Propylene mass fraction prediction method, system and medium

Abstract

The invention provides a propylene mass fraction prediction method, a propylene mass fraction prediction system and a computer readable storage medium. The method for predicting the mass fraction of propylene comprises the following steps: acquiring data of a propylene rectifying tower; constructing a time delay matrix based on the time sequence characteristics of the propylene rectifying tower data, and performing dimension reduction on the propylene rectifying tower data by adopting principal component analysis based on the time delay matrix so as to extract characteristic data; and inputting the characteristic data into a bidirectional long-short-term memory network model to predict the propylene mass fraction, and determining parameters of the bidirectional long-short-term memory network model according to a pelican optimization algorithm. According to the prediction method for the propylene mass fraction provided by the invention, the prediction precision and the model robustness can be improved, and a more reliable and efficient solution is provided for process control of a propylene rectifying tower.

Description

Technical Field

[0001] The invention relates to the field of ethylene production distillation towers, and in particular to a propylene mass fraction prediction method, a propylene mass fraction prediction system, and a computer-readable storage medium. Background Art

[0002] Propylene is an important chemical raw material, widely used in industries such as plastics, rubber and textiles. The propylene distillation tower is a key equipment in the petrochemical production process, mainly used to separate and purify propylene. The operating parameters of the propylene distillation tower, such as temperature, pressure and feed flow, directly affect the purity and yield of the product. Therefore, accurate monitoring and control of the operating parameters of the propylene distillation tower is of great significance to improving propylene quality and production efficiency.

[0003] The traditional hard measurement method of propylene mass fraction relies on the installation of a large number of expensive and complex sensor equipment, which not only increases costs but may also cause production interruptions when equipment fails. In addition, in the actual production process, the operating environment of the propylene distillation tower is complex and changeable, and there are highly nonlinear and coupled relationships between the operating parameters, which makes it very difficult to predict and control the parameters through traditional physical models. Existing soft measurement technology uses easy-to-measure variables to indirectly predict parameters that are difficult to measure directly by establishing mathematical models, such as multivariate linear regression and support vector machines. However, existing technologies are often unable to fully capture the dynamic changes and nonlinear characteristics of the operating parameters of the propylene distillation tower, resulting in insufficient prediction accuracy and model robustness, limiting its application in actual production.

[0004] In order to overcome the above-mentioned defects of the prior art, there is an urgent need in the art for a propylene mass fraction prediction technology that can improve the prediction accuracy and the robustness of the model and provide a more reliable and efficient solution for the process control of the propylene distillation tower. Summary of the invention

[0005] A brief summary of one or more aspects is given below to provide a basic understanding of these aspects. This summary is not an exhaustive overview of all conceived aspects, and is neither intended to identify the key or critical elements of all aspects nor to define the scope of any or all aspects. Its only purpose is to give some concepts of one or more aspects in a simplified form as a prelude to a more detailed description that will be given later.

[0006] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method for predicting the mass fraction of propylene, a system for predicting the mass fraction of propylene, and a computer-readable storage medium, which can improve the prediction accuracy and the robustness of the model, and provide a more reliable and efficient solution for the process control of the propylene distillation tower.

[0007] Specifically, the above-mentioned propylene mass fraction prediction method provided according to the first aspect of the present invention includes the steps of: acquiring propylene distillation tower data; constructing a time delay matrix based on the timing characteristics of the propylene distillation tower data, and using principal component analysis based on the time delay matrix to reduce the dimension of the propylene distillation tower data to extract feature data; and inputting the feature data into a bidirectional long short-term memory network model to predict the propylene mass fraction, and the hyperparameters of the bidirectional long short-term memory network model are determined according to the Pelican optimization algorithm.

[0008] Preferably, in one embodiment of the present invention, the bidirectional long short-term memory network model includes a forward long short-term memory network calculation process and a reverse long short-term memory network calculation process.

[0009] Preferably, in one embodiment of the present invention, the forward long short-term memory network calculation process and the reverse long short-term memory network calculation process include the calculation of a forget gate, an input gate, candidate memory cells, memory cell updates, an output gate, and hidden state updates.

[0010] Preferably, in one embodiment of the present invention, the calculation process of the forget gate of the forward long short-term memory network calculation process is:

[0011]

[0012] Among them, σ is the sigmoid activation function, is the weight matrix of the forget gate in the forward LSTM network calculation process, x t Represents the input at the current moment, represents the hidden state of the previous moment, b f is the bias matrix;

[0013] The calculation process of the input gate of the forward long short-term memory network calculation process is:

[0014]

[0015] in, is the weight matrix of the input gate of the forward LSTM network calculation process, b i is the bias matrix;

[0016] The calculation process of the candidate memory units in the forward long short-term memory network calculation process is as follows:

[0017]

[0018] in, is the weight matrix of the candidate hidden state in the forward LSTM network calculation process, b C is the bias matrix;

[0019] The calculation process of the memory unit update in the forward long short-term memory network calculation process is:

[0020]

[0021] ⊙ is the element-wise product, is the output of the input gate;

[0022] The calculation process of the output gate of the forward long short-term memory network calculation process is:

[0023]

[0024] in, is the weight matrix of the output gate of the forward LSTM network calculation process, b o is the bias matrix;

[0025] The calculation process of the hidden state update of the forward long short-term memory network calculation process is:

[0026]

[0027] in, is the update result of the memory unit at the current moment;

[0028] The calculation process of the forget gate of the reverse long short-term memory network calculation process is:

[0029]

[0030] Among them, σ is the sigmoid activation function, is the weight matrix of the forget gate in the reverse long short-term memory network calculation process, x t Represents the input at the current moment, represents the hidden state at the next moment, b f is the bias matrix;

[0031] The calculation process of the input gate of the reverse long short-term memory network calculation process is:

[0032]

[0033] in, is the weight matrix of the input gate of the reverse long short-term memory network calculation process, b i is the bias matrix;

[0034] The calculation process of the candidate memory units in the reverse long short-term memory network calculation process is as follows:

[0035]

[0036] in, is the weight matrix of the candidate hidden state in the reverse long short-term memory network calculation process, b C is the bias matrix;

[0037] The calculation process of memory unit update in the reverse long short-term memory network calculation process is:

[0038]

[0039] ⊙ is the element-wise product, is the output of the input gate;

[0040] The calculation process of the output gate of the reverse long short-term memory network calculation process is:

[0041]

[0042] in, is the weight matrix of the output gate of the reverse long short-term memory network calculation process, b o is the bias matrix;

[0043] The calculation process of the hidden state update of the reverse long short-term memory network calculation process is:

[0044]

[0045] in, is the update result of the memory unit at the current moment.

