Ethylene section parameter prediction method and device, storage medium and program product

Abstract

The invention relates to the technical field of industrial control, in particular to an ethylene section parameter prediction method and device, a storage medium and a program product. According to the method, a workshop section prediction model is trained in advance, real-time operation data of an ethylene production workshop section are obtained and input into the trained workshop section prediction model, and a target prediction value, output by the workshop section prediction model, of the ethylene production workshop section is obtained; and the section prediction model is obtained by training a comprehensive loss function combining an original scale error and a standardized error, so that the model is ensured to have the fitting precision of standardized data, and a predicted value is ensured to conform to a physical rule in actual production, thereby improving the prediction accuracy and efficiency of the ethylene production section parameters.

Description

Technical Field

[0001] The present application relates to the field of industrial production control, and particularly to a method, device, storage medium and program product for predicting parameters in an ethylene section. Background Art

[0002] In the process of ethylene production, it is crucial to accurately predict various parameters in the section to optimize the production process and reduce energy consumption.

[0003] However, traditional modeling methods for the ethylene section usually rely on physical and chemical process models or empirical models. Physical and chemical models require detailed reaction mechanisms and thermodynamic parameters, which are time-consuming to develop and have insufficient accuracy under complex working conditions. Empirical models, on the other hand, are overly dependent on historical data and are extremely vulnerable to data quality and generalization ability limitations, resulting in the existing prediction of ethylene section parameters often facing problems of insufficient accuracy and high computational complexity.

[0004] Therefore, there is an urgent need for a method for predicting parameters in the ethylene section to improve the prediction accuracy and efficiency of ethylene section parameters. Summary of the Invention

[0005] The present invention provides a method, device, storage medium and program product for predicting parameters in the ethylene section, so as to improve the prediction accuracy and efficiency of ethylene production parameters, thereby improving ethylene production efficiency.

[0006] In a first aspect, the present application provides a method for predicting parameters in the ethylene section, the method comprising:

[0007] Obtain real-time operation data of the ethylene production section, and input the real-time operation data into a trained section prediction model;

[0008] Obtain a target prediction value of the ethylene production section output by the section prediction model;

[0009] Wherein, the section prediction model is obtained by respectively performing prediction processing on the historical operation data of the ethylene production section and the standardized historical operation data to obtain the original prediction value and the standardized prediction value of the compressor power, and constructing a comprehensive loss function of the section prediction model based on the original prediction value and the standardized prediction value, and training the model based on the comprehensive loss function.

[0010] Optionally, the section prediction model is trained in the following manner:

[0011] Obtain the historical operation data of the ethylene production section, the historical operation data including the operation parameter sets of multiple compression sections in the ethylene production process and the true value of the compressor power;

[0012] Standardize the historical operation data, and perform prediction processing based on the historical operation data and the standardized historical operation data to obtain the original predicted value and the standardized predicted value of the compressor power respectively;

[0013] Based on the original error between the original predicted value and the true value, and the standardized error between the standardized predicted value and the true value, obtain the comprehensive loss function of the section prediction model;

[0014] Based on the comprehensive loss function, adjust the parameters of the section prediction model until the section prediction model meets the preset convergence condition to obtain the trained section prediction model.

[0015] Optionally, the obtaining of the comprehensive loss function of the section prediction model based on the original error between the original predicted value and the true value, and the standardized error between the standardized predicted value and the true value includes:

[0016] Based on the standardized error, construct the standardized loss of the section prediction model;

[0017] Based on the original error, construct the inverse standardized loss of the section prediction model;

[0018] Perform weighted combination based on the standardized loss and the inverse standardized loss to obtain the comprehensive loss function.

[0019] Optionally, the standardizing of the historical operation data includes:

[0020] Perform anomaly filtering on the historical operation data, and perform normalization processing on the historical operation data based on the mean and standard deviation of the filtered historical operation data;

[0021] Perform data partitioning on the normalized historical operation data to obtain the corresponding training set and test set.

[0022] Optionally, the section prediction model is determined based on a preset Long Short-Term Memory (LSTM) model architecture, and the LSTM model architecture includes:

[0023] An input layer for receiving the standardized input data;

[0024] An LSTM layer for extracting the temporal features in the input data and transmitting the temporal features to the fully connected layer;

[0025] The fully connected layer for performing non-linear transformation on the temporal features and transmitting the obtained one-dimensional features to the output layer;

[0026] The output layer is used to generate corresponding predicted values based on the one-dimensional features.

[0027] Optionally, the LSTM model architecture further includes a dropout layer for randomly dropping out some neurons.

[0028] Optionally, the model parameters of the section prediction model at least include one or more combinations of the following parameters:

[0029] The input feature dimension, which represents the number of features of the input data received by the model each time;

[0030] The number of LSTM network layers;

[0031] The hidden layer dimension, which represents the hidden state dimension of each LSTM layer;

[0032] The sequence length, which represents the length of the input data received by the model each time in the time dimension;

[0033] The batch size, which represents the number of samples to be processed in each model training;

[0034] The dropout probability, which represents the probability of randomly dropping out neuron connections in the dropout layer;

[0035] The learning rate, which represents the step size for updating model parameters;

[0036] The number of training epochs, which represents the number of times the training set is traversed;

[0037] The batch dimension order, which corresponds to the format of the input data.

