A method, system and storage medium for predicting ethylene concentration

Abstract

The application provides an ethylene concentration prediction method, an ethylene concentration prediction system and a computer readable storage medium. The ethylene concentration prediction method comprises the steps of: obtaining ethylene rectification column data, the ethylene rectification column data comprising variables related to ethylene rectification column products; performing dimensionality reduction extraction and noise reduction processing on the ethylene rectification column data by using principal component analysis and a denoising autoencoder to determine characteristic data; and inputting the characteristic data into a hybrid neural network model to predict the ethylene concentration, wherein the hybrid neural network model is constructed based on a bidirectional gated recurrent unit, and the preferred hyperparameters of the hybrid neural network model are determined according to a whale optimization algorithm. The ethylene concentration prediction method provided by the application can accurately predict the ethylene concentration in real time, and solves the problems of low prediction accuracy and poor real-time performance caused by changes in process state and the like in the ethylene separation process.

Description

Technical Field

[0001] This invention relates to the field of ethylene production, and more particularly to a method for predicting ethylene concentration, a system for predicting ethylene concentration, and a computer-readable storage medium. Background Technology

[0002] Ethylene is a crucial raw material in the organic chemical industry. The feedstock for ethylene plants includes refinery gas from primary refinery processing and hydrotreated tail oil from secondary processing, with products including methane and ethylene. Ethylene can be polymerized to produce polyethylene and ethylene copolymers, oxidized to produce ethylene oxide and acetaldehyde, alkylated to produce ethylbenzene and alkylaluminum, and halogenated to produce dichloroethane. Therefore, ethylene is often referred to as the "mother of chemicals."

[0003] The key to optimizing the ethylene separation process is real-time monitoring of the ethylene concentration at the top of the ethylene distillation column. By monitoring the ethylene concentration in real time, the parameters of the ethylene separation process can be controlled, thereby increasing product purity.

[0004] Traditional methods for monitoring ethylene concentration rely on instruments such as online chromatographs to measure ethylene concentration. While instrument-based methods are accurate, they suffer from significant time lags, thus failing to meet the requirements for real-time monitoring of ethylene concentration.

[0005] Existing soft sensor techniques estimate the ethylene concentration at the top of ethylene distillation columns by establishing mathematical models based on easily measurable process variables and target variables. This allows for the prediction of ethylene concentrations that are difficult or impossible to measure directly during ethylene production, thereby optimizing production control. Neural networks, a type of soft sensor technique, possess powerful nonlinear mapping capabilities and are widely used in chemical process modeling. However, the complexity and variability of ethylene production processes pose challenges to the adaptability and predictive accuracy of these models. In particular, existing neural network models cannot accurately predict changes in production conditions and require retraining to adapt to the new conditions, which is both time-consuming and costly.

[0006] In order to overcome the above-mentioned defects of the existing technology, there is an urgent need in the field for an ethylene concentration prediction technology that can predict ethylene concentration in real time and accurately, and solve the problems of low prediction accuracy and poor real-time performance caused by changes in process status during ethylene separation. Summary of the Invention

[0007] The following provides a brief overview of one or more aspects to offer a basic understanding of them. This overview is not an exhaustive summary of all conceived aspects, nor is it intended to identify key or decisive elements of all aspects, nor to define the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form to prepare for the more detailed descriptions that follow.

[0008] To overcome the aforementioned deficiencies in the existing technology, the present invention provides a method for predicting ethylene concentration, a system for predicting ethylene concentration, and a computer-readable storage medium, which can predict ethylene concentration accurately in real time and solve the problems of low prediction accuracy and poor real-time performance caused by changes in process status during ethylene separation.

[0009] Specifically, the method for predicting ethylene concentration according to the first aspect of the present invention includes the steps of: acquiring ethylene distillation column data, the ethylene distillation column data including variables related to the products of the ethylene distillation column; performing dimensionality reduction extraction and denoising processing on the ethylene distillation column data using principal component analysis and a noise reduction autoencoder to determine feature data; and inputting the feature data into a hybrid neural network model to predict ethylene concentration, wherein the hybrid neural network model is constructed based on a bidirectional gated recurrent unit, and the preferred hyperparameters of the hybrid neural network model are determined according to the whale optimization algorithm.

[0010] Preferably, in one embodiment of the present invention, the noise reduction autoencoder includes an encoder and a decoder, the encoder and the decoder include two fully connected neural networks, and the encoder and the decoder have a symmetrical structure.

[0011] Preferably, in one embodiment of the present invention, the encoder construction process is as follows:

[0012] X encoder = f(f(W1X+B1)W2+B2),

[0013] Where f is the activation function, W1 is the weight of the first fully connected layer of the encoder's neural network, W2 is the weight of the second fully connected layer of the encoder's neural network, B1 is the bias of the first fully connected layer of the encoder's neural network, B2 is the bias of the second fully connected layer of the encoder's neural network, and X... encoder The first feature is obtained by the encoder encoding the dimension-reduced ethylene distillation column data;

[0014] The construction process of the decoder is as follows:

[0015] X decoder =f(f(W2X) encoder +B2)W1+B1),

[0016] Among them, X decoder The feature data output by the noise reduction autoencoder.

[0017] Preferably, in one embodiment of the present invention, the step of determining the preferred hyperparameters of the hybrid neural network model according to the whale optimization algorithm includes: in response to the prediction accuracy of the hybrid neural network model not meeting the requirements, determining the hyperparameters of the hybrid neural network model according to the whale optimization algorithm, wherein the prediction accuracy of the hybrid neural network model is determined by an error discriminant function; training the hybrid neural network model according to the hyperparameters to update the parameters of the hybrid neural network model, and determining the prediction accuracy of the hybrid neural network model after updating the parameters; and repeating the above steps until the prediction accuracy of the hybrid neural network model meets the requirements, and determining the hyperparameters for training the hybrid neural network model as the preferred hyperparameters.

