic cracking regeneration flue gas nitrogen oxide initial prediction model, and a catalytic cracking regeneration flue gas nitrogen oxide target prediction model is obtained.

[0250] The test sample set is predicted using the target prediction model.

[0251] In the embodiment, the preprocessing of the variable data comprises:

[0252] all the variable data is obtained;

[0253] the variable data is cleaned;

[0254] the cleaned variable data is normalized to obtain normalized data;

[0255] the normalized data is reduced in dimension to obtain an input variable data group;

[0256] a preset number of data groups in the input variable data group are selected as a sample set.

[0257] In the embodiment, the cleaning of the variable data comprises:

[0258] abnormal values in each variable data are removed according to a threshold range of each variable data;

[0259] missing data in each variable data is filled by using a linear interpolation algorithm to obtain cleaned variable data.

[0260] In the embodiment, the reducing in dimension of the normalized data to obtain the input variable data group comprises:

[0261] principal component analysis is used to reduce the normalized data in dimension to obtain an input variable data group composed of variables with a correlation coefficient and a contribution rate greater than a threshold.

[0262] In the embodiment, the constructing of the initial prediction model of the nitrogen oxide in the FCC regenerator flue gas according to the training sample set comprises:

[0263] an initial prediction model of the nitrogen oxide in the FCC regenerator flue gas is constructed based on a fuzzy neural network (FNN):

[0264]

[0265] wherein y(t) is an actual output of the nitrogen oxide at time t, w(t)=[w1(t), w2(t), …, wP(t)] is an output weight of the fuzzy neural network, wP(t) is a weight value connecting an output neuron and an lth rule layer neuron at time t, P is a total number of neurons, v(t) is an output of the lth rule layer neuron at time t, and v(t) is calculated according to the following formula: P l l l

[0266]

[0267] ​​​​

[0268] In the formula, c j (t) = [c 1j (t), c 2j (t), …, c mj (t)] is the center of the jth radial basis function neuron at time t, σ j (t) = [σ 1j (t), σ 2j (t), …, σ mj (t)] is the width of the jth radial basis function neuron at time t, is the output value of the jth radial basis function neuron at time t, x(t) = [x1(t), x2(t), …, x m (t)] is the input of the nitrogen oxide prediction model at time t;

[0269] The error function expression is defined as:

[0270]

[0271] wherein z = 1, 2, …, Z, Z is the number of test samples, y d is the expected output of nitrogen oxide, y is the actual output of nitrogen oxide, and e is the error of the nitrogen oxide prediction model;

[0272] The training sample set is input into the catalytic cracking regeneration flue gas nitrogen oxide basic model for training, to obtain a catalytic cracking regeneration flue gas nitrogen oxide initial prediction model.

[0273] In the embodiment, the catalytic cracking regeneration flue gas nitrogen oxide initial prediction model is optimized by using a particle swarm optimization algorithm based on flight information, to obtain a catalytic cracking regeneration flue gas nitrogen oxide target prediction model, which includes:

[0274] A particle represents a neural network, the population size of the particle is n, and the position expression of the particle is:

[0275] x i = {(w i，1 , σ i，1 , c i，1 ), (w i,2 , σ i,2 , c i,2 )...(w i,j , σ i,j , c i,j )};

[0276] In the formula, w i,1 is the weight of the first neuron in the fuzzy neural network represented by the ith particle, σ i,1width of the first neuron in the fuzzy neural network represented by the i-th particle, c i,1 center of the first neuron in the fuzzy neural network represented by the i-th particle;

[0277] initialize the position, velocity and inertia weight of each particle in the particle swarm, define the population size and the maximum iteration number of the function;

[0278] according to the error function, calculate the error function value of each particle, compare the error function value of each particle with the current global optimal position g best (k), if the error function value of the particle is better, update g best (k), and compare the error function value of each particle with the current historical optimal position p i (k), if the error function value of the particle is better, update p i (k), calculate SP(k); wherein SP is an index representing the diversity of particles, and the expression is:

[0279]

[0280] SP(k+1) is the diversity of the particle swarm, d i (k+1) is the minimum Euclidean distance of the i-th particle from other particles in the k+1-th iteration, is the average value of all d i (k+1);

[0281] update the position x i (k) and velocity v i (k) of the particles in the particle swarm according to the velocity update formula and the position update formula, and calculate the inertia weight ω i (k);

[0282] the position update formula of the i-th particle is:

[0283] x i (k+1) = x i (k) + v i (k+1);

[0284] the velocity update formula of the i-th particle is:

[0285] v i (k+1) = ω i (k) v i (k) + ε1R1(p i (k) - x i (k)) + ε2R2(g best (k) - x i (k));

[0286] where ω(k) is the inertia weight, ε1 and ε2 are learning factors, R1 and R2 are random values between 0 and 1, p i (k) is the history optimal position of the particle at the kth iteration, g best (k) is the global optimal position found by the entire population at the kth iteration, x i (k) is the position of the i-th particle at the kth iteration, v i (k) is the velocity of the i-th particle at the kth iteration.