[0046] Preferably, in one embodiment of the present invention, the output of the bidirectional long short-term memory network model is the concatenation of the hidden state of the forward long short-term memory network calculation process and the hidden state of the reverse long short-term memory network calculation process.

[0047] Preferably, in one embodiment of the present invention, the step of determining the hyperparameters of the bidirectional long short-term memory network model according to the Pelican optimization algorithm includes: step S1: initializing the initial position, speed and fitness of each individual of the Pelican optimization algorithm, and the individual is a set of hyperparameters of the bidirectional long short-term memory network model; step S2: using each individual to train the bidirectional long short-term memory network model, and determining the fitness value of each individual by verifying the error of the bidirectional long short-term memory network model after training; step S3: determining the historical optimal position and the global optimal position of each individual according to the fitness value; step S4: updating the position and speed of each individual based on the historical optimal position and the global optimal position of each individual; and step S5: repeating steps S2 to S4 until the optimal solution is determined when the termination condition is met, and determining the hyperparameters of the bidirectional long short-term memory network model according to the hyperparameter set corresponding to the optimal solution.

[0048] Preferably, in an embodiment of the present invention, in step S4, the formula for updating the position and speed of each individual is:

[0049]

[0050] X i (t+1)=X i (t)+V i (t+1),

[0051] Among them, ω represents the inertia weight, c1, c2, c3 are acceleration constants used to control the acceleration of the individual to the historical optimal position and the global optimal position, r1, r2, r2 are random numbers between 0 and 1, represents the historical optimal position of the individual i, represents the global optimal position, l i is used to help the individual i to guide to a target solution position during the search process, X i (t) is the position of the individual i at the tth iteration, V i (t) is the velocity of the individual i at the tth iteration.

[0052] Preferably, in one embodiment of the present invention, the step of constructing a time delay matrix based on the timing characteristics of the propylene distillation tower data, and reducing the dimension of the propylene distillation tower data by principal component analysis based on the time delay matrix to extract characteristic data includes: constructing a time delay matrix based on the timing characteristics of the propylene distillation tower data, and calculating the covariance matrix of the time delay matrix; performing eigenvalue decomposition based on the covariance matrix, and selecting eigenvectors corresponding to the first K eigenvalues ​​as principal components; reducing the dimension of the propylene distillation tower data by using the principal components to obtain reduced dimension data; and performing cluster analysis on the reduced dimension data to extract the characteristic data.

[0053] Preferably, in an embodiment of the present invention, the step of acquiring the propylene distillation tower data comprises: preprocessing the sample data to determine the propylene distillation tower data, wherein the preprocessing comprises processing of abnormal values ​​and processing of missing values.

[0054] Preferably, in an embodiment of the present invention, the processing of missing values ​​includes filling the missing values ​​of the sample data using a multiple interpolation algorithm.

[0055] Preferably, in one embodiment of the present invention, the propylene distillation tower data include one or more of propylene feed amount, propane feed amount, methane feed amount, acetylene feed amount, ethane feed amount, ethylene feed amount, propadiene feed amount, propyne feed amount, butadiene feed amount, 1-butene feed amount, 1-3 butadiene feed amount, diacetylene feed amount, n-butane feed amount, isobutane feed amount, feed temperature, reflux ratio, reboiler heat duty and distillate flow rate.

[0056] In addition, the prediction system for propylene mass fraction provided according to the second aspect of the present invention includes a memory and a processor. The memory stores computer instructions. The processor is connected to the memory and is configured to execute the computer instructions stored in the memory to implement the prediction method for propylene mass fraction provided in any one of the above embodiments.

[0057] In addition, the computer-readable storage medium provided according to the third aspect of the present invention stores computer instructions, which, when executed by a processor, implement the method for predicting the mass fraction of propylene provided in any one of the above embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The above features and advantages of the present invention can be better understood after reading the detailed description of the embodiments of the present disclosure in conjunction with the following drawings. In the drawings, the components are not necessarily drawn to scale, and components with similar related properties or features may have the same or similar reference numerals.

[0059] Figure 1A schematic diagram of a prediction system for propylene mass fraction provided according to some embodiments of the present invention is shown;

[0060] Figure 2 A flow chart showing a method for predicting propylene mass fraction according to some embodiments of the present invention is shown;

[0061] Figure 3 A structural diagram of a bidirectional long short-term memory network model provided according to some embodiments of the present invention is shown;

[0062] Figure 4 A flowchart of the Pelican optimization algorithm provided according to some embodiments of the present invention is shown;

[0063] Figure 5 A prediction error diagram of a method for predicting propylene mass fraction according to an embodiment of the present invention is shown;

[0064] Figure 6 A comparison diagram of evaluating multiple prediction models provided in a preferred embodiment 1 of the present invention is shown; and

[0065] Figure 7 A comparative test diagram of multiple prediction models provided according to the second preferred embodiment of the present invention is shown.

[0066] Reference numerals:

[0067] 100: Propylene mass fraction prediction system;

[0068] 110: memory;

[0069] 111: Computer storage media;

[0070] 120: Processor;

[0071] 301: Forward long short-term memory network unit layer;

[0072] 302: Reverse long short-term memory network unit layer;

[0073] 701: Model;

[0074] 702: Bidirectional Gated Recurrent Unit Neural Network;

[0075] 703: Traditional Long Short-Term Memory Networks;

[0076] S1-S5: steps; and

[0077] S210~S230: steps. DETAILED DESCRIPTION

[0078] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. Note that the aspects described below in conjunction with the accompanying drawings and specific embodiments are only exemplary and should not be construed as limiting the scope of protection of the present invention in any way.

[0079] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0080] In addition, the terms "upper", "lower", "left", "right", "top", "bottom", "horizontal" and "vertical" used in the following description should be understood as the directions shown in the paragraph and the related drawings. Such relative terms are only used for the convenience of description and do not mean that the device described therein must be manufactured or operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0081] It is understood that although the terms "first", "second", "third", etc. may be used herein to describe various components, regions, layers and / or parts, these components, regions, layers and / or parts should not be limited by these terms, and these terms are only used to distinguish different components, regions, layers and / or parts. Therefore, the first component, region, layer and / or part discussed below may be referred to as a second component, region, layer and / or part without departing from some embodiments of the present invention.