[0018] Preferably, in one embodiment of the present invention, the step of determining the hyperparameters of the hybrid neural network model according to the whale optimization algorithm in response to the prediction accuracy of the hybrid neural network model not meeting the requirements includes: S1: In response to the accuracy of the hybrid neural network model not meeting the requirements, initializing the number of whale individuals, whale positions, and algorithm parameters, wherein the whale positions represent the hyperparameters of the hybrid neural network model; S2: Calculating the fitness of the whale individuals based on their whale positions to determine the initial optimal whale individual position as the preferred solution; S3: Updating the algorithm parameters and updating the whale positions of the whale individuals by adopting bubble net predation, encirclement and predation, or search and predation methods according to the algorithm parameters; S4: Calculating the fitness of the whale individuals based on the updated whale positions to determine the optimal whale individual position; S5: In response to the optimal whale individual position being better than the preferred solution, using the optimal whale individual position as the preferred solution; and S6: Repeating steps S3 to S5 until the termination condition is met, and outputting the whale position of the preferred solution to determine the hyperparameters of the hybrid neural network model.

[0019] Preferably, in one embodiment of the present invention, step S3 includes: updating the whale position of the individual whale by using a bubble net for predation in response to a probability p being greater than or equal to 0.5; updating the whale position of the individual whale by using a predation trap in response to a probability p being less than 0.5 and an absolute value of the coefficient vector A being less than 1; and updating the whale position of the individual whale by using a search and predation trap in response to a probability p being less than 0.5 and an absolute value of the coefficient vector A being greater than or equal to 1.

[0020] Preferably, in one embodiment of the present invention, the error discrimination function includes root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination.

[0021] Preferably, in one embodiment of the present invention, the variables related to the products of the ethylene distillation column include ethylene feed rate, ethane feed rate, feed temperature, feed pressure, reflux ratio, reboiler heat load, and column top condenser heat load.

[0022] Furthermore, the ethylene concentration prediction system 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 configured to execute the computer instructions stored in the memory to implement the ethylene concentration prediction method provided in any of the above embodiments.

[0023] Furthermore, the computer-readable storage medium provided according to the third aspect of the present invention stores computer instructions. When the computer instructions are executed by a processor, the ethylene concentration prediction method provided in any of the above embodiments is implemented. Attached Figure Description

[0024] The above-described features and advantages of the present invention will be better understood after reading the following detailed description of embodiments of the present disclosure in conjunction with the accompanying drawings. In the drawings, components are not necessarily drawn to scale, and components having similar related properties or features may have the same or similar reference numerals.

[0025] Figure 1 A schematic diagram of an ethylene concentration prediction system provided according to some embodiments of the present invention is shown;

[0026] Figure 2 A flowchart of a method for predicting ethylene concentration according to some embodiments of the present invention is shown;

[0027] Figure 3 A schematic diagram of a hybrid neural network model provided according to some embodiments of the present invention is shown;

[0028] Figure 4 A flowchart illustrating the updating of parameters of a hybrid neural network model according to some embodiments of the present invention is shown;

[0029] Figure 5 A flowchart of a whale optimization algorithm provided according to some embodiments of the present invention is shown;

[0030] Figure 6 A comparison graph showing the prediction results of a hybrid neural network model provided according to an embodiment of the present invention with the prediction results of other models is shown; and

[0031] Figure 7 A comparison graph is shown of the prediction results of a hybrid neural network model provided according to an embodiment of the present invention with the prediction results of other models.

[0032] Figure label:

[0033] 100: Ethylene concentration prediction system;

[0034] 110: Memory;

[0035] 111: Computer-readable storage medium;

[0036] 120: Processor;

[0037] 200: Methods for predicting ethylene concentration;

[0038] 300: Hybrid Neural Network Model;

[0039] 310: Forward-gated loop unit;

[0040] 320: Backward gated loop unit;

[0041] S210~S230: Steps; and

[0042] S410~S440: Steps. Detailed Implementation

[0043] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should be noted that the aspects described below with reference to the accompanying drawings and specific embodiments are merely exemplary and should not be construed as limiting the scope of protection of the present invention in any way.

[0044] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0045] Furthermore, the terms "upper," "lower," "left," "right," "top," "bottom," "horizontal," and "vertical" used in the following description should be understood as the orientations shown in the relevant paragraphs and accompanying drawings. These relative terms are for illustrative purposes only and do not imply that the described apparatus must be manufactured or operated in a specific orientation, and therefore should not be construed as limiting the invention.

[0046] It is understood that although terms such as "first," "second," and "third" 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 components, regions, layers, and / or parts discussed below may be referred to as second components, regions, layers, and / or parts without departing from some embodiments of the present invention.

[0047] As mentioned above, existing soft sensor techniques estimate the ethylene concentration at the top of the ethylene distillation column by establishing mathematical models based on easily measurable process variables and target variables. This allows for the prediction of ethylene concentration that is difficult or impossible to measure directly during ethylene production, thereby optimizing production control. Neural networks, a type of soft sensor technique, possess powerful nonlinear mapping capabilities and are widely used in chemical process modeling. However, the complexity and variability of the ethylene production process pose challenges to the adaptability and prediction accuracy of these models. In particular, when production conditions change, existing neural network models cannot achieve accurate predictions and require retraining to adapt to the new operating conditions, which is not only time-consuming but also costly.