[0287] The error function value of the current position of the particle is calculated, the history optimal position p i (k) and the global optimal position g d (k) are updated according to the global optimal position g best (k) and the position of the particle x i (k) is calculated, and the inertia weight ω is updated according to the following formula:

[0288]

[0289] where g best (k+1) is the global optimal position of the i-th particle at the k+1th iteration, p i (k+1) is the history optimal position of the i-th particle at the k+1th iteration, f(p i (k+1)) is the fitness value of the i-th particle at the k+1th iteration, x i (k+1) is the position of the i-th particle at the k+1th iteration.

[0290] When the number of iterations reaches the maximum number of iterations, the algorithm stops, and the optimal parameter value of the initial prediction model of the nitrogen oxides in the flue gas of the catalytic cracking regeneration is obtained.

[0291] In this embodiment, after obtaining the optimal parameter value of the initial prediction model of the nitrogen oxides in the flue gas of the catalytic cracking regeneration, the performance of the catalytic cracking regeneration flue gas nitrogen oxides prediction model using the optimal parameters is evaluated according to the root mean square error and the accuracy, and it is judged whether the prediction error and the accuracy are within the preset range;

[0292] If yes, the catalytic cracking regeneration flue gas nitrogen oxides prediction model using the optimal parameters is used as the target prediction model of the catalytic cracking regeneration flue gas nitrogen oxides;

[0293] The expression of RMSE is:

[0294]

[0295] The calculation formula of the prediction accuracy is:

[0296]

[0297] wherein Z is the number of samples tested, y d is the desired output of nitrogen oxides, y is the actual output of nitrogen oxides.

[0298] The third aspect of the present application provides a catalytic cracking regeneration flue gas nitrogen oxides prediction system, as shown in the figure, the system comprises: Figure 3

[0299] a data processing module, configured to preprocess variable data to obtain a training sample set and a test sample set;

[0300] an initial prediction model construction module, configured to construct an initial prediction model of catalytic cracking regeneration flue gas nitrogen oxides according to the training sample set;

[0301] an initial prediction model optimization module, configured to optimize the initial prediction model of catalytic cracking regeneration flue gas nitrogen oxides by using a particle swarm optimization algorithm based on flight information to obtain a target prediction model of catalytic cracking regeneration flue gas nitrogen oxides;

[0302] a data prediction module, configured to predict the test sample set by using the target prediction model.

[0303] In another aspect, the present application provides a machine readable storage medium, which stores instructions for causing a machine to execute the catalytic cracking regeneration flue gas nitrogen oxides prediction method described in the present application.

[0304] Those skilled in the art can understand that all or part of the steps of the method for implementing the above-mentioned embodiments can be instructed by a program to relevant hardware, the program is stored in a storage medium, including a plurality of instructions for causing a single-chip microcomputer, chip or processor to execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk and various program code storage media.

[0305] ​The optional embodiments of the present application are described in detail above with reference to the drawings, but the embodiments of the present application are not limited to the specific details in the above-described embodiments. Within the technical concept of the embodiments of the present application, various simple modifications can be made to the technical solutions of the embodiments of the present application, and these simple modifications all belong to the protection scope of the embodiments of the present application. In addition, it should be noted that, in the above-described specific embodiments, various specific technical features can be combined in any appropriate manner without contradiction. In order to avoid unnecessary repetition, various possible combinations are not described again in the embodiments of the present application.

[0306] In addition, various different embodiments of the present application can also be combined in any appropriate manner, as long as it does not deviate from the idea of the embodiments of the present application, and it should also be considered as disclosed by the embodiments of the present application.