[0082] As mentioned above, the traditional hard measurement method of propylene mass fraction relies on the installation of a large number of expensive and complex sensor equipment, which not only increases the cost but also may cause production interruption when the equipment fails. In addition, in the actual production process, the operating environment of the propylene distillation tower is complex and changeable, and there are highly nonlinear and coupled relationships between the operating parameters, which makes it very difficult to predict and control the parameters through traditional physical models. Existing soft measurement technology uses easy-to-measure variables to indirectly predict parameters that are difficult to measure directly by establishing mathematical models, such as multivariate linear regression and support vector machines. However, existing technologies are often unable to fully capture the dynamic changes and nonlinear characteristics of the operating parameters of the propylene distillation tower, resulting in insufficient prediction accuracy and model robustness, which limits its application in actual production.

[0083] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method for predicting the mass fraction of propylene, a system for predicting the mass fraction of propylene, and a computer-readable storage medium, which can improve the prediction accuracy and the robustness of the model, and provide a more reliable and efficient solution for the process control of the propylene distillation tower.

[0084] In some non-limiting embodiments, the propylene mass fraction prediction method provided in the first aspect of the present invention may be implemented via the propylene mass fraction prediction system provided in the second aspect of the present invention.

[0085] Please refer to Figure 1 , Figure 1 A schematic diagram of a system for predicting propylene mass fraction according to some embodiments of the present invention is shown.

[0086] like Figure 1 As shown, the prediction system 100 for propylene mass fraction may be configured with a memory 110 and a processor 120. The memory 110 includes but is not limited to the computer-readable storage medium 111 provided in the third aspect of the present invention, on which computer instructions are stored. The processor 120 is connected to the memory 110 and is configured to execute the computer instructions stored in the memory 110 to implement the prediction method for propylene mass fraction provided in the first aspect of the present invention.

[0087] In some embodiments, the prediction system 100 for propylene mass fraction may include multiple program modules. The program modules may be stored in the memory 110. The program modules include, but are not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or a combination thereof may include the implementation of a network environment. The program modules generally perform the functions and / or methods of the embodiments described in the present invention.

[0088] In a non-limiting embodiment, the prediction system 100 for propylene mass fraction may include an online analysis unit and an offline training unit. The online analysis unit may include a data input module and a prediction value output module, and the offline training unit may include a model input module, a modeling module, a pelican optimization module, and a model verification module. The model input module may be used to obtain a data set for building a model; the modeling module may fit a data model based on the training data; the pelican optimization module may be used to optimize the hyperparameters of the fitted bidirectional long short-term memory network model; and the model verification module may be used to verify the optimized model. This embodiment only provides a specific solution that is convenient for the public to implement, and is not used to limit all functions or all working methods of the prediction system 100 for propylene mass fraction.

[0089] The following will first describe the working principle of the above-mentioned propylene mass fraction prediction system in conjunction with some embodiments of the propylene mass fraction prediction method. Those skilled in the art will understand that these embodiments of the propylene mass fraction prediction method are only some non-limiting implementation methods provided by the present invention, which are intended to clearly demonstrate the main concept of the present invention and provide some specific schemes that are convenient for the public to implement, rather than to limit all functions or all working methods of the propylene mass fraction prediction system. Similarly, the propylene mass fraction prediction system is also only a non-limiting implementation method provided by the present invention, and does not constitute a limitation on the execution subject and execution order of each step in these propylene mass fraction prediction methods.

[0090] Please refer to Figure 2 , Figure 2 A flow chart of a method for predicting propylene mass fraction provided according to some embodiments of the present invention is shown.

[0091] like Figure 2 As shown, the method 200 for predicting the mass fraction of propylene includes step S210: acquiring propylene distillation tower data.

[0092] The prediction system of propylene mass fraction can obtain sample data from the distributed control system of the actual factory, and the propylene distillation tower data can be determined according to the sample data. The propylene distillation tower data can be the auxiliary variables preliminarily selected from the sample data, and specifically, the prediction system can preliminarily select the relevant auxiliary variables according to the mechanism analysis and prior knowledge. In some embodiments, the selected auxiliary variables may include one or more of propylene feed amount, propane feed amount, methane feed amount, acetylene feed amount, ethane feed amount, ethylene feed amount, propadiene feed amount, propyne feed amount, butadiene feed amount, 1-butene feed amount, 1-3 butadiene feed amount, diacetylene feed amount, n-butane feed amount, isobutane feed amount, feed temperature, reflux ratio, reboiler heat load and distillate flow rate.

[0093] The collected sample data often contain random errors and negligent errors, including the influence of noise and some missing values ​​or outliers. Although the probability of these values ​​appearing is small, they will significantly reduce the data quality and thus affect the final model performance.

[0094] Step S210 may also include preprocessing of the sample data, where the preprocessing may include processing of outliers and processing of missing values.

[0095] The prediction system can perform data cleaning on the acquired sample data. Specifically, the prediction system can remove outliers in the sample data. First, calculate the mean and standard deviation of the data. For the data set X = [x1, x2, ..., x A ], the data set includes A data x jk , data xjk represents the k-th dimension variable of the j-th data. Calculate each data x jk The mean μ and standard deviation s of :

[0096]

[0097] An outlier can be a data point that exceeds the mean μ plus or minus q times the standard deviation s, that is, a data point that exceeds the upper bound or the lower bound:

[0098] Lower Bound=μ-qs,

[0099] Upper Bound=μ+qs,

[0100] In some embodiments, q may be 3. In this way, the prediction system may remove outliers in the sample data.

[0101] Afterwards, the prediction system can process the missing values ​​in the sample data. The processing of missing values ​​can include filling the missing values ​​of the sample data using a multiple interpolation algorithm.

[0102] First, the prediction system can check for missing values ​​in the sample data. Specifically, a logical variable M is used to represent the missing values. M(I,J)=1 indicates that the data point x in the sample data is IJ Missing, M(I,J)=0 means data point x IJ exists. Then, for missing values, the prediction system can use a multiple imputation algorithm to fill in the missing values. Based on the multiple imputation algorithm, the calculation formula for the data value used for filling is as follows:

[0103]

[0104] T0 and T1 can respectively represent the missing values ​​at the time near the corresponding time point T, and x(T) represents the data for filling corresponding to time T. Repeat the steps until the missing values ​​in the data set are filled with the determined data values.