[0048] To overcome the aforementioned deficiencies in the existing technology, the present invention provides a method for predicting ethylene concentration, a system for predicting ethylene concentration, and a computer-readable storage medium, which can predict ethylene concentration accurately in real time and solve the problems of low prediction accuracy and poor real-time performance caused by changes in process status during ethylene separation.

[0049] In some non-limiting embodiments, the ethylene concentration prediction method provided in the first aspect of the present invention can be implemented via the ethylene concentration prediction system provided in the second aspect of the present invention.

[0050] Please refer to Figure 1 , Figure 1 A schematic diagram of an ethylene concentration prediction system provided according to some embodiments of the present invention is shown.

[0051] like Figure 1 As shown, the ethylene concentration prediction system 100 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 described 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 ethylene concentration prediction method provided in the first aspect of the present invention.

[0052] The working principle of the ethylene concentration prediction system will first be described using examples of some ethylene concentration prediction methods. Those skilled in the art will understand that these examples of ethylene concentration prediction methods are merely non-limiting embodiments provided by this invention, intended to clearly demonstrate the main concepts of the invention and provide specific solutions convenient for public implementation, rather than limiting all functions or operating methods of the ethylene concentration prediction system. Similarly, the ethylene concentration prediction system is also only a non-limiting embodiment provided by this invention, and does not limit the executing entities or execution order of the steps in these ethylene concentration prediction methods.

[0053] Please refer to Figure 2 , Figure 2 A flowchart of a method for predicting ethylene concentration according to some embodiments of the present invention is shown.

[0054] like Figure 2 As shown, the ethylene concentration prediction method 200 may include step S210: acquiring ethylene distillation column data, which includes variables related to the ethylene distillation column products.

[0055] In some embodiments, the method for predicting ethylene concentration provided in the first aspect of the present invention can be implemented in two parts: an offline training phase and an online analysis phase.

[0056] During the online analysis phase, the ethylene concentration prediction system can select variables related to the ethylene distillation column as inputs for predicting ethylene concentration. During the offline training phase, the ethylene distillation column data can also include the actual ethylene concentration. The ethylene concentration prediction system can acquire the actual ethylene concentration as output while acquiring the ethylene distillation column data to achieve the offline training phase of the ethylene concentration prediction method. In some embodiments, the actual ethylene concentration can be the concentration of impurities at the top of the column measured by an analyzer. Variables related to the ethylene distillation column products can include ethylene feed rate, ethane feed rate, feed temperature, feed pressure, reflux ratio, reboiler heat load, and top condenser heat load.

[0057] During the offline training phase, ethylene distillation column data, including variables related to ethylene distillation column products and actual ethylene concentrations, can be divided into training, testing, and validation sets. Furthermore, the ethylene concentration prediction system can perform standardized preprocessing on the ethylene distillation column data.

[0058] like Figure 2 As shown, the ethylene concentration prediction method 200 may include step S220: using principal component analysis and a noise reduction autoencoder to perform dimensionality reduction extraction and noise reduction processing on the ethylene distillation column data to determine feature data.

[0059] First, the ethylene concentration prediction system can perform principal component analysis on the original ethylene distillation column data to reduce the dimensionality of the dataset X. Specifically, the system can calculate the feature covariance matrix based on the ethylene distillation column data to determine the linear relationship between the features of the data. Then, the system can calculate the eigenvalues ​​and corresponding eigenvectors of the covariance matrix. Eigenvectors with a cumulative contribution exceeding a preset threshold are then selected to obtain the dimensionality-reduced dataset X. In some embodiments, this preset threshold can be 80%.

[0060] Subsequently, the ethylene concentration prediction system can denoise the dimensionality-reduced dataset X based on the constructed and trained denoising autoencoder.

[0061] A noise reduction autoencoder may include an encoder and a decoder, which may include two fully connected neural networks, and the encoder and decoder may be symmetrical structures.

[0062] Specifically, in a non-limiting embodiment, the construction of the noise-reducing autoencoder includes the construction of an encoder and a decoder. The encoder construction process is as follows:

[0063] X encoder = f(f(W1X+B1)W2+B2),

[0064] Where f is the activation function, W1 is the weight of the first fully connected layer of the encoder, W2 is the weight of the second fully connected layer of the encoder, B1 is the bias of the first fully connected layer of the encoder, B2 is the bias of the second fully connected layer of the encoder, and X... encoder The first feature is obtained by the encoder from the dimension-reduced ethylene distillation column data, that is, the first feature obtained by the encoder from the dataset X.

[0065] The decoder construction process is as follows:

[0066] X decoder =f(f(W2X) encoder +B2)W1+B1),

[0067] Among them, X decoder The output feature data is the noise reduction autoencoder. In this non-limiting embodiment, the decoder and encoder have a symmetrical structure and include two fully connected neural networks. Preferably, the activation function f can be the ReLU (Rectified Linear Unit) activation function.

[0068] Next, the trained denoising autoencoder is used to denoise the dimensionality-reduced ethylene distillation column data. In some embodiments, the ethylene concentration prediction system can use mean squared error as an evaluation metric for assessing the denoising effect. The feature data obtained after dimensionality reduction and denoising of the ethylene distillation column data can be used as input to the hybrid neural network model provided by the ethylene concentration prediction method.

[0069] Thus, by performing dimensionality reduction and noise reduction on the input ethylene distillation tower data, the ethylene concentration prediction system can not only significantly reduce the computational complexity of the subsequent hybrid neural network model, but also significantly enhance the generalization ability of the hybrid neural network model to new data, thereby improving the effectiveness and reliability of the hybrid neural network model in practical industrial applications.

[0070] Please continue to refer to this. Figure 2 The ethylene concentration prediction method 200 may include step S230: inputting feature data into a hybrid neural network model to predict ethylene concentration, wherein the hybrid neural network model is constructed based on bidirectional gated recurrent units, and the preferred hyperparameters of the hybrid neural network model are determined according to the whale optimization algorithm.