Claims

1. A method for predicting nitrogen oxides in catalytic cracking regeneration flue gas, characterized in that, The method includes: Preprocess the variable data to obtain the processed sample set; An initial prediction model for nitrogen oxides in catalytic cracking regenerated flue gas was constructed based on the sample set. The initial prediction model for nitrogen oxides in the catalytic cracking regeneration flue gas was optimized using a particle swarm optimization algorithm based on flight information to obtain the target prediction model for nitrogen oxides in the catalytic cracking regeneration flue gas, including: Define a particle as representing a neural network, and the population size of the particles is... n The position expression of the particle is: ； In the formula, w i,1 For the first i The weight σ of the first neuron in the fuzzy neural network, represented by each particle. i,1 For the first i c represents the width of the first neuron in the fuzzy neural network, where each particle represents a neuron. i,1 For the first i The center of the first neuron in the fuzzy neural network is represented by a particle; Initialize the position, velocity, and inertial weights of each particle in the particle swarm, and define the population size and the maximum number of iterations of the function; Based on the error function, calculate the error function value for each particle, and then compare the error function value of each particle with the current global optimal position. g best ( k In comparison, if the particle's error function value is better, then update. g best ( k Simultaneously, the error function value of each particle is compared with the current historical best position. p i ( k In comparison, if the particle's error function value is better, then update. p i ( k ),calculate SP ( k ); Update the positions of particles in the particle swarm according to the velocity update formula and the position update formula. x i ( k ) and speed v i ( k ), calculate inertia weight ω i ( k ); No. i The formula for updating the position of each particle is: ； No. i The velocity update formula for each particle is: ； In the formula, ω ( k ) is the inertia weight. ε 1 and ε 2 is the learning factor. R 1 and R 2 is a random value between [0, 1]. p i ( k ) is the particle in the first k The historical best position in the next iteration. g best ( k ) is the first k The globally optimal position found by the entire population in the next iteration. x i ( k ) is the first i The particle in the first k The position of the next iteration. v i ( k ) is the first i The particle is at the... k The speed of each iteration; Calculate the error function value of the particle's current position and update the historical best position. p i ( k and the global optimal position g d ( k Based on the globally optimal position g best ( k and the position of the particles x i ( k Calculate the value of the inertia weight for the current iteration. ω The updated formula is: in, g best ( k +1) is the first i The particle in the first k+ The global optimal position in one iteration. p i ( k+ 1) is the first i The particle in the first k+ The historical best position in one iteration f ( p i ( k+ 1)) is the first i The particle in the first k+ Fitness value in 1st iteration x i ( k+ 1) is the first i The particle in the first k+ The position of the first iteration, SP ( k +1) represents the diversity of the particle swarm; The algorithm stops when the number of iterations reaches the maximum number of iterations, and the optimal parameter values ​​of the initial prediction model for nitrogen oxides in the catalytic cracking regenerated flue gas are obtained. The target prediction model is used to predict the preprocessed test data to obtain the predicted value of nitrogen oxides.

2. The method for predicting nitrogen oxides in catalytic cracking regenerated flue gas according to claim 1, characterized in that, The preprocessing of variable data to obtain the processed sample set includes: Get all variable data; The variable data is cleaned; The cleaned variable data is normalized to obtain normalized data; The normalized data is dimensionality reduced to obtain the input variable data set. A preset number of data sets are selected from the input variable data set as a sample set.

3. The method for predicting nitrogen oxides in catalytic cracking regenerated flue gas according to claim 2, characterized in that, The cleaning of the variable data includes: Based on the threshold range of each variable's data, outliers in the variable's data are removed; The missing data in each variable data is filled in using a linear interpolation algorithm to obtain cleaned variable data.

4. The method for predicting nitrogen oxides in catalytic cracking regenerated flue gas according to claim 2, characterized in that, The dimensionality reduction process performed on the normalized data yields the input variable data set, including: Principal component analysis was used to reduce the dimensionality of the normalized data, and the variables with correlation coefficients and contribution rates greater than a threshold were used to form the input variable data set.

5. The method for predicting nitrogen oxides in catalytic cracking regenerated flue gas according to claim 2, characterized in that, The variable data includes: pollutants at the inlet and outlet of regenerated flue gas and reactor data, regenerator data, desulfurization and denitrification facility data, product distribution data, and raw material data of the treatment facilities during the operation of the catalytic cracking unit; The reactor data, regenerator data, desulfurization and denitrification facility data, product distribution data, and raw material data of the treatment facilities all include: raw material nitrogen content, regenerator oxygen content, regenerator dense phase content, regeneration main air volume, riser oil slurry feed rate, riser top temperature, outlet flue gas temperature, total feed rate, dust concentration, riser top temperature, regenerator bottom dense phase temperature, and regenerator dilute phase section pressure.