[0038] In a second aspect, the present application provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements any one of the ethylene production section parameter prediction methods in the first aspect above.

[0039] In a third aspect, the present application provides a computer storage medium. The computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed by the processor, they implement any one of the ethylene production section parameter prediction methods in the first aspect above.

[0040] In a fourth aspect, a computer program product provided by an embodiment of the present application includes computer program instructions. When the computer program instructions are executed by the processor, they implement any one of the ethylene production section parameter prediction methods in the first aspect above.

[0041] The beneficial effects of the present invention are as follows:

[0042] An embodiment of the present application provides a method for predicting parameters in an ethylene production section. In this method, a section prediction model is pre-trained. By obtaining the real-time operation data of the ethylene production section and inputting it into the trained section prediction model, a target prediction value of the ethylene production section output by the section prediction model is obtained. Moreover, the section prediction model is trained by a comprehensive loss function that combines the original scale error and the standardized error, which not only ensures the fitting accuracy of the model for standardized data but also ensures that the predicted value conforms to the physical laws in actual production, thereby improving the prediction accuracy and efficiency of the parameters in the ethylene production section. Description of the Drawings

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only the embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on the provided drawings.

[0044] Figure 1 It is a schematic diagram of the training process of a section prediction model provided by an embodiment of the present application;

[0045] Figure 2 It is a schematic diagram of the neural network structure of a section prediction model provided by an embodiment of the present application;

[0046] Figure 3 It is a schematic diagram of the downward trend of the model loss function provided by an embodiment of the present application;

[0047] Figure 4 It is a flowchart of a method for predicting parameters in an ethylene production section provided by an embodiment of the present application;

[0048] Figures 5(a)-(e) are respectively schematic diagrams of the comparison between the predicted values and the true values of five compressors in a multi-stage compression section of ethylene production provided by an embodiment of the present application;

[0049] Figure 6 It is a schematic diagram of the structure of a computer device provided by an embodiment of the present application. Detailed Embodiments

[0050] To make the objectives, technical solutions and advantages of the present application more clearly understood, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part rather than all of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without making creative efforts shall fall within the scope of protection of the present application. Without conflict, the embodiments in the present application and the features in the embodiments may be combined arbitrarily with each other. And although the logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than here.

[0051] The terms "first" and "second" in the description and claims of the present application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the term "comprising" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products or devices. "Multiple" in the present application may mean at least two, for example, it may be two, three or more, and the embodiments of the present application do not make limitations.

[0052] The term "and / or" in the embodiments of the present application is only a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this article generally represents an "or" relationship between the associated objects before and after.

[0053] It can be understood that in the following specific embodiments of the present application, related to data such as the ethylene production process, when the embodiments of the present application are applied to specific products or technologies, relevant permissions or approvals need to be obtained, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions. For example, relevant volunteers can be recruited and relevant agreements on authorizing data of the volunteers can be signed, and then the data of these volunteers can be used for implementation; or, implementation can be carried out within the scope of an authorized organization, and the data of the internal members of the organization can be used to implement the following embodiments for data management; or, the relevant data used in specific implementation are all simulated data, for example, simulated data generated in a virtual scenario.

[0054] The embodiments of the present application relate to artificial intelligence and machine learning (ML) technologies, and are mainly designed based on machine learning in artificial intelligence.

[0055] Artificial intelligence uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, including the theory, methods, technologies, and application systems for perceiving the environment, acquiring knowledge, and using knowledge to obtain results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a way similar to human intelligence. Artificial intelligence also studies the design principles and implementation methods of various intelligent machines, enabling machines to have the functions of perception, reasoning, and decision-making.

[0056] Artificial intelligence technology is an interdisciplinary subject with a wide range of fields involved, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. The software technologies of artificial intelligence mainly include several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0057] Machine learning is an interdisciplinary subject that involves multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve their own performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent, and its applications cover all fields of artificial intelligence. Machine learning and deep learning usually include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and rote learning.

[0058] Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent, and its applications cover all fields of artificial intelligence. Machine learning and deep learning usually include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, etc. An artificial neural network (ANN) abstracts the human brain neuron network from the perspective of information processing, establishes a simple model, and forms different networks according to different connection methods. A neural network is an operation model composed of a large number of nodes (or neurons) connected to each other. Each node represents a specific output function, called an activation function. The connection between each two nodes represents a weighted value for the signal passing through this connection, called a weight, which is equivalent to the memory of the artificial neural network. The output of the network varies depending on the connection method, weight value, and activation function of the network, and the network itself usually approximates a certain algorithm or function in nature or may be an expression of a logical strategy.

[0059] The design concept of the embodiments of the present application will be briefly introduced below.

[0060] In the process of ethylene production, it is crucial to accurately predict various parameters in the process section to optimize the production process and reduce energy consumption.

[0061] However, traditional modeling methods for the ethylene process section usually rely on physical and chemical process models or empirical models. Building a physical and chemical model requires detailed reaction mechanisms and thermodynamic and kinetic parameters, and the construction and development process is both time-consuming and challenging. Especially under complex working conditions, the accuracy of the model often fails to meet the actual production requirements. Although empirical models can simplify the modeling process, they rely too much on a large amount of historical data, and the prediction accuracy is severely limited by the data quality and the model generalization ability. For example, empirical models are sensitive to data noise and cannot capture non-linear dynamic processes, resulting in low prediction accuracy. Therefore, the existing parameter prediction processes for the ethylene process section generally have the defects of insufficient accuracy and high computational complexity.