[0071] LSTM (Long Short-Term Memory) networks can reduce gradient vanishing by introducing special gate structures, allowing errors to propagate over long distances and thus possessing long-term memory capabilities. GRU (Gated Recurrent Unit) is an improved LSTM algorithm. GRU combines the forget gate and input gate into a single update gate and merges the data state and hidden state. Compared to LSTM, GRU has a simpler structure and faster training speed.

[0072] The ethylene concentration prediction system can be based on a hybrid neural network model constructed from bidirectional gated recurrent units. In some embodiments, the hybrid neural network model can be a WOA-BiGRU (Whale Optimization Algorithm-Bidirectional Gated Recurrent Unit) model.

[0073] Please refer to Figure 3 , Figure 3 A schematic diagram of a hybrid neural network model provided according to some embodiments of the present invention is shown.

[0074] like Figure 3As shown, the hybrid neural network model 300 can be composed of bidirectional gated recurrent units, including a forward gated recurrent unit 310 and a backward gated recurrent unit 320. Each of the forward and backward gated recurrent units can consist of multiple gated recurrent units (GRUs). The input to the hybrid neural network model 300 can include inputs at different time steps, x t x is the input for the current time step. t-1 As the input of the previous time step, x t+1 This serves as the input for the next time step. The output of the hybrid neural network model 300 can be the hidden state at different time steps, h. t h represents the hidden state at the current time step. t-1 h is the hidden state of the previous time step. t+1 This is the hidden state for the next time step.

[0075] The Gated Recurrent Unit (GRU) combines the forget gate and input gate of the LSTM into an update gate. The update gate can be used to control which information from the current time step's state information needs to flow into the candidate state of the next time step.

[0076] like Figure 3 As shown, taking the forward-gated loop unit 310 as an example, the inputs to the GRU of the forward-gated loop unit 310 include the hidden state h output by the GRU in the previous time step. t-1 and the input x at the current time step t The specific formula for updating the door is as follows:

[0077] z i = sigmod(W z [h t-1 x t ]+b z ),

[0078] Among them, z t It updates the output of the gate, h. t-1 It is the hidden state of the previous time step, x t It is the input at the current time step, W z It is the weight matrix of the updated gate, b z It is the bias vector for updating the gate.

[0079] GRU candidate states As shown below:

[0080]

[0081] Among them, W z W h b is the weight parameter of the gated recurrent unit. z bh For the bias of the gated loop unit, r t It is the output of the reset door.

[0082] The reset gate determines the output h of the hidden layer at the previous time step. t-1 That is, the hidden state h of the previous time step. t-1 For candidate states The effect. The specific formula is as follows:

[0083] r t = sigmod(W r [h t-1 x t ]+b r ),

[0084] Among them, W r To reset the weight parameters of the gate; b r To reset the door offset.

[0085] Therefore, the output of the GRU of the forward-gated loop unit 310 can be determined by the output of the hidden layer in the previous time step, the output of the update gate, and the candidate state in the current time step. The output of the GRU of the forward-gated loop unit 310 is the forward hidden state.

[0086]

[0087] Accordingly, in the backward-gated loop unit 320, the input to the GRU of the backward-gated loop unit 320 includes the hidden state h of the GRU output in the next time step. t+1 and the input x at the current time step t Subsequently, based on the calculations of the update gate, candidate state, and reset gate, the ethylene concentration prediction system can determine the output of the backward-gated loop unit 320, i.e., the reverse hidden state, based on the output of the hidden layer in the next time step, the output of the update gate, and the candidate state in the current time step.

[0088] Please continue to refer to this. Figure 3 In the bidirectional gated recurrent unit of the hybrid neural network model 300, feature data can be fed into the forward gated recurrent unit 310 and the backward gated recurrent unit 320 respectively, and the forward hidden state at the current time step t... and reverse hidden state The combination of these is the output of the hybrid neural network model 300, i.e.

[0089] In some embodiments, the preferred hyperparameters of the constructed hybrid neural network model can be determined using the whale optimization algorithm. Through the whale optimization algorithm, the ethylene concentration prediction system can optimize and improve the performance of the bidirectional gated recurrent units constituting the hybrid neural network model, obtaining preferred hyperparameter weights for training the hybrid neural network model, and then determining the parameters of the hybrid neural network model through training and validation. Hyperparameters may include the number of hidden layer nodes, the number of training rounds, and the learning rate.

[0090] Please refer to Figure 4 , Figure 4 A flowchart illustrating the updating of parameters for a hybrid neural network model according to some embodiments of the present invention is shown.

[0091] like Figure 4 As shown, the method for predicting ethylene concentration may further include step S410: determining the preferred hyperparameters of a hybrid neural network model in response to the prediction accuracy of the hybrid neural network model not meeting the requirements, determining the hyperparameters of the hybrid neural network model according to the whale optimization algorithm, wherein the prediction accuracy of the hybrid neural network model is determined by the error discriminant function. Then, step S420 is performed: training the hybrid neural network model based on the hyperparameters to update the parameters of the hybrid neural network model, and determining the prediction accuracy of the hybrid neural network model after updating the parameters. Steps S410 and S420 are repeated until the prediction accuracy of the hybrid neural network model meets the requirements, and the hyperparameters used to train the hybrid neural network model with the required prediction accuracy are determined as the preferred hyperparameters.