[0105] In some embodiments, the prediction system may repeatedly perform the process of removing outliers in the sample data and processing missing values ​​in the sample data until there are no outliers in the sample data. Furthermore, the prediction system may also perform standardization processing on the sample data.

[0106] The prediction system improves the availability and accuracy of the processed sample data by filtering outliers in the sample data and performing multiple interpolation on missing values. Through data processing, the prediction system improves the fitness of the collected original sample data.

[0107] like Figure 2As shown, the prediction method 200 for propylene mass fraction includes step S220: constructing a time delay matrix based on the time series characteristics of the propylene distillation tower data, and performing dimension reduction on the propylene distillation tower data by principal component analysis based on the time delay matrix to extract characteristic data.

[0108] Since there may be coupling and redundancy between the variables of the propylene distillation tower data, and too many input features will lead to a decrease in the performance of the neural network, in order to simplify the model structure and reduce the number of input variables of the neural network, principal component analysis is used to reduce the dimension of the propylene distillation tower data. The prediction system can perform principal component analysis on the processed sample data.

[0109] In addition, the numerical differences between the features are large, with different dimensions and value ranges. If the propylene distillation tower data is not normalized, the features with larger scales may dominate the modeling process, resulting in a decrease in model performance. Therefore, the processed sample data for principal component analysis can be the data that has been standardized.

[0110] In some embodiments, the prediction system may use normalization to unify all feature scales to accelerate model convergence and improve the efficiency of the training process. For example, the prediction system may use minimum-maximum normalization to map all features to a range of 0 to 1. The formula for minimum-maximum normalization is as follows:

[0111]

[0112] Among them, Y z is the normalized eigenvalue, X z is the eigenvalue before normalization, X min and X max are the minimum and maximum values ​​of each feature, respectively. In some embodiments, the prediction system may divide the normalized data during offline training to construct subsequent prediction models, for example, using 70% of the data as a training set and the remaining 30% of the data as a test set.

[0113] The prediction system can construct a time delay matrix based on the time series characteristics of the propylene distillation tower data, thereby capturing the time dynamic characteristics of the data. The calculation method of the constructed time delay matrix can be as follows:

[0114]

[0115] Among them, X represents the original sample data matrix, the size of X is n×m, where n is the number of samples, m is the number of features, and t crepresents the current moment, Ts is the number of valid time steps in the data matrix, defined as Ts = nd, that is, the number of time steps remaining after removing the first d moments. Each row in the matrix contains the data of the current moment and the first d moments.

[0116] Afterwards, the prediction system can calculate the covariance matrix C of the time delay matrix:

[0117]

[0118] Then, the prediction system can perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalue matrix Λ and the eigenvector matrix V:

[0119] CV=VΛ,

[0120] Where Λ is the diagonal eigenvalue matrix and V is the eigenvector matrix.

[0121] The prediction system can select the eigenvectors corresponding to the first K eigenvalues ​​as the principal components, and then use the principal components to reduce the dimension of the propylene distillation tower data to obtain reduced-dimensional data. Specifically, select the eigenvectors corresponding to the first K largest eigenvalues ​​to form a new eigenvector matrix V K ; Use the selected principal components to reduce the dimension of the processed sample data to obtain the reduced dimension data Z:

[0122] Z=X d V K .

[0123] In this way, the prediction system can reduce data redundancy by sorting the contributions of each principal component and selecting the main components.

[0124] Finally, the prediction system can perform cluster analysis on the dimension-reduced data Z to extract feature data.

[0125] In view of the complexity and high-dimensional characteristics of the data, the prediction system uses principal component analysis to reduce the dimension of the propylene distillation tower data, which can reduce the complexity of the propylene distillation tower data while retaining the characteristics of the propylene distillation tower data. Moreover, during the training process of the prediction system, principal component analysis of sample data can effectively reduce the training time of the model, which facilitates subsequent process control and optimization.

[0126] Please continue to refer to Figure 2 The method 200 for predicting the mass fraction of propylene includes step S230: inputting the characteristic data into a bidirectional long short-term memory network model to predict the mass fraction of propylene, and the hyperparameters of the bidirectional long short-term memory network model are determined according to the Pelican optimization algorithm.

[0127] Deep learning technology has shown strong capabilities in processing nonlinear and high-dimensional data and is widely used in fields such as image recognition and natural language processing. In particular, recurrent neural networks (RNN) and their variants, such as long short-term memory networks (LSTM) and gated recurrent units (GRU), have performed well in time series prediction.

[0128] During the online analysis process of the prediction system, the prediction system can predict the propylene mass fraction based on the trained and verified bidirectional long short-term memory network (Bi-LSTM) model through the input processed propylene distillation tower data. The hyperparameters of the bidirectional long short-term memory network model used for propylene mass fraction prediction in the prediction system can be determined according to the Pelican optimization algorithm.

[0129] Furthermore, during the offline training process of the prediction system, the prediction system may use the input processed propylene distillation tower data as training data to fit the bidirectional long short-term memory network model.

[0130] Please refer to Figure 3 , Figure 3 A structural diagram of a bidirectional long short-term memory network model provided according to some embodiments of the present invention is shown.

[0131] like Figure 3 As shown, the bidirectional LSTM model may include a layer of forward LSTM unit layer 301 and a layer of reverse LSTM unit layer 302. The forward LSTM unit layer 301 and the reverse LSTM unit layer 302 include multiple LSTM units. The input at the current time t is x t , the input at the previous time t-1 is x t-1 , the input at the next moment t+1 is x t+1 The hidden state at the current time t is h t , the hidden state at the previous moment t-1 is h t-1 , the hidden state at the next moment t+1 is h t+1 In the forward LSTM network unit layer 301, the hidden state h at the previous moment t-1 is passed to the LSTM network unit at the next moment t; in the reverse LSTM network unit layer 302, the hidden state h at the next moment t+1 Passed to the long short-term memory network unit of the previous moment.

[0132] The bidirectional long short-term memory network model may include a forward long short-term memory network calculation process and a reverse long short-term memory network calculation process. In some embodiments, the forward long short-term memory network calculation process and the reverse long short-term memory network calculation process may include calculations of a forget gate, an input gate, a candidate memory cell, a memory cell update, an output gate, and a hidden state update.