[0062] In view of the above problems, the embodiments of the present application provide a method for predicting parameters in the ethylene production process section. The method pre-trains a process section prediction model, obtains real-time operation data of the ethylene production process section, and inputs it into the pre-trained process section prediction model to obtain the target prediction value of the ethylene production process section output by the process section prediction model. The process section prediction model is trained by a comprehensive loss function that combines the original scale error and the standardized error, which not only ensures the fitting accuracy of the model for standardized data but also ensures that the prediction value conforms to the physical laws in actual production, thereby improving the prediction accuracy and efficiency of the parameters in the ethylene production process section.

[0063] Furthermore, in order to further improve the prediction accuracy and efficiency of the parameters in the ethylene production section, this application can construct a section prediction model through the LSTM model architecture, so that the section prediction model in this application combines the advantages of data-driven modeling and traditional process modeling, and through the powerful time series processing ability of the LSTM network, realizes the efficient prediction of the complex dynamic process in the ethylene section. During the model application process, this application can collect and process the actual operation data of the factory, use the LSTM network to capture the time series characteristics, and optimize the model parameters through the comprehensive loss function. The experimental results show that the model optimized in this way can accurately predict the key parameters in the production process through learning the actual operation data of the factory, and show high prediction accuracy and stability under various operating conditions, providing effective support for the optimal control and monitoring decision-making of the ethylene production process. Moreover, compared with the traditional method, the section prediction model of this application overcomes the problems of high computational complexity, low prediction accuracy, and long model construction in the traditional modeling method, improves the prediction accuracy and efficiency, and has higher computational efficiency and stronger adaptability. It can accurately predict the production parameters of the ethylene section without relying on an accurate reaction mechanism or a cumbersome physical and chemical process model.

[0064] Next, the method provided by the exemplary embodiment of this application will be described with reference to the accompanying drawings. It should be noted that the above application scenarios are only shown for the convenience of understanding the spirit and principle of this application, and the embodiments of this application are not limited in this regard.

[0065] In the embodiment of this application, the section prediction model can be used to implement the prediction process of the production parameters in the above-mentioned ethylene production section. Before the section prediction model is put into use, it is necessary to pre-train the model to make the section prediction model converge. Next, for the convenience of describing the model application process, the training process of the section prediction model will be introduced first.

[0066] Please refer to Figure 1 , which is a schematic diagram of the training process of a section prediction model provided by the embodiment of this application. The specific implementation process of this training process is as follows:

[0067] Step 101: Obtain the historical operation data of the ethylene production section.

[0068] In the embodiment of this application, the historical operation data of the ethylene production section may include the operation parameter sets of multiple compression sections in the ethylene production process and the true values of the compressor power.

[0069] Specifically, the compression section refers to the process stage in ethylene production where the raw material gas (such as pyrolysis gas) is compressed in multiple stages. Ethylene production usually involves multiple compressors in series or parallel, compressing the gas to the pressure conditions required for subsequent separation, rectification, and other processes through step-by-step pressure increase. Each compression section corresponds to an independent compression unit, and its operating parameters directly affect the overall production efficiency and energy consumption. The compressor power represents the energy input required to drive the compressor operation, which directly reflects the energy consumption level of the compression section. It is the core index for optimizing production efficiency and cost, and also the prediction target of this application. By predicting the compressor power value in real time through the section prediction model, it can provide data support for process optimization.

[0070] In a possible implementation, the operating parameters of the compression section represent the real-time physical state and operating conditions of the ethylene compression section, and may include key parameters in the production process such as temperature, pressure, molar flow rate, hydrogen molar fraction, methane molar fraction, ethylene molar fraction, etc.

[0071] In a possible implementation, this application can collect a large amount of ethylene production data through process simulation software, including but not limited to chemical process simulation and optimization software such as HYSYS software under AspenPlus V12, and simulate by raising the feed stream to a certain methane molar fraction (for example, 20% methane molar fraction) as the fluctuated data to obtain multiple sets of simulation data (for example, 2740 sets of simulation data). The generated simulation data can be used as the time series data required for model training to learn the non-linear relationship between compressor power and input parameters, solving the problem of insufficient data in actual factories.

[0072] Step 102: Standardize the historical operation data.

[0073] In the embodiments of this application, before inputting the historical operation data into the section prediction model trained this time, the historical operation data will also be standardized to improve the model training efficiency and model performance.

[0074] In a possible implementation, in order to ensure the quality and consistency of the input data, this application can perform preprocessing processes such as outlier filtering and normalization processing on the historical operation data, and divide the input data into a training set and a test set. Use normalization methods such as the MinMaxScaler class to standardize the training data, validation, and test data, thereby improving the model training effect.

[0075] Specifically, the main steps of data preprocessing in this application include:

[0076] (1) Data loading: The original historical operation data containing characteristic parameters such as molar flow rate, temperature, and pressure can be read from a Comma-Separated Values (CSV) file through methods like the getData function. These characteristic parameters will also serve as the input data for the model.

[0077] (2) Normalization processing: The input data will undergo normalization to ensure consistent scale during training, preventing the values of certain features from being too large or too small and affecting the training effect of the model.