[0092] In some embodiments, during the offline training phase, the ethylene concentration prediction system can train and validate the constructed hybrid neural network model. If the prediction accuracy of the established hybrid neural network model does not meet the requirements, the hyperparameters of the hybrid neural network model can be determined according to the whale optimization algorithm, and the hybrid neural network model can be retrained and validated to update the parameters until the prediction accuracy meets the requirements. During the online analysis phase, the measured actual ethylene concentration value can be verified against the predicted ethylene concentration. If the prediction accuracy of the established hybrid neural network model does not meet the requirements, the hyperparameters of the hybrid neural network model can be determined according to the whale optimization algorithm, and the hybrid neural network model can be retrained and validated to update the parameters. Until the hybrid neural network model meets the prediction accuracy requirements, the ethylene concentration prediction system can determine the hyperparameters used to train the hybrid neural network model as the preferred hyperparameters, and then determine the updated parameters based on the hybrid neural network model, thereby realizing the dynamic adjustment of the parameters of the hybrid neural network model.

[0093] Preferably, the ethylene concentration prediction system can update the connection weights of the hybrid neural network model through backpropagation during the training process, and measure the difference between the predicted and actual ethylene concentrations using an appropriate loss function, until the maximum number of training rounds or the error requirement is met. Then, historical or simulation data are used to evaluate the performance of the optimized hybrid neural network model. In some embodiments, simulation data can be obtained by simulating the operation of an ethylene distillation column using Aspenplus software.

[0094] Please refer to Figure 5 , Figure 5 A flowchart of a whale optimization algorithm provided according to some embodiments of the present invention is shown.

[0095] like Figure 5 As shown, in response to the fact that the prediction accuracy of the hybrid neural network model does not meet the requirements, the steps for determining the hyperparameters of the hybrid neural network model based on the whale optimization algorithm may include step S1: In response to the fact that the accuracy of the hybrid neural network model does not meet the requirements, the number of individual whales, the whale positions, and the algorithm parameters are initialized, where the whale positions represent the hyperparameters of the hybrid neural network model. The algorithm parameters of the whale optimization algorithm may include the number of individual whales, the maximum number of iterations, parameter a, coefficient vector A, coefficient vector C, probability p, and random number l.

[0096] Then, the ethylene concentration prediction system can perform step S2: calculate the fitness of the whale individual based on its position to determine the initial optimal position of the whale individual as the preferred solution.

[0097] The ethylene concentration prediction system can calculate the fitness of an individual whale based on its location, using a defined objective function.

[0098] In some embodiments, the objective function can be:

[0099]

[0100] Where N is the sample size, i.e., the number of individual whales. Let be the predicted output of the hybrid neural network model, yi be the actual output value of the i-th sample, W be the set of all weights in the hybrid neural network model, b be the set of all biases in the hybrid neural network model, and η be the learning rate of the hybrid neural network model. min η is the lower bound of the learning rate. max Here, E represents the upper limit of the learning rate, and E is the number of training epochs for the hybrid neural network model. min E is the lower limit of the number of training epochs. max This represents the upper limit for the number of training rounds.

[0101] Next, step S3 is executed: update the algorithm parameters and, based on the algorithm parameters, update the whale's position by using bubble net hunting, encircling prey, or searching for prey.

[0102] The ethylene concentration prediction system can update the parameters 'a', coefficient vector A, coefficient vector C, probability 'p', and random number 'l' of the whale optimization algorithm. Based on the updated probability 'p' and coefficient vector A, it updates the whale's position accordingly. During this process, the current whale determines whether to swim towards prey in a spiral or a closed-circuit approach based on the probability 'p' between (0, 1). When probability 'p' is greater than or equal to 0.5, the whale can swim towards prey in a spiral, i.e., use a bubble net approach to update its position; otherwise, when probability 'p' is less than 0.5, it swims towards prey in a closed-circuit approach, i.e., the method of encircling or searching for prey is determined by the absolute value of coefficient vector A to update the whale's position.

[0103] Depending on the value of probability p, the whale optimization algorithm can switch between spiral and circular motion to update the position of an individual whale. When probability p is greater than or equal to 0.5, bubble net hunting is used to update the position of the individual whale. Bubble net hunting is a method of updating the position of an individual whale based on the hunting behavior of humpback whales; in bubble net hunting, the whale swims towards its prey using spiral motion. After updating using spiral motion, the position X(n+1) of the individual whale is as follows:

[0104] D k =|X * (n)-X(n)|,

[0105] X(n+1)=X * (n)+D k ·e bl ·cos 2πl,

[0106] Among them, D k This represents the current whale position X(n) and the current optimal whale position X. * The distance between (n), b is the logarithmic spiral shape constant, and l is a random number between [-1, 1].

[0107] In response to a probability p less than 0.5 and an absolute value of coefficient vector A less than 1, the ethylene concentration prediction system can update the whale position of an individual whale by hunting its prey.

[0108] The whale optimization algorithm searches within the global solution space, requiring the prey range to be determined first for encirclement. However, the optimal position within the search space is unknown. Therefore, the whale optimization algorithm can assume that the current optimal whale's position is the target prey or close to the optimal solution. After defining the target prey's position, other whales can attempt to encircle it, i.e., hunt the prey. The position of the encircling prey to update the whale's position X(n+1) can be as follows:

[0109] D = |C × X * (n)-X(n)|,

[0110] X(n+1)=X * (n)-A·D,

[0111] Where n represents the current iteration number, A and C are the coefficient vectors of the whale optimization algorithm, and D represents the distance between the current whale position and the current optimal whale position.

[0112] In response to a probability p less than 0.5 and an absolute value of coefficient vector A greater than or equal to 1, the ethylene concentration prediction system can update the whale position of individual whales by searching and preying on them.