[0133] Specifically, the calculation process of the forget gate in the forward long short-term memory network calculation process can be:

[0134]

[0135] Among them, σ is the sigmoid activation function, is the weight matrix of the forget gate in the forward LSTM network calculation process, x t Represents the input at the current moment, represents the hidden state of the previous moment, b f is the bias matrix;

[0136] The calculation process of the input gate of the forward long short-term memory network calculation process can be:

[0137]

[0138] in, is the weight matrix of the input gate of the forward LSTM network calculation process, b i is the bias matrix;

[0139] The calculation process of candidate memory units in the forward long short-term memory network calculation process can be:

[0140]

[0141] in, is the weight matrix of the candidate hidden state in the forward long short-term memory network calculation process, b C is the bias matrix;

[0142] The calculation process of memory unit update in the forward long short-term memory network calculation process can be:

[0143]

[0144] ⊙ is the element-wise product, It is the output of the input gate, which is used to determine how much of the current input content is added to the memory;

[0145] The calculation process of the output gate of the forward long short-term memory network calculation process can be:

[0146]

[0147] in, is the weight matrix of the output gate of the forward LSTM network calculation process, b o is the bias matrix;

[0148] The calculation process of the hidden state update in the forward long short-term memory network calculation process can be:

[0149]

[0150] in, is the update result of the memory unit at the current moment.

[0151] The calculation process of the forget gate of the reverse long short-term memory network calculation process can be:

[0152]

[0153] Among them, σ is the sigmoid activation function, is the weight matrix of the forget gate in the reverse long short-term memory network calculation process, x t Represents the input at the current moment, represents the hidden state at the next moment, b f is the bias matrix;

[0154] The calculation process of the input gate of the reverse long short-term memory network calculation process can be:

[0155]

[0156] in, is the weight matrix of the input gate of the reverse long short-term memory network calculation process, b i is the bias matrix;

[0157] The calculation process of candidate memory units in the reverse long short-term memory network calculation process can be:

[0158]

[0159] in, is the weight matrix of the candidate hidden state in the reverse long short-term memory network calculation process, b C is the bias matrix;

[0160] The calculation process of memory unit update in the reverse long short-term memory network calculation process can be:

[0161]

[0162] ⊙ is the element-wise product, It is the output of the input gate, which is used to determine how much of the current input content is added to the memory;

[0163] The calculation process of the output gate of the reverse long short-term memory network calculation process can be:

[0164]

[0165] in, is the weight matrix of the output gate of the reverse long short-term memory network calculation process, b o is the bias matrix;

[0166] The calculation process of the hidden state update of the reverse long short-term memory network calculation process can be:

[0167]

[0168] in, is the update result of the memory unit at the current moment.

[0169] The output of the bidirectional long short-term memory network model can be the concatenation of the hidden state of the forward long short-term memory network calculation process and the hidden state of the reverse long short-term memory network calculation process. For example, in the above embodiment, the final output of the bidirectional long short-term memory network model is the concatenation of the hidden states of the forward long short-term memory network and the reverse long short-term memory network:

[0170]

[0171] Relying only on the bidirectional long short-term memory network model for soft measurement modeling still faces problems such as difficulty in optimizing model parameters and long training time. Therefore, during offline training, the prediction system can adjust and determine the optimized hyperparameters according to the Pelican optimization algorithm.

[0172] As a global optimization algorithm, the Pelican Optimization Algorithm (POA) can effectively optimize the hyperparameters of complex bidirectional long short-term memory network models by simulating the flight and foraging behavior of pelicans to find the optimal solution. Therefore, combining the deep learning bidirectional long short-term memory network model with the Pelican Optimization Algorithm can improve the prediction accuracy of the prediction system and the robustness of the bidirectional long short-term memory network model, providing a more reliable and efficient solution for the process control of the propylene distillation tower.

[0173] Please refer to Figure 4 , Figure 4 A flowchart of the Pelican optimization algorithm provided according to some embodiments of the present invention is shown.

[0174] like Figure 4As shown, the step of adjusting the hyperparameters according to the Pelican optimization algorithm of the prediction system may include step S1: initializing the initial position, speed and fitness of each individual of the Pelican optimization algorithm, where the individual is a hyperparameter set of the bidirectional long short-term memory network model. Here, the hyperparameter set may include the learning rate, number of trainings and / or number of hidden layer nodes of the bidirectional long short-term memory network model.

[0175] Initializing the individual population can randomly initialize a group of individuals, each of which contains different hyperparameter settings of the Bi-LSTM model. These hyperparameters can include two hidden layer nodes, the number of training times, and / or the learning rate.

[0176] Afterwards, the prediction system can execute step S2: use each individual to train the bidirectional long short-term memory network model, and determine the fitness value of each individual by verifying the error of the trained bidirectional long short-term memory network model.

[0177] For each individual, the prediction system can train the bidirectional long short-term memory network model according to the hyperparameters of the bidirectional long short-term memory network model represented by it. Afterwards, the performance of the bidirectional long short-term memory network model is evaluated using the validation set, and the error (e.g., mean square error) is calculated as the fitness value. The smaller the fitness value, the better the performance of the bidirectional long short-term memory network model. The prediction system can adjust the hyperparameters of the bidirectional long short-term memory network model according to the validation result (i.e., the fitness value) until the performance of the bidirectional long short-term memory network model meets the expected requirements.

[0178] In some embodiments, the prediction system can set various parameters of the Pelican optimization algorithm, determine the initial position and speed of each individual of the Pelican optimization algorithm, train the bidirectional long short-term memory network model and verify it to calculate the initial fitness of each individual. For example, the fitness function value calculation of each individual can be determined by the mean square error (MSE, Mean Squared Error), the formula is as follows:

[0179]

[0180] Where N represents the total number of individuals in the data set, that is, how many data points are there for error calculation. i Represents the true value (or target value) of the i-th individual. Represents the actual result predicted by the model, that is, the predicted value of the i-th individual. It is the result predicted by the model.

[0181] In addition, commonly used model evaluation indicators include root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R2 ), these evaluation indicators can comprehensively evaluate the prediction performance of the model. Those skilled in the art can understand that in other embodiments, those skilled in the art can also use other equivalent methods to determine the fitness function of the Pelican optimization algorithm based on the above concept and relevant knowledge in the field to calculate the error of the bidirectional long short-term memory network and evaluate the prediction performance of the bidirectional long short-term memory network.