[0078] In a possible implementation, the normalization process may include subtracting the mean of the data and dividing by the standard deviation.

[0079] (3) Data splitting: The input data is divided into a training set and a test set. The training set is used for model training, and the test set is used to evaluate the performance of the model.

[0080] In this way, after normalization processing, it can be ensured that the input data in this application has a unified scale range, which is beneficial for the network to train more quickly and stably. For example, in this model, the shape of the input data can be (batch_size, sequence_length, input_size), where batch_size is the number of samples per batch, sequence_length is the length of the time series, and input_size is the number of features at each time step.

[0081] Step 103: Perform prediction processing based on the historical operation data and the normalized historical operation data to obtain the original predicted value and the normalized predicted value of the compressor power respectively.

[0082] In the embodiment of this application, after normalizing the historical operation data, the historical operation data and the normalized historical operation data can be input into the process section prediction model trained this time respectively, so as to obtain the corresponding original predicted value and normalized predicted value.

[0083] In a possible implementation, this application can use the LSTM model as the core framework of the process section prediction model. By first applying the LSTM network to the dynamic parameter prediction of the ethylene multi-stage compression process section, its long-term dependence capture ability can be used to process complex time series data, thereby improving the ability of the process section prediction model to handle long-term dependence relationships in time series data.

[0084] Specifically, the LSTM model in this application may include an Input Layer, an LSTM Layer, a Fully Connected Layer, and an Output Layer. Among them, the input layer is mainly used to receive the standardized input data, and the dimension of the input data is determined by the input size parameter. The LSTM layer is defined by the LSTM Model class. This layer contains multiple LSTM units, and the input and output sizes of each unit are determined by the input size and hidden size parameters. The LSTM layer can handle the long-term dependencies of time series data and transfer historical information to future time steps. One or more fully connected layers (Linear layers) are connected after the LSTM layer. These layers are responsible for extracting features from the output of the LSTM and performing non-linear mapping to output the prediction results of the model. The output layer is used to output the final predicted value through a linear activation function to generate the predicted value for each time step.

[0085] Specifically, referring to Figure 2 The following is a schematic diagram of the neural network structure of a section prediction model provided by an embodiment of this application. The Figure 2 structure of the LSTM model in it includes an Input layer, an LSTM layer, fully connected layers (Dense1 and Dense2), and an Output layer. The above layers are stacked in sequence to fully exploit the features of time series data and ensure that the model can effectively learn and predict the dynamic evolution of the ethylene section. Among them,

[0086] (1) The input layer is used to receive the input data after being standardized, such as various parameters such as temperature, pressure, flow rate, and mole fraction in the ethylene production section.

[0087] (2) The LSTM layer is used to capture the dynamic characteristics in time series data and transfer the information to the fully connected layer to extract deeper features. Therefore, the LSTM layer is the core part of this model, responsible for processing time series data and capturing the long-term dependencies in the data. In this way, the LSTM network can effectively remember and transfer the historical information in the sequence, ensuring that the prediction at the current moment depends not only on the current input but also combines the information of historical time steps.

[0088] In a possible implementation manner, the hidden state size of the LSTM layer in this application can be 64, that is, the LSTM layer in this application can map the input of each time step to a 64-dimensional high-dimensional space, and the output shape is (batch_size, sequence_length, hidden_size).

[0089] (3) Fully connected layer, which is used to further extract the features of the data and perform non-linear transformation. Multiple fully connected layers may be included in this model. For example, Dense1 and Dense2 shown above. Each fully connected layer can use the Rectified Linear Unit (ReLU) activation function to extract the data features. This activation function can effectively introduce non-linear features, thereby increasing the expressive power of the model. Figure 2 As shown above, for each fully connected layer, the Rectified Linear Unit (ReLU) activation function can be used to extract data features. This activation function can effectively introduce non-linear features, thereby increasing the expressive power of the model.

[0090] In a possible implementation, the fully connected layer in this application can map the input data from 64 dimensions to 128 dimensions and then to 1 dimension to achieve the final prediction.

[0091] (4) Output layer, which is used to generate the prediction value results at each time point according to the output of the LSTM layer.

[0092] In a possible implementation, the output layer in this application can adopt a linear activation function to generate the specific prediction values at each time step. This output layer is responsible for converting the internal state of the LSTM layer into the final prediction result, that is, the numerical value output by the model. In a possible implementation, the output shape of this layer can be (batch_size, 1), that is, only one prediction value is output for each batch of data, representing the final prediction result of the time series.

[0093] In a possible implementation, to prevent overfitting, a Dropout layer can be further added after the LSTM layer in the model of this application (the Dropout layer, that is, Dropout shown above). This layer can reduce the complexity of the model by randomly discarding some neurons and enhance its generalization ability. Figure 2 This layer can reduce the complexity of the model by randomly discarding some neurons and enhance its generalization ability.

[0094] In a possible implementation, the dropout rate of the Dropout layer in this application can be set to 0.2, that is, each neuron is discarded with a probability of 20%, thereby improving the robustness of the model training process, especially when the data is less or the model is more complex.

[0095] In this way, the model network structure in this application combines the actual production data with deep learning technology, and can capture the complex dynamics in the time series data through the advantages of the LSTM network, improving the prediction accuracy and efficiency of the model.