[0113] To ensure sufficient searching, individual whales can randomly choose the location of another whale as a reference to update their own position, rather than using the current best whale's position as a reference. The whale position X(n+1) updated using the search-and-prey method is as follows:

[0114] X(n+1)=X rand (n)-A·D,

[0115] D = |C × X * (n)-X(n)|,

[0116] Among them, X rand (n) represents the location of a randomly selected individual whale.

[0117] Please continue to refer to this. Figure 5 The ethylene concentration prediction system can perform step S4: calculate the fitness of a whale individual based on the updated whale individual's whale position to determine the optimal whale individual's position.

[0118] Then, proceed to step S5: in response to the fact that the position of the optimal whale individual is better than the preferred solution, the position of the optimal whale individual is taken as the preferred solution.

[0119] After the position update, the ethylene concentration prediction system can recalculate the fitness of each individual whale. Based on the fitness, it determines the position of the optimal whale after the update iteration. Then, it compares the position of the optimal whale after the update iteration with the position of the optimal whale from the previous iteration (i.e., the current preferred solution). If the position of the optimal whale after the update iteration is better than the preferred solution, then the position of the optimal whale after the update iteration replaces the current preferred solution as the new preferred solution.

[0120] Then, the ethylene concentration prediction system can execute step S6: repeat steps S3 to S5 until the termination condition is met, and output the whale position of the optimal solution to determine the hyperparameters of the hybrid neural network model.

[0121] The ethylene concentration prediction system can determine whether the current iteration count has reached the maximum number of iterations as a termination condition. In some embodiments, the termination condition can be that the fitness of the optimal whale individual no longer decreases with increasing iteration count. If the termination condition is met, the ethylene concentration prediction system can output the preferred solution as the optimal solution using the whale optimization algorithm; if the termination condition is not met, the system proceeds to the next iteration.

[0122] Thus, by employing the whale optimization algorithm as a hyperparameter optimization strategy, the performance of the hybrid neural network model was further optimized by precisely adjusting key non-learning parameters, such as the learning rate and the number of nodes in the hidden layers. This optimization not only enhanced the generalization ability of the hybrid neural network model and avoided overfitting, but also accelerated its convergence speed and improved training efficiency.

[0123] like Figure 4 As shown, to comprehensively evaluate the predictive performance of the hybrid neural network model, the ethylene concentration prediction system uses an error discriminant function to determine the prediction accuracy of the hybrid neural network model. By minimizing the error discriminant function, the ethylene concentration prediction system can dynamically adjust the hyperparameters of the hybrid neural network model using the whale optimization algorithm to train the hybrid neural network model and update the parameters, thereby achieving accurate prediction of ethylene concentration.

[0124] The error discrimination function employs various evaluation metrics, including but not limited to root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R²). 2Through multi-dimensional evaluation, the ethylene concentration prediction system can integrate the advantages of different measurement criteria to ensure that the established hybrid neural network model achieves optimal prediction accuracy and robustness.

[0125] In the specific calculation of the error discrimination function, the root mean square error is the square root of the ratio of the square of the predicted ethylene concentration value and the square of the actual ethylene concentration value to the number of observations (i.e., the number of samples) N, which is the predicted value. and actual value t i The square root of the ratio of the square of the deviation to the number of observations N:

[0126]

[0127] When the predicted value Close to the actual value t i When the predicted value is low, the RMSE value will decrease; when the predicted value is low... Far from the actual value t i When the RMSE value increases, the RMSE value will increase. Therefore, in predicting ethylene concentration, the smaller the RMSE, the better.

[0128] The mean absolute error is the average of the absolute errors, which is the predicted value. and actual value t i Average of the errors between:

[0129]

[0130] When the predicted value Close to the actual value t i When the forecast value is high, the MAE value will decrease; as the forecast value increases... Compared with the actual value t i As the error between values ​​increases, the MAE value will increase linearly. Therefore, in predicting ethylene concentration, a smaller MAE is better.

[0131] Mean absolute percentage error is the predicted value Compared with the actual value t i The average percentage of the relative error between them:

[0132]

[0133] Mean absolute percentage error uses percentages to represent predicted values. and actual value t i The relative errors between them are more comparable for prediction problems of different magnitudes.

[0134] The coefficient of determination is the proportion of the sum of squares explained by the hybrid neural network model:

[0135]

[0136] in, This represents the average of the actual values ​​of ethylene concentration.

[0137] The coefficient of determination reflects the goodness of fit of a hybrid neural network model, measuring its explanatory power for changes in the dependent variable. Its value is typically between 0 and 1. A higher coefficient of determination indicates a better fit of the hybrid neural network model to the data.

[0138] The whale optimization algorithm is used to optimize the hyperparameters of the hybrid neural network model, and the model parameters are dynamically adjusted based on the error discriminant function, which can improve the error caused by the weakening correlation between training data and actual prediction data.

[0139] Thus, the ethylene concentration prediction system can update the weights of the hybrid neural network model using the backpropagation algorithm, and evaluate the performance of the hybrid neural network model using the root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination from the error discriminant function. Furthermore, through the error discriminant function, the ethylene concentration prediction system can dynamically adjust the parameters of the hybrid neural network model, thereby effectively predicting ethylene concentration, for example, effectively predicting the concentration of impurities at the top of an ethylene distillation column.

[0140] The following are specific, non-limiting preferred embodiments, which are used to further illustrate the ethylene concentration prediction system proposed in this invention.

[0141] Please refer to Figure 6 , Figure 6 A flowchart for constructing a hybrid neural network model according to an embodiment of the present invention is shown.

[0142] like Figure 6 As shown, the ethylene concentration prediction system can first acquire historical datasets, which can be simulation data or actual scenario data. Then, the system can process the historical datasets using principal component analysis and a denoising autoencoder to obtain feature data. Finally, the feature data and a bidirectional gated recurrent unit (BIGRU) are used to construct a hybrid neural network model.