[0182] Afterwards, the prediction system may execute step S3: determine the historical optimal position and the global optimal position of each individual according to the fitness value, and adjust the speed and position of the individual according to the historical optimal position and the global optimal position of the individual to iteratively find the optimal solution.

[0183] Specifically, the prediction system can execute step S4: based on the historical optimal position and the global optimal position, update the position and speed of each individual. The formula for updating the position and speed of each individual is:

[0184]

[0185] X i (t+1)=X i (t)+V i (t+1),

[0186] Among them, ω represents the inertia weight, c1, c2, c3 are acceleration constants used to control the acceleration of the individual to the historical optimal position and the global optimal position, r1, r2, r3 are random numbers between 0 and 1, represents the historical optimal position of individual i, represents the global optimal position, l i is used to help individual i to guide to a target solution during the search process. i (t) is the position of individual i in the tth iteration, V i (t) is the velocity of individual i at the tth iteration.

[0187] like Figure 4 As shown, the prediction system may further include step S5: repeating steps S2 to S4 until an optimal solution is determined when a termination condition is met, and determining hyperparameters of the bidirectional long short-term memory network model according to a hyperparameter set corresponding to the optimal solution.

[0188] Specifically, the prediction system repeatedly evaluates the fitness and the position of the updated individual until the termination condition is met, and determines the optimal solution formed by the iteration. The termination condition may include reaching the maximum number of iterations or error convergence. The bidirectional long short-term memory network model is trained according to the hyperparameter set corresponding to the optimal solution to determine the parameters of the bidirectional long short-term memory network model.

[0189] The offline training process of the prediction system can also include training using an optimized bidirectional long short-term memory network model and evaluating performance through a validation set.

[0190] The root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R 2 ) are used as evaluation indicators to conduct a comprehensive performance evaluation of the bidirectional long short-term memory network model, and the accuracy of the bidirectional long short-term memory network model is dynamically determined based on these indicators.

[0191] In this way, the prediction system combines the Pelican optimization algorithm to perform hyperparameter optimization on the number of hidden layer nodes, learning rate, and training times of the bidirectional long short-term memory network unit, and obtains the bidirectional long short-term memory network model with the best performance, and finally realizes the prediction of the purity of the propylene distillation tower product.

[0192] Please refer to Figure 5 , Figure 5 A prediction error diagram of a method for predicting propylene mass fraction provided according to an embodiment of the present invention is shown.

[0193] Figure 5 The figure shows the loss curve of the prediction method of propylene mass fraction provided by the present invention, with the horizontal axis representing the sample data and the vertical axis representing the propylene mass fraction (%). The curve with the symbol △ is the true curve, i.e., the true value corresponding to the data set, and the curve with the symbol * is the prediction curve, which is used to represent the prediction result value using the prediction model. Figure 5 As shown, the error between the true value of the propylene mass fraction corresponding to the sample data and the predicted value determined by the propylene mass fraction prediction method provided by the present invention is small, and the predicted curve and the true curve have a high degree of overlap. The propylene mass fraction prediction method provided by the present invention has high prediction accuracy.

[0194] Compared with traditional modeling methods, the prediction method based on the bidirectional long short-term memory network model and the Pelican optimization algorithm provided by the present invention can cope with more working conditions, has a wider range of applicable scenarios, and has a longer life cycle.

[0195] The following are a number of specific non-limiting preferred embodiments, based on which the bidirectional long short-term memory network model optimized by the Pelican optimization algorithm in the propylene mass fraction prediction system proposed in the present invention is further described.

[0196] In preferred embodiment 1, the prediction system can collect the running propylene distillation tower data through the distributed control system to obtain 600 groups of sample data. Among them, the sample data contains a large amount of variable data. Directly selecting all data as auxiliary variables has a poor effect. The prediction system can preliminarily select relevant auxiliary variables according to mechanism analysis and prior knowledge. In some embodiments, the selected auxiliary variables include propylene feed amount, propane feed amount, methane feed amount, acetylene feed amount, ethane feed amount, ethylene feed amount, propadiene feed amount, propyne feed amount, butadiene feed amount, 1-butene feed amount, 1-3 butadiene feed amount, diacetylene feed amount, normal butane feed amount, isobutane feed amount, feed temperature, reflux ratio, reboiler heat load and distillate flow rate. These data are used as historical data used in the offline training process of the prediction system to train the bidirectional long short-term memory network model.

[0197] In preferred embodiment 1, those skilled in the art may select an LSTM neural network and a Bi-LSTM neural network that has not been optimized by the Pelican optimization algorithm as comparisons to conduct comparative experiments with the bidirectional long short-term memory network model POA-BiLSTM optimized by the Pelican optimization algorithm in the method for predicting propylene mass fraction provided by the present invention.

[0198] Specifically, those skilled in the art can use the same historical data to train the above three models. After training, they can make predictions on the same test data set and obtain the predicted values ​​of each model. The fairness of the evaluation can be ensured by training, testing and verifying on the same data set.

[0199] Please refer to Figure 6 , Figure 6 A comparison diagram of evaluating multiple prediction models provided according to a preferred embodiment 1 of the present invention is shown.

[0200] In the preferred embodiment 1, the comparative experiment can use the root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and determination coefficient (R 2 ) are used to evaluate the above three models.

[0201] like Figure 6 As shown in the figure, the root mean square error (RMSE) of the bidirectional long short-term memory network model POA-BiLSTM optimized by the Pelican optimization algorithm in the prediction method of propylene mass fraction provided by the present invention is 0.045, the mean absolute error (MAE) is 0.036, the mean absolute percentage error (MAPE) is 2.38%, and the determination coefficient (R 2) is 0.94; the root mean square error (RMSE) of the LSTM neural network is 0.057, the mean absolute error (MAE) is 0.044, the mean absolute percentage error (MAPE) is 2.87%, and the coefficient of determination (R 2 ) is 0.89; and the root mean square error (RMSE) of the BiLSTM neural network that has not been optimized by the Pelican optimization algorithm is 0.052, the mean absolute error (MAE) is 0.041, the mean absolute percentage error (MAPE) is 2.78%, and the coefficient of determination (R 2 ) is 0.90.