[0096] In a possible implementation, the model parameters in this application may include:

[0097] (1) Input feature dimension: the number of input features at each time step, that is, the dimension of the input vector received by the model each time. For example, if the input data has 6 features (such as temperature, pressure, etc.), the input size should be set to 6. In this application, the input feature dimension can be defaulted to 6.

[0098] (2) Number of LSTM layers: The number of layers in the LSTM network. A deeper network has stronger expressive power but will also increase the computational burden accordingly. In this application, the number of LSTM layers can be defaulted to 2.

[0099] (3) Hidden layer dimension: The dimension of the hidden state of each layer of LSTM, which controls the output size of each LSTM cell. A larger hidden size can capture more feature information. In this application, the hidden layer dimension value can be defaulted to 64.

[0100] (4) Sequence length: The length of the time series input to the model each time, that is, the length of the input data in the time dimension, and also the number of data processed by the LSTM at each time step. In this application, the sequence length value can be defaulted to 12.

[0101] (5) Batch size: The number of samples processed in each training. A smaller batch can improve the stability of training, while a larger batch helps to accelerate training. In this application, the batch size can be defaulted to 4.

[0102] (6) Dropout probability (dropout): The probability of randomly discarding neuron connections to prevent overfitting. During training, 20% of the neurons will be randomly discarded to enhance the generalization ability of the model. In this application, the dropout probability value can be defaulted to 0.2.

[0103] (7) Learning rate (Learning Rate, lr): The step size for updating the model parameters. A smaller learning rate will cause the model to converge slowly, while a larger learning rate may lead to unstable training. In this application, the learning rate (lr) can be defaulted to 0.01.

[0104] (8) Number of training epochs: The number of times the dataset is traversed. In each epoch of training, the model will process the entire training set until the maximum number of training epochs is reached. In this application, the number of training epochs can be defaulted to 100.

[0105] (9) Batch dimension order (batch first): If set to True, the format of the input data is (batchsize, sequence length, input size), that is, the batch dimension is at the front. If set to False, the format is (sequence length, batch size, input size). In this application, the batch dimension order can be defaulted to True.

[0106] Thus, after constructing the process prediction model for this training, the original historical operation data and the standardized historical operation data can be used as input data and input into the process prediction model respectively, so as to obtain the original predicted values corresponding to the original input data and the standardized predicted values corresponding to the standardized data output by the model.

[0107] Step 104: Obtain the combined loss function of the process prediction model based on the original error between the original predicted value and the true value, and the standardized error between the standardized predicted value and the true value.

[0108] In this application, in order to ensure the prediction accuracy of the process prediction model and conform to physical laws, a combined loss function (CombinedMSELoss) that combines the original error and the standardized error is used to balance the adaptability of the model to the standardized data and the original data.

[0109] In a possible implementation manner, this application can construct the standardized loss of the process prediction model through the standardized error, and construct the inverse standardized loss of the process prediction model through the original error, so as to perform weighted combination on the standardized loss and the inverse standardized loss to obtain the combined loss function.

[0110] In a possible implementation manner, the standardized error between the standardized predicted value and the true value in this application can be represented by the standardized mean squared error (MSE). The standardized MSE calculates the performance of the model on the standardized data. Standardization is to handle the scale differences between different features and ensure that the training of the model is not affected by the scale of a single feature. Thus, through this standardized error, the standardized loss of the process prediction model can be constructed.

[0111] Specifically, the formula for the standardized loss is as follows:

[0112]

[0113] Among them, represents the predicted value of the model on the standardized data;

[0114] y i represents the true value after standardization;

[0115] N represents the total number of samples.

[0116] In a possible implementation, the original error between the original predicted value and the true value in this application can be represented by the mean squared error (MSE) of the original scale after inverse normalization. The MSE of the original scale after inverse normalization takes into account the performance of the model after restoring to the original scale. Since the standardization process is carried out through the mean and standard deviation, inverse normalization can restore the predicted value and the true value to the original physical unit or numerical range. Thus, through this original scale MSE, the inverse normalization loss of the process prediction model can be constructed.

[0117] Specifically, the formula for the inverse normalization loss can be shown as follows:

[0118]

[0119] Where: is the predicted value restored to the original scale.

[0120] y original,i = y i ×std + mean is the true value restored to the original scale.

[0121] std and mean are the standard deviation and mean of the training data respectively.

[0122] Thus, by restoring the predicted value and the true value to the original scale, it can be ensured that the model can make effective predictions on the original data (such as physical quantities in the actual industrial process) in practical applications.

[0123] In a possible implementation, the comprehensive loss function in this application can be obtained by weighted combination of the standardized MSE and the inverse normalization MSE.

[0124] Specifically, the weight coefficients α and β can be used to balance the influence of the two parts of the original loss and the standardized loss respectively. The formula for the comprehensive loss function can be shown as follows:

[0125] LOSS combined = α × LOSS standardized + β × LOSS original

[0126] Where α and β are the weights of the loss function, used to control the relative influence of the standardized loss and the inverse normalization loss (i.e., the original scale loss). And α + β = 1, which can be flexibly set according to actual needs. For example, setting α = 0.5 and β = 0.5 means that the weights of the two parts of the loss are equal.