[0143] The ethylene concentration prediction system initializes the constructed hybrid neural network model and determines its accuracy. When the accuracy of the hybrid neural network model meets the requirements, the prediction result is directly output. When the accuracy of the hybrid neural network model does not meet the requirements, the hyperparameters of the constructed hybrid neural network model are determined using the whale optimization algorithm, and the initialized parameters of the constructed hybrid neural network model are updated based on these hyperparameters.

[0144] Furthermore, such as Figure 6As shown, the whale optimization algorithm first initializes the population, calculates the fitness of each individual whale, and updates the algorithm parameters. Based on the updated algorithm parameters, the method for updating the whale position of each individual whale is determined. When the probability p is greater than or equal to 0.5, bubble net predation is used to update the whale position; when the probability p is less than 0.5, the coefficient vector A is then evaluated. When the probability p is less than 0.5 and the absolute value of coefficient vector A is less than 1, predation encirclement is used to update the whale position; when the probability p is less than 0.5 and the absolute value of coefficient vector A is greater than or equal to 1, search-predation is used to update the whale position.

[0145] Then, the whale position is continuously updated until the termination condition is met, at which point the whale position is output as the optimal solution of the hyperparameters of the hybrid neural network model.

[0146] The parameters of the constructed hybrid neural network model are updated based on the hyperparameters determined by the whale optimization algorithm. Then, the accuracy of the constructed hybrid neural network model is determined. If the accuracy of the constructed hybrid neural network model does not meet the requirements, the whale optimization algorithm is used again to determine the hyperparameters of the constructed hybrid neural network model. When the accuracy of the constructed hybrid neural network model meets the requirements, the predicted ethylene concentration is output using the constructed hybrid neural network model.

[0147] Please refer to Figure 7 , Figure 7 A comparison graph is shown of the prediction results of a hybrid neural network model provided according to an embodiment of the present invention with the prediction results of other models.

[0148] like Figure 7 As shown, data from an ethylene distillation column was acquired under stable operating conditions at a petrochemical plant. To ensure fair comparison with other models, the feature data input to the hybrid neural network model and other models were obtained using the same data processing method: principal component analysis and denoising autoencoder processing. Other models could be LSTM or BiGRU (Bidirectional Gated Recurrent Unit) models. The hybrid neural network model, LSTM model, and BiGRU model were trained using the training dataset from the acquired ethylene distillation column data. During validation, the validation dataset from the ethylene distillation column data was input to predict and calculate the ethylene concentration at the top of the ethylene distillation column under the current ethylene feed flow rate, ethane feed flow rate, feed temperature, and feed pressure. The predicted ethylene concentration was compared with the actual ethylene concentration in the validation dataset to evaluate the hybrid neural network model, LSTM model, and BiGRU model.

[0149] Among the metrics used to evaluate the predictive performance of the hybrid neural network model, namely the WOA-BiGRU model, the root mean square error (RMSE) is 0.045, the mean absolute error (MAE) is 0.036, the mean absolute percentage error (MAPE) is 2.38%, and the coefficient of determination (R²) is [missing data]. 2 The root mean square error (RMSE) was 0.94; among the metrics for evaluating the predictive performance of the LSTM model, the RMSE was 0.057, the mean absolute error (MAE) was 0.044, the mean absolute percentage error (MAPE) was 2.87%, and the coefficient of determination (R²) was 0.94. 2 The root mean square error (RMSE) was 0.89; among the metrics for evaluating the predictive performance of the BiGRU model, the root mean square error (RMSE) was 0.052, the mean absolute error (MAE) was 0.041, the mean absolute percentage error (MAPE) was 2.78%, and the coefficient of determination (R²) was 0.89. 2 The score is 0.90. It can be seen that the WOA-BiGRU model outperforms the other two models on all evaluation metrics.

[0150] In summary, the ethylene concentration prediction method provided by this invention can construct a hybrid neural network model based on a bidirectional gated recurrent unit and optimize the parameters of the hybrid neural network model using the whale optimization algorithm to improve the model's prediction accuracy and training efficiency. This allows for adjustments to the hybrid neural network model in response to changes in process conditions, thereby solving the problems of low prediction accuracy and poor real-time performance caused by changes in process conditions during ethylene separation. Furthermore, by combining principal component analysis and a denoising autoencoder for data preprocessing, the ethylene concentration prediction method provided by this invention can effectively remove noise from the data, enhancing the stability of subsequent predictions.

[0151] Although the methods described above are illustrated and depicted as a series of actions for the sake of simplicity, it should be understood and appreciated that these methods are not limited by the order of the actions, as some actions may occur in a different order and / or concurrently with other actions from the illustrations and descriptions herein or not illustrated and described herein but which may be understood by those skilled in the art, according to one or more embodiments.

[0152] Those skilled in the art will understand that information, signals, and data can be represented using any of a variety of different techniques and arts. For example, the data, instructions, commands, information, signals, bits, symbols, and chips described throughout the above description can be represented by voltage, current, electromagnetic waves, magnetic fields or magnetic particles, light fields or optical particles, or any combination thereof.

[0153] 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 can be implemented as electronic hardware, computer software, or a combination of both. To clearly illustrate this interchangeability between hardware and software, the various illustrative components, blocks, modules, circuits, and steps are described above in a generalized manner 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. Those skilled in the art may implement the described functionality in different ways for each specific application, but such implementation decisions should not be construed as departing from the scope of the invention.

[0154] The various illustrative logic modules and circuits described in conjunction with the embodiments disclosed herein may be implemented or performed using 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. The general-purpose processor may be a microprocessor, but in alternatives, it 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, multiple microprocessors, one or more microprocessors cooperating with a DSP core, or any other such configuration.