6. The method for predicting nitrogen oxides in catalytic cracking regeneration flue gas according to claim 1, characterized in that, The step of constructing an initial prediction model for nitrogen oxides in catalytic cracking regeneration flue gas based on the sample set includes: Constructing a basic model for nitrogen oxide emissions from catalytic cracking regeneration flue gas based on fuzzy neural networks: ; In the formula, y ( t )yes t The actual output of nitrogen oxides at time t, w( t )=[ w 1( t ), w 2( t ), …, w P ( t [] represents the output weights of the fuzzy neural network. w l ( t )yes t Connect the output neuron and the first l The weights of neurons in each rule layer. P It is the total number of neurons. v l ( t )yes t Time of the first l The output of each rule layer neuron v l ( t The calculation formula for ) is as follows: ； ； In the formula, c j ( t )=[ c 1j ( t ), c 2j ( t ),…, c mj ( t )]yes t Time of the first j The center of each radial basis function neuron, σ j ( t )=[ σ 1j ( t ), σ 2j ( t ),…, σ mj ( t )] yes t Time of the first j The width of a radial basis function neuron. φ j ( t )yes t Time of the first j The output value of each radial basis function neuron, x( t )=[ x 1( t ), x 2( t ),…, x m ( t )]yes t Input to the nitrogen oxide prediction model at any given time; Define the expression for the error function: ； in, z= 1,2,…, Z , Z It is the number of test samples. y d This is the expected output of nitrogen oxides. y This is the actual output of nitrogen oxides. e This is the error in the nitrogen oxide prediction model; The sample set is input into the basic model of nitrogen oxides in the catalytic cracking regeneration flue gas for training to obtain the initial prediction model of nitrogen oxides in the catalytic cracking regeneration flue gas.

7. The method for predicting nitrogen oxides in catalytic cracking regeneration flue gas according to claim 6, characterized in that, SP It is an index characterizing particle diversity, expressed as: ; SP ( k +1) represents the diversity of the particle swarm. d i ( k +1) is the first i The particle and other particles in the second k The minimum Euclidean distance in +1 iterations It is all d i ( k The average value of (+1).

8. The method for predicting nitrogen oxides in catalytic cracking regenerated flue gas according to claim 7, characterized in that, After obtaining the optimal parameter values ​​of the initial prediction model for nitrogen oxides in the catalytic cracking regenerated flue gas, the performance of the prediction model using the optimal parameters is evaluated based on the root mean square error and accuracy, and it is determined whether the prediction error and accuracy are within the preset range. If so, the catalytic cracking regeneration flue gas nitrogen oxide prediction model using the aforementioned optimal parameters will be used as the target prediction model for catalytic cracking regeneration flue gas nitrogen oxides. RMSE The expression is: ； The formula for calculating its prediction accuracy is: ； in, Z It is the number of test samples. y d This is the expected output of nitrogen oxides. y This is the actual output of nitrogen oxides.