[0127] Thus, by weighted combination of the standardized loss and the original scale loss, considering the prediction errors of two different scales, and adjusting the proportion of the two in the total loss according to the given weights, the process prediction model of this application can take into account the fitting accuracy of the standardized data and the actual prediction ability of the original data during the training process.

[0128] Step 105: Based on the comprehensive loss function, adjust the parameters of the section prediction model until the section prediction model meets the preset convergence condition, and obtain the trained section prediction model.

[0129] In the embodiment of the present application, the parameters of the model are adjusted by minimizing the comprehensive loss function, and when the number of training times of the section prediction model reaches the preset number of iterations, or when the loss function no longer changes, the trained exponential prediction model can be output.

[0130] Specifically, the architecture of the section prediction model constructed in the present application can be shown in Table 1 below:

[0131]

[0132] Table 1

[0133] In Table 1 above, Layer represents the constituent unit of the neural network structure;

[0134] The activation function is a non-linear function used to determine the output form of the neuron, including the hyperbolic tangent function (Tanh), the rectified linear unit (ReLU), and the linear function (Linear). Among them, Tanh represents the S-shaped activation function with an output range of (-1, 1), and ReLU represents the activation function with an output of max(0, x);

[0135] The number of neurons represents the total amount of connection weight parameters of the neurons in this layer;

[0136] The output shape includes parameters in three dimensions: the number of batches, the time step, and the feature dimension.

[0137] In a possible implementation manner, the training process of the section prediction model in the present application can be realized through the following steps:

[0138] (1) Data loading: Load batch data from the training set and the test set respectively, and ensure that the data is processed batch by batch through the DataLoader module.

[0139] (2) Forward propagation: Send the input data into the section prediction model and process it through the LSTM network to obtain the prediction result.

[0140] (3) Loss calculation: Calculate the loss between the predicted value and the actual value of the model, and use the custom comprehensive loss function.

[0141] (4) Backward propagation: Calculate the gradient through the backward propagation algorithm and update the parameters of the model using an optimizer (including but not limited to the Adam optimizer).

[0142] (5) Learning rate scheduling: Accelerate the training process by adjusting the learning rate (ExponentialLR) to avoid oscillations during training.

[0143] (6) Model saving: Save the weights of the current model every specified number of training epochs to ensure that the intermediate results during training can be saved and restored when needed.

[0144] In a possible implementation, after the model training is completed, the present application can also evaluate the model through the following steps to test the model performance:

[0145] (1) Load the trained model: Use the saved model weight file to restore the model parameters.

[0146] (2) Load the test data: Load the data from the test set for prediction.

[0147] (3) Calculate the error: Use the Root Mean Squared Error (RMSE) to calculate and evaluate the prediction error of the model in the standardized space and the original scale space. Among them, the test set error during the evaluation process is crucial for detecting the model performance.

[0148] In a possible implementation, in order to further improve the performance of the model, especially to enable the model to obtain better prediction effects on different tasks and datasets, the present application can optimize and adjust the model through the following optimization means:

[0149] (1) Batch normalization: Add a batch normalization layer after the fully connected layer to optimize the training process of the network and accelerate the convergence speed.

[0150] (2) Dropout layer: Add a Dropout layer between the fully connected layer and the Time Distributed layer to prevent overfitting and improve the generalization ability of the model.

[0151] (3) Parameter adjustment: Optimize the model performance by adjusting hyperparameters such as the number of LSTM layers, hidden layer size, learning rate, and dropout probability.

[0152] In summary, the process section prediction model in the present application can be effectively applied to time series prediction problems, especially in the application scenarios involving ethylene process section prediction, to ensure the accuracy and physical rationality of the prediction.

[0153] In a possible implementation, refer to Figure 3The figure shows a schematic diagram of the downward trend of the model loss function provided by the embodiment of the present application. In the present application, five compressor data can be imported into the model respectively, and after parameter optimization, the LSTM model is trained. Above Figure 3 Taking the training loss curve of Compressor 1 after parameter optimization as an example, the horizontal axis is the number of training epochs, and the vertical axis is the training loss, which intuitively reflects the optimization process of the comprehensive loss function. From this Figure 3 It can be seen that Figure 3 It can be seen that the training loss of the model drops rapidly in the initial stage, indicating that the model has learned more information in the early iterations and quickly adapted to the training data. This rapid loss drop usually occurs at the beginning of training, proving that the network's learning of the data is very effective. In the subsequent training process, the loss value gradually levels off, indicating that the model has entered the convergence stage and there is limited room for further optimization. The loss value always remains at a low level and has small fluctuations, indicating that the prediction accuracy of the model has tended to be stable. Thus, the training results show that the model has successfully extracted effective features from the training data, and in the later stage of training, the convergence of the loss function is good, and the model shows good stability and reliability after 100 rounds of training.

[0154] After obtaining the trained process prediction model, the embodiment of the present application can use the process prediction model to perform prediction processing on the actual operation data of the ethylene production process to be predicted, so as to obtain the target prediction value of the ethylene production process output by the process prediction model.

[0155] Refer to Figure 4 The figure shows a flowchart of a method for predicting parameters of an ethylene production process provided by the embodiment of the present application. The specific implementation process of this method is as follows:

[0156] Step 401: Obtain the real-time operation data of the ethylene production process and input the real-time operation data into the trained process prediction model.