[0155] The steps of the methods or algorithms 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 both. The software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor such that the processor can read and write information to / from the storage medium. In an alternative, the storage medium may be integrated into the processor. The processor and storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and storage medium may reside as discrete components in the user terminal.

[0156] In one or more exemplary embodiments, the described functionality may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functionality may be stored or transmitted as one or more instructions or code on or through a computer-readable medium. A computer-readable medium includes both computer storage media and communication media, encompassing any medium that facilitates the transfer of a computer program from one location to another. A storage medium may be any available medium accessible to a computer. By way of example and not limitation, such a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disc storage, disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and is accessible to a computer. Any connection is also legitimately referred to as a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of a medium. As used in this article, disk and disc include compact discs (CDs), laser discs, optical discs, digital multi-purpose discs (DVDs), floppy disks, and Blu-ray discs. Disks typically reproduce data magnetically, while discs reproduce data optically using lasers. Combinations of these should also be included within the scope of computer-readable media.

[0157] The prior description of this disclosure is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to this 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 this disclosure. Therefore, this disclosure is not intended to be limited to the examples and designs described herein, but should be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting ethylene concentration, characterized in that, Including the following steps: Acquire ethylene distillation column data, which includes variables related to the ethylene distillation column products, including ethylene feed rate, ethane feed rate, feed temperature, feed pressure, reflux ratio, reboiler heat load, and column top condenser heat load. Principal component analysis and a noise-reducing autoencoder were used to perform dimensionality reduction and noise reduction on the ethylene distillation column data to determine the feature data. as well as The feature data is input into a hybrid neural network model to predict ethylene concentration. The hybrid neural network model is constructed based on bidirectional gated recurrent units, and its optimal hyperparameters are determined using the whale optimization algorithm. The steps for determining the preferred hyperparameters of the hybrid neural network model based on the whale optimization algorithm include: In response to the fact that the prediction accuracy of the hybrid neural network model does not meet the requirements, the hyperparameters of the hybrid neural network model are determined according to the whale optimization algorithm, and the prediction accuracy of the hybrid neural network model is determined by the error discriminant function; The hybrid neural network model is trained based on the hyperparameters to update the parameters of the hybrid neural network model, and the prediction accuracy of the hybrid neural network model after updating the parameters is determined; and Repeat the above steps until the prediction accuracy of the hybrid neural network model meets the requirements, and determine the hyperparameters for training the hybrid neural network model as the preferred hyperparameters.

2. The method for predicting ethylene concentration as described in claim 1, characterized in that, The noise reduction autoencoder includes an encoder and a decoder, and the encoder and the decoder each include a two-layer fully connected neural network. The encoder and the decoder have a symmetrical structure.

3. The method for predicting ethylene concentration as described in claim 2, characterized in that, The encoder is constructed as follows: X encoder =f(f(W1X+B1)W2+B2), Where f is the activation function, W1 is the weight of the first fully connected layer of the encoder's neural network, W2 is the weight of the second fully connected layer of the encoder's neural network, B1 is the bias of the first fully connected layer of the encoder's neural network, B2 is the bias of the second fully connected layer of the encoder's neural network, and X... encoder The first feature is obtained by the encoder encoding the dimension-reduced ethylene distillation column data; The construction process of the decoder is as follows: X decoder =f(f(W2X encoder +B2)W1+B1), Among them, X decoder The feature data output by the noise reduction autoencoder.

4. The method for predicting ethylene concentration as described in claim 1, characterized in that, The step of determining the hyperparameters of the hybrid neural network model based on the whale optimization algorithm when the prediction accuracy of the hybrid neural network model does not meet the requirements includes: S1: In response to the fact that the accuracy of the hybrid neural network model does not meet the requirements, initialize the number of individual whales, the whale positions, and the algorithm parameters, where the whale positions represent the hyperparameters of the hybrid neural network model; S2: Calculate the fitness of the whale individual based on its position to determine the initial optimal position of the whale individual as the preferred solution; S3: Update the algorithm parameters and, based on the algorithm parameters, update the whale position of the individual whale by means of bubble net hunting, encircling prey, or searching for prey; S4: Calculate the fitness of the whale individual based on the updated whale position to determine the optimal position of the whale individual; S5: In response to the optimal whale individual's position being better than the preferred solution, the optimal whale individual's position is taken as the preferred solution; and S6: Repeat steps S3 to S5 until the termination condition is met, and output the whale position of the preferred solution to determine the hyperparameters of the hybrid neural network model.

5. The method for predicting ethylene concentration as described in claim 4, characterized in that, Step S3 includes: In response to a probability p greater than or equal to 0.5, the whale position of the individual whale is updated by using a bubble net feeding method; In response to the probability p being less than 0.5 and the absolute value of the coefficient vector A being less than 1, the whale position of the individual whale is updated by employing a predator-hunting strategy; and In response to the probability p being less than 0.5 and the absolute value of the coefficient vector A being greater than or equal to 1, the whale position of the individual whale is updated by adopting a search and predation method.

6. The method for predicting ethylene concentration as described in claim 1, characterized in that, The error discrimination function includes root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination.

7. A system for predicting ethylene concentration, characterized in that, include: Memory, on which computer instructions are stored; and A processor, connected to the memory, and configured to execute computer instructions stored in the memory to implement the method for predicting ethylene concentration as described in any one of claims 1 to 6.

8. A computer-readable storage medium storing computer instructions thereon, characterized in that, When the computer instructions are executed by the processor, the method for predicting ethylene concentration as described in any one of claims 1 to 6 is implemented.

Images