9. A device for predicting nitrogen oxides in catalytic cracking regenerated flue gas, characterized in that, include: Controller, for Preprocess the variable data to obtain the processed sample set; An initial prediction model for nitrogen oxides in catalytic cracking regenerated flue gas was constructed based on the sample set. The initial prediction model for nitrogen oxides in the catalytic cracking regeneration flue gas was optimized using a particle swarm optimization algorithm based on flight information to obtain the target prediction model for nitrogen oxides in the catalytic cracking regeneration flue gas, including: Define a particle as representing a neural network, and the population size of the particles is... n The position expression of the particle is: ； In the formula, w i,1 For the first i The weight σ of the first neuron in the fuzzy neural network, represented by each particle. i,1 For the first i c represents the width of the first neuron in the fuzzy neural network, where each particle represents a neuron. i,1 For the first i The center of the first neuron in the fuzzy neural network is represented by a particle; Initialize the position, velocity, and inertial weights of each particle in the particle swarm, and define the population size and the maximum number of iterations of the function; Based on the error function, calculate the error function value for each particle, and then compare the error function value of each particle with the current global optimal position. g best ( k In comparison, if the particle's error function value is better, then update. g best ( k Simultaneously, the error function value of each particle is compared with the current historical best position. p i ( k In comparison, if the particle's error function value is better, then update. p i ( k ),calculate SP ( k ); Update the positions of particles in the particle swarm according to the velocity update formula and the position update formula. x i ( k ) and speed v i ( k ), calculate inertia weight ω i ( k ); No. i The formula for updating the position of each particle is: ； No. i The velocity update formula for each particle is: In the formula, ω ( k ) is the inertia weight. ε 1 and ε 2 is the learning factor. R 1 and R 2 is a random value between [0, 1]. p i ( k ) is the particle in the first k The historical best position in the next iteration. g best ( k ) is the first k The globally optimal position found by the entire population in the next iteration. x i ( k ) is the first i The particle in the first k The position of the next iteration. v i ( k ) is the first i The particle is at the... k The speed of each iteration; Calculate the error function value of the particle's current position and update the historical best position. p i ( k and the global optimal position g d ( k Based on the globally optimal position g best ( k and the position of the particles x i ( k Calculate the value of the inertia weight for the current iteration. ω The updated formula is: in, g best ( k +1) is the first i The particle in the first k+ The global optimal position in one iteration. p i ( k+ 1) is the first i The particle in the first k+ The historical best position in one iteration f ( p i ( k+ 1)) is the first i The particle in the first k+ Fitness value in 1st iteration x i ( k+ 1) is the first i The particle in the first k+ The position of the first iteration, SP ( k +1) represents the diversity of the particle swarm; The algorithm stops when the number of iterations reaches the maximum number of iterations, and the optimal parameter values ​​of the initial prediction model for nitrogen oxides in the catalytic cracking regenerated flue gas are obtained. The target prediction model is used to predict the preprocessed test data to obtain the predicted value of nitrogen oxides.

10. The catalytic cracking regeneration flue gas nitrogen oxide prediction device according to claim 9, characterized in that, The preprocessing of variable data to obtain the processed sample set includes: Get all variable data; The variable data is cleaned; The cleaned variable data is normalized to obtain normalized data; The normalized data is dimensionality reduced to obtain the input variable data set. A preset number of data sets are selected from the input variable data set as a sample set.

11. The catalytic cracking regeneration flue gas nitrogen oxide prediction device according to claim 10, characterized in that, The cleaning of the variable data includes: Based on the threshold range of each variable's data, outliers in the variable's data are removed; The missing data in each variable data is filled in using a linear interpolation algorithm to obtain cleaned variable data.

12. The catalytic cracking regeneration flue gas nitrogen oxide prediction device according to claim 10, characterized in that, The dimensionality reduction process performed on the normalized data yields the input variable data set, including: Principal component analysis was used to reduce the dimensionality of the normalized data, and the variables with correlation coefficients and contribution rates greater than a threshold were used to form the input variable data set.

13. The catalytic cracking regeneration flue gas nitrogen oxide prediction device according to claim 9, characterized in that, The step of constructing an initial prediction model for nitrogen oxides in catalytic cracking regeneration flue gas based on the sample set includes: Constructing a basic model for nitrogen oxide emissions from catalytic cracking regeneration flue gas based on fuzzy neural networks: ; In the formula, y ( t )yes t The actual output of nitrogen oxides at time t, w( t )=[ w 1( t ), w 2( t ), …, w P ( t [] represents the output weights of the fuzzy neural network. w l ( t )yes t Connect the output neuron and the first l The weights of neurons in each rule layer. P It is the total number of neurons. v l ( t )yes t Time of the first l The output of each rule layer neuron v l ( t The calculation formula for ) is as follows: ； ； In the formula, c j ( t )=[ c 1j ( t ), c 2j ( t ),…, c mj ( t )]yes t Time of the first j The center of each radial basis function neuron, σ j ( t )=[ σ 1j ( t ), σ 2j ( t ),…, σ mj ( t )] yes t Time of the first j The width of a radial basis function neuron. φ j ( t )yes t Time of the first j The output value of each radial basis function neuron, x( t )=[ x 1( t ), x 2( t ),…, x m ( t )]yes t Input to the nitrogen oxide prediction model at any given time; Define the expression for the error function: ； in, z= 1,2,…, Z , Z It is the number of test samples. y d This is the expected output of nitrogen oxides. y This is the actual output of nitrogen oxides. e This is the error in the nitrogen oxide prediction model; The sample set is input into the basic model of nitrogen oxides in the catalytic cracking regeneration flue gas for training to obtain the initial prediction model of nitrogen oxides in the catalytic cracking regeneration flue gas.