[0157] In the embodiment of the present application, after respectively performing prediction processing on the historical operation data and the standardized historical operation data of the ethylene production process to obtain the original prediction value and the standardized prediction value of the compressor power, a comprehensive loss function of the process prediction model can be constructed through the original prediction value and the standardized prediction value, so as to perform model training through the comprehensive loss function to obtain a trained process prediction model. After obtaining the trained process prediction model, the process prediction model can be applied to the prediction of actual parameters of the ethylene production process, that is, the actual operation parameters of the ethylene production process to be predicted are input into the process prediction model.

[0158] Step 402: Obtain the target prediction value of the ethylene production section output by the section prediction model.

[0159] In the embodiment of the present application, through the trained section prediction model, the actual operating parameters of the ethylene production section are predicted to obtain the target prediction value of the ethylene production section. By outputting the target prediction value of the ethylene production section to the operator or the relevant process control system, the production process can be adjusted in real time, the process parameters can be optimized, and the ethylene production efficiency and the quality of related products can be improved.

[0160] It is worth mentioning that in the embodiment of the present application, the prediction process of the ethylene production section during the training process is the same as that during the actual application process. Therefore, the detailed introduction of the foregoing training process can be referred to for this process, and no further elaboration will be provided here.

[0161] In a possible implementation manner, refer to FIGS. 5(a)-(e), which are respectively schematic diagrams comparing the predicted values and the true values of the five compressors numbered 1-5 in a multi-stage compression section of ethylene production provided by the embodiment of the present application. After the model training in the present application, the trained model is imported and the prediction set is predicted. Although the actual power of the compressors numbered 1-5 is different, according to FIGS. 5(a)-(e) above, the difference between the model predicted value and the actual value of each compressor is extremely small, and the errors between the vast majority of the predicted values and the actual values are maintained at a low level, verifying the accuracy of the section prediction model of the embodiment of the present application in capturing data features. For example, taking the compressor 1 shown in FIG. 5(a) as an example, the difference between the model predicted value 13010.20 and the actual value 13008.58 is only 1.62, and the errors between most other predicted values and the actual values are maintained at a low level. And the mean square error of the standardized test set is 0.0007, which means that on the standardized data, the prediction error of the model is very low, indicating that the model can well adapt to the changes of the input features. The root mean square error of the non-standardized test set is 1.1280, indicating that on the actual scale of the physical quantity, the prediction error of the model is also relatively small, showing its physical rationality. Similarly, the actual power sizes of the standardized test set mean square errors of the other compressors numbered 2-5 are 0.0003, 0.0002, 0.0001, 0.0001 respectively, and the root mean square errors of the non-standardized test sets are 2.3264, 0.9963, 0.1398, 0.6995 respectively, all of which are small values, further verifying that the prediction errors of the model on the standardized data and the actual scale of the physical quantity are both small. At the same time, combined with the Figure 3 loss curve above, the model converges rapidly in the initial stage of training, and the loss tends to be stable in the later stage, indicating that the training process is successful and the model can achieve a good fitting effect on the training set.

[0162] Please refer to Figure 6 As shown, based on the same inventive concept, an embodiment of the present application further provides a computer device 60. In one embodiment, the computer device may be a device dedicated to predicting the parameters of the ethylene production section, or a control device for overall control of the ethylene production process. The computer device is as Figure 6 shown, and includes a memory 601, a communication module 603, and one or more processors 602.

[0163] The memory 601 is used to store the computer program executed by the processor 602. The memory 601 may mainly include a program storage area and a data storage area. Among them, the program storage area may store an operating system, programs required to run the instant messaging function, etc.; the data storage area may store various instant messaging information and operation instruction sets, etc.

[0164] The memory 601 may be a volatile memory, such as a random-access memory (RAM); the memory 601 may also be a non-volatile memory, such as a read-only memory, a flash memory, a hard disk drive (HDD), or a solid-state drive (SSD); or the memory 601 is any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 601 may be a combination of the above memories.

[0165] The processor 602 may include one or more central processing units (CPUs) or be a digital processing unit, etc. The processor 602 is used to implement the above-mentioned method for predicting the parameters of the ethylene production section when calling the computer program stored in the memory 601.

[0166] The communication module 603 is used to communicate with chemical industry dispatching equipment or other control systems.

[0167] In the embodiment of the present application, the specific connection medium between the above-mentioned memory 601, communication module 603, and processor 602 is not limited. In the embodiment of the present application Figure 6 it is shown that the memory 601 and the processor 602 are connected through a bus 604. The bus 604 is described in thick lines in Figure 6 The connection manners of other components are only for illustrative purposes and are not to be construed as limiting. The bus 604 may be divided into an address bus, a data bus, a control bus, etc. For ease of description, Figure 6It is only described by a thick line, but it does not describe that there is only one bus or one type of bus.

[0168] The computer storage medium is stored in the memory 601. The computer executable instructions are stored in the computer storage medium. The computer executable instructions are used to implement the ethylene production section parameter prediction method of the embodiments of the present application. The processor 602 is used to execute the ethylene production section parameter prediction methods of the above embodiments.

[0169] Based on the same inventive concept, the embodiments of the present application also provide a storage medium. The storage medium stores a computer program. When the computer program runs on a computer, the computer is caused to execute the steps in the ethylene production section parameter prediction method according to various exemplary embodiments of the present application described above in this specification.