[0202] according to Figure 6 According to the various evaluation indicators shown, the prediction accuracy of the bidirectional long short-term memory network model POA-BiLSTM optimized by the Pelican optimization algorithm in the prediction method of propylene mass fraction provided by the present invention is significantly better than that of the LSTM neural network and the BiLSTM neural network that has not been optimized by the Pelican optimization algorithm. This is because the bidirectional long short-term memory network can simultaneously capture the forward and backward dependencies in the time series data, ensuring that the model understands the dynamic characteristics of the data more comprehensively. In contrast, although LSTM can also capture time dependencies, it requires more computing resources and time when processing bidirectional information. In addition, models such as LSTM and BiLSTM usually rely on traditional optimization algorithms (such as gradient descent) for parameter optimization, which is easy to fall into local optimality. POA-BiLSTM uses the Pelican optimization algorithm to perform hyperparameter optimization, so that the time dependency capture ability of the bidirectional long short-term memory network model is further enhanced, so that data can be predicted more accurately. In addition, the Pelican optimization algorithm can accelerate the training process of the bidirectional long short-term memory network model, improve the convergence speed and computational efficiency of the bidirectional long short-term memory network model, and enhance the ability of the bidirectional long short-term memory network model to handle complex nonlinear relationships. Therefore, POA-BiLSTM can well complete the prediction task in propylene separation and help the subsequent optimization of the production process.

[0203] Please refer to Figure 7 , Figure 7 A comparative test diagram of multiple prediction models provided according to the second preferred embodiment of the present invention is shown.

[0204] like Figure 7 As shown, in the preferred embodiment 2, those skilled in the art can select the bidirectional long short-term memory network 702 and the traditional long short-term memory network 703 as benchmark methods to conduct comparative experiments with the bidirectional long short-term memory network model 701 optimized by the Pelican optimization algorithm in the prediction method of propylene mass fraction provided by the present invention.

[0205] The comparative experiment uses the box plot of the prediction errors of the three prediction models involved in the preferred embodiment 2 to evaluate the performance of the bidirectional long short-term memory network model 701 optimized by the Pelican optimization algorithm in the prediction method of propylene mass fraction provided by the present invention.

[0206] The obtained bidirectional long short-term memory network model 701 optimized by the Pelican optimization algorithm is compared with the bidirectional long short-term memory network 702 and the traditional long short-term memory network 703. The upper and lower boundaries of the box represent the first quartile (Q1) and the third quartile (Q3) of the error, respectively, and the length of the box (interquartile range, IQR) reflects the degree of dispersion of the model error, and the median line in the box plot represents the middle value of the error, that is, the distribution center of 50% error data. Figure 7 As shown, the median of model 701 is around 0, indicating that the prediction error of model 701 is generally small, that is, model 701 can more accurately predict the key indicators of the propylene distillation tower as a whole. And the error box of model 701 is narrower than that of the bidirectional long short-term memory network 702 and the traditional long short-term memory network 703 (that is, the IQR is smaller), which means that the error distribution of model 701 is more concentrated, indicating that the model prediction results are stable, and most of the prediction errors are within a smaller range. Therefore, it can be explained that the bidirectional long short-term memory network model 701 optimized by the Pelican optimization algorithm proposed in the present invention has stronger prediction ability and faster convergence speed, can better realize the accurate prediction of the key indicators of the propylene distillation tower, and is more suitable for the optimization and control technology of the propylene distillation tower.

[0207] In summary, the prediction method for propylene mass fraction provided by the present invention combines principal component analysis with the bidirectional long short-term memory network model, and optimizes the hyperparameters of the bidirectional long short-term memory network model using the Pelican optimization algorithm, so as to efficiently construct a bidirectional long short-term memory network model and improve the prediction accuracy and robustness of the bidirectional long short-term memory network model.

[0208] Although the above methods are illustrated and described as a series of actions for simplicity of explanation, it should be understood and appreciated that these methods are not limited by the order of the actions, because according to one or more embodiments, some actions may occur in a different order and / or concurrently with other actions from those illustrated and described herein or not illustrated and described herein but understandable to those skilled in the art.

[0209] Those skilled in the art will appreciate that information, signals, and data may be represented using any of a variety of different technologies and techniques. For example, data, instructions, commands, information, signals, bits, symbols, and chips cited throughout the above description may be represented by voltage, current, electromagnetic waves, magnetic fields or magnetic particles, optical fields or optical particles, or any combination thereof.

[0210] Those skilled in the art will further appreciate that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or a combination of the two. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps are generally described above in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. The technician may implement the described functionality in different ways for each specific application, but such implementation decisions should not be interpreted as resulting in a departure from the scope of the present invention.

[0211] The various illustrative logic modules and circuits described in conjunction with the embodiments disclosed herein may be implemented or executed with a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in cooperation with a DSP core, or any other such configuration.

[0212] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. The software module may reside in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor so that the processor can read and write information from / to the storage medium. In an alternative, a storage medium may be integrated into a processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and the storage medium may reside in a user terminal as discrete components.

[0213] In one or more exemplary embodiments, the functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented as a computer program product in software, each function may be stored on or transmitted by a computer-readable medium as one or more instructions or codes. Computer-readable media include both computer storage media and communication media, including any medium that facilitates the transfer of a computer program from one place to another. Storage media may be any available medium that can be accessed by a computer. As an example and not limitation, such a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, disk storage or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of an instruction or data structure and can be accessed by a computer. Any connection is also properly referred to as a computer-readable medium. For example, if the software is transmitted from a website, a server, or other remote source using a coaxial cable, a fiber optic cable, a twisted pair, a digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwaves, the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwaves are included in the definition of the medium. Disk and disc as used herein include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray disc, wherein disk often reproduces data magnetically, while disc reproduces data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

[0214] The previous description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the disclosure is not intended to be limited to the examples and designs described herein, but should be granted the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting propylene mass fraction, characterized in that: Includes steps: Get propylene distillation tower data; constructing a time delay matrix based on the time series characteristics of the propylene distillation tower data, and performing dimension reduction on the propylene distillation tower data using principal component analysis based on the time delay matrix to extract characteristic data; as well as The characteristic data is input into a bidirectional long short-term memory network model to predict the propylene mass fraction, and the hyperparameters of the bidirectional long short-term memory network model are determined according to the Pelican optimization algorithm.

2. The method for predicting the mass fraction of propylene according to claim 1, characterized in that: The bidirectional long short-term memory network model includes a forward long short-term memory network calculation process and a reverse long short-term memory network calculation process.

