Prediction method and system of oil product blending attribute and storage medium

Abstract

The invention provides an oil product blending attribute prediction method, an oil product blending attribute prediction system and a computer readable storage medium. The oil product blending attribute prediction method comprises the following steps: acquiring a data set comprising key attribute values and an actual blended oil formula in an oil product blending process, wherein the key attribute values comprise attribute values of nonlinear attributes; and inputting the data set into an oil product blending attribute prediction model to predict a blended oil attribute value, the oil product blending attribute prediction model being established based on an artificial neural network, the artificial neural network comprising an activation function. According to the prediction method for the oil product blending attributes, a model which does not depend on the physicochemical principle of the nonlinear attributes can be established, and the nonlinear attributes in oil product blending can be better predicted.

Description

Technical Field

[0001] The present invention relates to the field of oil blending, and in particular to a method for predicting oil blending properties, a system for predicting oil blending properties, and a computer-readable storage medium. Background Art

[0002] In the oil refining industry, oil blending is a key production link, which aims to produce diversified oil products that meet market demand, including gasoline, diesel, lubricating oil, etc., by reasonably matching different component oils and additives. Oil blending involves mixing multiple component oils with different properties and indicators in a certain proportion. The goal of oil blending is to ensure that the mixed finished oil can meet the established quality standards and performance requirements, while effectively reducing the cost of finished oil that meets the requirements.

[0003] At present, most oil refining companies use pipeline blending for their oil blending process. Pipeline blending requires the support of information technology such as blending optimization technology, and how to accurately predict the properties of finished oil is the core of the optimization system. Accurate prediction of finished oil properties helps to improve the first-time success rate of finished oil blending, reduce the excess of key quality indicators of finished oil, and effectively shorten the cycle of finished oil blending.

[0004] In the prior art, the soft measurement model of oil blending properties built according to the experimental method requires a large number of experiments to fit the data for the nonlinear properties in the oil blending process, and the parameters in the formula need to be adjusted according to the actual situation. For some strong nonlinear properties, there are a variety of different soft measurement models for oil blending properties to meet the needs of different working conditions. For example, for the strong nonlinear property viscosity, the Kendai-Morne model is suitable for mixtures with similar component properties and similar molecular weights, and the Arrhenius model is suitable for mixtures with high viscosity ratios. It performs well when calculating the viscosity of heavy oil (asphalt) and light oil mixtures. The Double Logarithmic model cannot be applied to non-Newtonian mixed oils and mixed oils with large differences in component oil viscosity index.

[0005] The existing mechanism-based modeling relies on a deep understanding of the physical and chemical principles in the blending process. There are also processes that are difficult to describe directly through the laws of physical and chemical methods. The modeling process is cumbersome and slow, and the model parameters need to be manually adjusted for different working conditions.

[0006] In order to overcome the above-mentioned defects of the prior art, there is an urgent need in the art for a prediction technology for oil blending properties, which can establish a model that does not rely on the physical and chemical principles of the nonlinear properties themselves, and better predict the nonlinear properties in oil blending. Summary of the invention

[0007] A brief summary of one or more aspects is given below to provide a basic understanding of these aspects. This summary is not an exhaustive overview of all conceived aspects, and is neither intended to identify the key or critical elements of all aspects nor to define the scope of any or all aspects. Its only purpose is to give some concepts of one or more aspects in a simplified form as a prelude to a more detailed description that will be given later.

[0008] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method for predicting oil blending properties, a system for predicting oil blending properties, and a computer-readable storage medium, which can establish a model that is independent of the physical and chemical principles of the nonlinear properties themselves, and better predict the nonlinear properties in oil blending.

[0009] Specifically, the above-mentioned method for predicting oil blending properties provided according to the first aspect of the present invention includes the steps of: obtaining a data set including key attribute values ​​in the oil blending process and an actual blending oil formula, wherein the key attribute values ​​include attribute values ​​of nonlinear attributes; and inputting the data set into an oil blending attribute prediction model to predict the blending oil attribute values, wherein the oil blending attribute prediction model is established based on an artificial neural network, and the artificial neural network includes an activation function.

[0010] Preferably, in one embodiment of the present invention, the artificial neural network is a residual network, and the residual network includes a convolutional layer, a pooling layer and a residual block.

[0011] Preferably, in an embodiment of the present invention, the artificial neural network is a multi-layer perceptron regression model.

[0012] Preferably, in one embodiment of the present invention, the training stage of the oil blending property prediction model includes the steps of: obtaining a sample set including key attribute values, actual blending oil formulas and blending oil attribute values, wherein the key attribute values ​​include attribute values ​​of nonlinear attributes; determining a hyperparameter combination of the oil blending property prediction model based on a grid search and cross-validation method; and training the oil blending property prediction model based on the activation function and the hyperparameter combination according to the sample set.

[0013] Preferably, in one embodiment of the present invention, the step of obtaining a sample set including key attribute values, actual blending oil formulas and blending oil attribute values ​​includes: preprocessing the sample set, the preprocessing includes eliminating interference data and incomplete data and removing abnormal data; and standardizing the sample set, the standardization method includes Min-Max standardization or Z-score standardization.

[0014] Preferably, in one embodiment of the present invention, the step of determining the hyperparameter combination of the oil blending property prediction model according to the grid search and cross-validation method includes: dividing the sample set into k subsets; taking subsets of the k subsets as validation subsets in turn, training the oil blending property prediction model with the remaining subsets according to the validation set hyperparameter combination corresponding to the validation subsets, and evaluating the trained oil blending property prediction model to form an evaluation index; and taking the average value of the k evaluation indexes as the final evaluation index to determine that the validation set hyperparameter combination with the highest accuracy in the corresponding validation subset is the hyperparameter combination of the oil blending property prediction model.