14. The catalytic cracking regeneration flue gas nitrogen oxide prediction device according to claim 13, characterized in that, SP It is an index characterizing particle diversity, expressed as: ; SP ( k +1) represents the diversity of the particle swarm. d i ( k +1) is the first i The particle and other particles in the second k The minimum Euclidean distance in +1 iterations It is all d i ( k The average value of (+1).

15. The catalytic cracking regeneration flue gas nitrogen oxide prediction device according to claim 14, characterized in that, After obtaining the optimal parameter values ​​of the initial prediction model for nitrogen oxides in the catalytic cracking regenerated flue gas, the performance of the prediction model using the optimal parameters is evaluated based on the root mean square error and accuracy, and it is determined whether the prediction error and accuracy are within the preset range. If so, the catalytic cracking regeneration flue gas nitrogen oxide prediction model using the aforementioned optimal parameters will be used as the target prediction model for catalytic cracking regeneration flue gas nitrogen oxides. RMSE The expression is: ； The formula for calculating its prediction accuracy is: ； in, Z It is the number of test samples. y d This is the expected output of nitrogen oxides. y This is the actual output of nitrogen oxides.

16. A nitrogen oxide prediction system for catalytic cracking regeneration flue gas, characterized in that, The system includes: The data processing module is used to preprocess the variable data to obtain the processed sample set; The initial prediction model construction module is used to construct an initial prediction model for nitrogen oxides in catalytic cracking regeneration flue gas based on the sample set. The initial prediction model optimization module is used to optimize the initial prediction model of nitrogen oxides in the catalytic cracking regeneration flue gas using a particle swarm optimization algorithm based on flight information, to obtain the target prediction model of nitrogen oxides in the catalytic cracking regeneration flue gas, including: Define a particle as representing a neural network, and the population size of the particles is... n The position expression of the particle is: ； In the formula, w i,1 For the first i The weight σ of the first neuron in the fuzzy neural network, represented by each particle. i,1 For the first i c represents the width of the first neuron in the fuzzy neural network, where each particle represents a neuron. i,1 For the first i The center of the first neuron in the fuzzy neural network is represented by a particle; Initialize the position, velocity, and inertial weights of each particle in the particle swarm, and define the population size and the maximum number of iterations of the function; Based on the error function, calculate the error function value for each particle, and then compare the error function value of each particle with the current global optimal position. g best ( k In comparison, if the particle's error function value is better, then update. g best ( k Simultaneously, the error function value of each particle is compared with the current historical best position. p i ( k In comparison, if the particle's error function value is better, then update. p i ( k ),calculate SP ( k ); Update the positions of particles in the particle swarm according to the velocity update formula and the position update formula. x i ( k ) and speed v i ( k ), calculate inertia weight ω i ( k ); No. i The formula for updating the position of each particle is: ； No. i The velocity update formula for each particle is: In the formula, ω ( k ) is the inertia weight. ε 1 and ε 2 is the learning factor. R 1 and R 2 is a random value between [0, 1]. p i ( k ) is the particle in the first k The historical best position in the next iteration. g best ( k ) is the first k The globally optimal position found by the entire population in the next iteration. x i ( k ) is the first i The particle in the first k The position of the next iteration. v i ( k ) is the first i The particle is at the... k The speed of each iteration; Calculate the error function value of the particle's current position and update the historical best position. p i ( k and the global optimal position g d ( k Based on the globally optimal position g best ( k and the position of the particles x i ( k Calculate the value of the inertia weight for the current iteration. ω The updated formula is: in, g best ( k +1) is the first i The particle in the first k+ The global optimal position in one iteration. p i ( k+ 1) is the first i The particle in the first k+ The historical best position in one iteration f ( p i ( k+ 1)) is the first i The particle in the first k+ Fitness value in 1st iteration x i ( k+ 1) is the first i The particle in the first k+ The position of the first iteration, SP ( k +1) represents the diversity of the particle swarm; The algorithm stops when the number of iterations reaches the maximum number of iterations, and the optimal parameter values ​​of the initial prediction model for nitrogen oxides in the catalytic cracking regenerated flue gas are obtained. The data prediction module is used to predict the preprocessed test data using the target prediction model to obtain the predicted value of nitrogen oxides.

17. A machine-readable storage medium storing instructions for causing a machine to perform the method for predicting nitrogen oxides in catalytic cracking regeneration flue gas according to any one of claims 1-8 of this application.

Images