3. The method for predicting the mass fraction of propylene according to claim 2, characterized in that: The forward long short-term memory network calculation process and the reverse long short-term memory network calculation process include the calculation of a forget gate, an input gate, candidate memory units, memory unit updates, an output gate, and hidden state updates.

4. The method for predicting the mass fraction of propylene according to claim 3, wherein: The calculation process of the forget gate of the forward long short-term memory network calculation process is: Among them, σ is the sigmoid activation function, W f (f) is the weight matrix of the forget gate in the forward LSTM network calculation process, x t Represents the input at the current moment, represents the hidden state of the previous moment, b f is the bias matrix; The calculation process of the input gate of the forward long short-term memory network calculation process is: Among them, W i (f) is the weight matrix of the input gate of the forward LSTM network calculation process, b i is the bias matrix; The calculation process of the candidate memory units in the forward long short-term memory network calculation process is as follows: in, is the weight matrix of the candidate hidden state in the forward LSTM network calculation process, b C is the bias matrix; The calculation process of the memory unit update in the forward long short-term memory network calculation process is: ⊙ is the element-wise product, is the output of the input gate; The calculation process of the output gate of the forward long short-term memory network calculation process is: in, is the weight matrix of the output gate of the forward LSTM network calculation process, b o is the bias matrix; The calculation process of the hidden state update of the forward long short-term memory network calculation process is: Among them, C t is the update result of the memory unit at the current moment; The calculation process of the forget gate of the reverse long short-term memory network calculation process is: Among them, σ is the sigmoid activation function, is the weight matrix of the forget gate in the reverse long short-term memory network calculation process, x t Represents the input at the current moment, represents the hidden state at the next moment, b f is the bias matrix; The calculation process of the input gate of the reverse long short-term memory network calculation process is: Among them, W i (b) is the weight matrix of the input gate of the reverse long short-term memory network calculation process, b i is the bias matrix; The calculation process of the candidate memory units in the reverse long short-term memory network calculation process is as follows: in, is the weight matrix of the candidate hidden state in the reverse long short-term memory network calculation process, b C is the bias matrix; The calculation process of memory unit update in the reverse long short-term memory network calculation process is: ⊙ is the element-wise product, is the output of the input gate; The calculation process of the output gate of the reverse long short-term memory network calculation process is: in, is the weight matrix of the output gate of the reverse long short-term memory network calculation process, b o is the bias matrix; The calculation process of the hidden state update of the reverse long short-term memory network calculation process is: in, is the update result of the memory unit at the current moment.

5. The method for predicting the mass fraction of propylene according to claim 2, wherein: The output of the bidirectional long short-term memory network model is the concatenation of the hidden state of the forward long short-term memory network calculation process and the hidden state of the reverse long short-term memory network calculation process.

6. The method for predicting the mass fraction of propylene according to claim 1, wherein: The steps of determining the hyperparameters of the bidirectional long short-term memory network model during training according to the Pelican optimization algorithm include: Step S1: Initialize the initial position, speed and fitness of each individual of the Pelican optimization algorithm, where the individual is a set of hyperparameters of the bidirectional long short-term memory network model; Step S2: using each individual to train the bidirectional long short-term memory network model, and determining the fitness value of each individual by verifying the error of the bidirectional long short-term memory network model after training; Step S3: determining the historical optimal position and the global optimal position of each individual according to the fitness value; Step S4: based on the historical optimal position of each individual and the global optimal position, updating the position and speed of each individual; and Step S5: Repeat steps S2 to S4 until an optimal solution is determined when a termination condition is met, and determine the hyperparameters of the bidirectional long short-term memory network model according to a hyperparameter set corresponding to the optimal solution.

7. The method for predicting the mass fraction of propylene according to claim 6, wherein: In step S4, the formula for updating the position and speed of each individual is: X i (t+1)=X i (t)+V i (t+1), Among them, ω represents the inertia weight, c1, c2, c3 are acceleration constants used to control the acceleration of the individual to the historical optimal position and the global optimal position, r1, r2, r3 are random numbers between 0 and 1, represents the historical optimal position of the individual i, represents the global optimal position, l i is used to help the individual i to guide to a target solution position during the search process, X i (t) is the position of the individual i at the tth iteration, V i (t) is the velocity of the individual i at the tth iteration.

8. The method for predicting the mass fraction of propylene according to claim 1, wherein: The steps of constructing a time delay matrix based on the time series characteristics of the propylene distillation tower data, and performing dimension reduction on the propylene distillation tower data by using principal component analysis based on the time delay matrix to extract characteristic data include: Constructing a time delay matrix based on the time series characteristics of the propylene distillation tower data, and calculating a covariance matrix of the time delay matrix; Perform eigenvalue decomposition based on the covariance matrix, and select eigenvectors corresponding to the first K eigenvalues ​​as principal components; Using the principal component to reduce the dimension of the propylene distillation tower data to obtain reduced-dimensional data; and Cluster analysis is performed on the dimension-reduced data to extract the feature data.

9. The method for predicting the mass fraction of propylene according to claim 1, wherein: The step of obtaining propylene distillation tower data comprises: The sample data is preprocessed to determine the propylene distillation tower data, wherein the preprocessing includes processing of abnormal values ​​and processing of missing values.

10. The method for predicting the mass fraction of propylene according to claim 9, characterized in that: The processing of missing values ​​includes filling the missing values ​​of the sample data using a multiple interpolation algorithm.

11. The method for predicting the mass fraction of propylene according to claim 1, characterized in that: The propylene distillation tower data include one or more of propylene feed amount, propane feed amount, methane feed amount, acetylene feed amount, ethane feed amount, ethylene feed amount, propadiene feed amount, propyne feed amount, butadiene feed amount, 1-butene feed amount, 1-3 butadiene feed amount, diacetylene feed amount, n-butane feed amount, isobutane feed amount, feed temperature, reflux ratio, reboiler heat duty and distillate flow rate.

12. A prediction system for propylene mass fraction, characterized in that: include: a memory having computer instructions stored thereon; and A processor is connected to the memory and configured to execute computer instructions stored in the memory to implement the method for predicting the mass fraction of propylene according to any one of claims 1 to 11.

13. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by a processor, the method for predicting the propylene mass fraction according to any one of claims 1 to 11 is implemented.

Images