[0170] In some possible implementation manners, each aspect of the ethylene production section parameter prediction method provided by the present application can also be implemented in the form of a computer program product, which includes a computer program. When the program product runs on a computer device, the computer program is used to cause the computer device to execute the steps in the ethylene production section parameter prediction method according to various exemplary embodiments of the present application described above in this specification. For example, the computer device can execute the steps of each embodiment.

[0171] The program product can adopt any combination of one or more readable media. The readable media can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0172] The program product of the embodiments of the present application can adopt a portable compact disk read-only memory (CD-ROM) and include a computer program, and can run on a computer device. However, the program product of the present application is not limited thereto. In the present application, the readable storage medium can be any tangible medium that contains or stores a program, and the computer program included therein can be used by or in combination with a command execution system, apparatus, or device.

[0173] A readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries a readable computer program. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable signal medium can also be any readable medium other than a readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with a command execution system, apparatus, or device.

[0174] The computer program contained on the readable medium can be transmitted with any appropriate medium, including but not limited to wireless, wired, optical fiber cable, RF, etc., or any suitable combination of the above.

[0175] The computer program for performing the operations of this application can be written in any combination of one or more programming languages, including object-oriented programming languages such as Java, C++, etc., and also including conventional procedural programming languages such as the "C" language or similar programming languages.

[0176] It should be noted that although several units or subunits of the apparatus are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of this application, the features and functions of the two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided and embodied by multiple units.

[0177] In addition, although the operations of the method of this application are described in a specific order in the drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, some steps can be omitted, multiple steps can be combined into one step for execution, and / or one step can be decomposed into multiple steps for execution.

[0178] Those skilled in the art should understand that the embodiments of this application can be provided as a method, system, or computer program product. Therefore, this application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0179] Although the preferred embodiments of this application have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of this application.

[0180] Obviously, those skilled in the art can make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalent technologies, this application is also intended to cover these changes and modifications.

Claims

1. A prediction method for an ethylene production section, characterized in that, The method includes: Obtaining real-time operation data of the ethylene production section and inputting the real-time operation data into a trained section prediction model; Obtaining a target prediction value of the ethylene production section output by the section prediction model; Wherein, the section prediction model is obtained by performing prediction processing on the historical operation data of the ethylene production section and the standardized historical operation data respectively to obtain the original prediction value and the standardized prediction value of the compressor power, and constructing a comprehensive loss function of the section prediction model based on the original prediction value and the standardized prediction value, and training the model based on the comprehensive loss function.

2. The method according to claim 1, characterized in that, The section prediction model is trained in the following manner: Obtaining the historical operation data of the ethylene production section, where the historical operation data includes the operation parameter sets of multiple compression sections in the ethylene production process and the true value of the compressor power; Standardizing the historical operation data, and performing prediction processing on the historical operation data and the standardized historical operation data to obtain the original prediction value and the standardized prediction value of the compressor power respectively; Obtaining a comprehensive loss function of the section prediction model based on the original error between the original prediction value and the true value, and the standardized error between the standardized prediction value and the true value; Based on the comprehensive loss function, adjusting the parameters of the section prediction model until the section prediction model meets the preset convergence condition to obtain a trained section prediction model.

3. The method according to claim 2, characterized in that, The obtaining of the comprehensive loss function of the section prediction model based on the original error between the original prediction value and the true value, and the standardized error between the standardized prediction value and the true value includes: Constructing a standardized loss of the section prediction model based on the standardized error; Constructing an inverse standardized loss of the section prediction model based on the original error; Performing weighted combination on the standardized loss and the inverse standardized loss to obtain the comprehensive loss function.

4. The method according to claim 2, wherein The standardizing of the historical operation data includes: Performing anomaly filtering on the historical operation data, and normalizing the historical operation data based on the mean and standard deviation of the filtered historical operation data; Performing data partitioning on the normalized historical operation data to obtain corresponding training sets and test sets.

5. The method according to claim 1, wherein The section prediction model is determined based on a preset Long Short-Term Memory (LSTM) model architecture, and the LSTM model architecture includes: An input layer for receiving standardized input data; An LSTM layer for extracting temporal features in the input data and transmitting the temporal features to a fully connected layer; The fully connected layer for performing a non-linear transformation on the temporal features and transmitting the obtained one-dimensional features to an output layer; The output layer for generating corresponding prediction values based on the one-dimensional features.

6. The method according to claim 5, wherein The LSTM model architecture further includes a random dropout layer for randomly dropping out some neurons.

7. The method according to claim 1, wherein The model parameters of the section prediction model at least include one or more combinations of the following parameters: The input feature dimension, which represents the number of features of the input data received by the model each time; The number of LSTM network layers; The hidden layer dimension, which represents the hidden state dimension of each LSTM layer; The sequence length, which represents the length of the input data received by the model each time in the time dimension; The batch size, which represents the number of samples to be processed in each model training; The random dropout probability, which represents the probability that the random dropout layer randomly drops neuron connections; The learning rate, which represents the step size for updating the model parameters; The number of training epochs, which represents the number of traversals of the training set; The batch dimension order, which corresponds to the format of the input data.

8. A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

9. A computer storage medium, on which computer program instructions are stored, wherein when the computer program instructions are executed by the processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product, comprising computer program instructions, wherein when the computer program instructions are executed by the processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images