[0015] Preferably, in one embodiment of the present invention, the steps of taking subsets of the k subsets as validation subsets in turn, training the oil blending property prediction model with the remaining subsets according to the validation set hyperparameter combination corresponding to the validation subset, and evaluating the trained oil blending property prediction model to form an evaluation index include: determining the hyperparameters according to commonly used values ​​or adjusting the hyperparameters of the validation set in turn within a specified parameter range in steps; determining a plurality of validation set hyperparameter combinations by exhaustive search of the hyperparameters; and evaluating the oil blending property prediction model using the plurality of validation set hyperparameter combinations on the validation subset to determine that the validation set hyperparameter combination with the best evaluation result is the validation set hyperparameter combination corresponding to the validation subset.

[0016] Preferably, in an embodiment of the present invention, the training phase further comprises the step of evaluating the oil blending property prediction model according to the mean square error and the coefficient of determination.

[0017] Preferably, in one embodiment of the present invention, the updating stage of the oil blending property prediction model includes the steps of: in response to the amount of data of the acquired data set exceeding the set acquisition quantity, evaluating the oil blending property prediction model based on the data set and the actual blended oil property value corresponding to the data set; and in response to the evaluation result being an updated parameter, performing transfer learning on the oil blending property prediction model to update the internal parameters of the oil blending property prediction model online.

[0018] Preferably, in one embodiment of the present invention, in response to the amount of data of the acquired data set exceeding the set acquisition quantity, the step of evaluating the oil blending property prediction model based on the data set and the actual blended oil property value corresponding to the data set includes: evaluating the oil blending property prediction model according to the mean square error and the determination coefficient; and when the mean square error is higher than a first threshold or the determination coefficient is lower than a second threshold, determining the evaluation result as an updated parameter.

[0019] Preferably, in one embodiment of the present invention, the step of performing transfer learning on the oil blending property prediction model in response to updating parameters of the evaluation result, and online updating internal parameters of the oil blending property prediction model includes: freezing part of the structure of the oil blending property prediction model to retain the features learned by the oil blending property prediction model during the training phase.

[0020] In addition, the above-mentioned prediction system for oil blending properties provided according to the second aspect of the present invention includes a memory and a processor. The memory stores computer instructions. The processor is connected to the memory and is configured to execute the computer instructions stored in the memory to implement the prediction method for oil blending properties provided by any one of the above-mentioned embodiments.

[0021] In addition, the computer-readable storage medium provided according to the third aspect of the present invention stores computer instructions, which, when executed by a processor, implement the method for predicting the blending properties of oil products provided in any one of the above embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The above features and advantages of the present invention can be better understood after reading the detailed description of the embodiments of the present disclosure in conjunction with the following drawings. In the drawings, the components are not necessarily drawn to scale, and components with similar related properties or features may have the same or similar reference numerals.

[0023] Figure 1 A schematic diagram of a prediction system for oil blending properties provided according to some embodiments of the present invention is shown;

[0024] Figure 2 A flow chart showing a method for predicting oil blending properties according to some embodiments of the present invention is shown;

[0025] Figure 3 A network structure diagram of an oil blending property prediction model provided according to a preferred embodiment of the present invention is shown;

[0026] Figure 4 A graph showing changes in the mean square error of the oil blending property prediction model during the training phase according to a preferred embodiment of the present invention is shown;

[0027] Figure 5 A comparison diagram between the predicted value and the true value of the oil blending property prediction model provided by a preferred embodiment of the present invention is shown;

[0028] Figure 6 A comparison diagram of the predicted value and the true value of the oil blending property prediction model provided by a preferred embodiment of the present invention under another working condition is shown; and

[0029] Figure 7 A comparison chart between the predicted value and the true value of the oil blending property prediction model after parameter update provided by a preferred embodiment of the present invention under another working condition is shown.

[0030] Reference numerals:

[0031] 100: Prediction system for oil blending properties;

[0032] 110: memory;

[0033] 111: Computer readable storage medium;

[0034] 120: Processor;

[0035] 200: Prediction methods for oil blending properties; and

[0036] S210~S220: steps. DETAILED DESCRIPTION

[0037] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. Note that the aspects described below in conjunction with the accompanying drawings and specific embodiments are only exemplary and should not be construed as limiting the scope of protection of the present invention in any way.

[0038] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0039] In addition, the terms "upper", "lower", "left", "right", "top", "bottom", "horizontal" and "vertical" used in the following description should be understood as the directions shown in the paragraph and the related drawings. Such relative terms are only used for the convenience of description and do not mean that the device described therein must be manufactured or operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0040] It is understood that although the terms "first", "second", "third", etc. may be used herein to describe various components, regions, layers and / or parts, these components, regions, layers and / or parts should not be limited by these terms, and these terms are only used to distinguish different components, regions, layers and / or parts. Therefore, the first component, region, layer and / or part discussed below may be referred to as a second component, region, layer and / or part without departing from some embodiments of the present invention.

[0041] As mentioned above, in the prior art, the soft measurement model of oil blending properties built according to the experimental method requires a large number of experiments to fit the data for the nonlinear properties in the oil blending process, and the parameters in the formula need to be adjusted according to the actual situation. For some strong nonlinear properties, there are a variety of different soft measurement models for oil blending properties to meet the needs of different working conditions. For example, for the strong nonlinear property viscosity, the Kendai-Morne model is suitable for mixtures with similar component properties and similar molecular weights, and the Arrhenius model is suitable for mixtures with high viscosity ratios. It performs better when calculating the viscosity of heavy oil (asphalt) and light oil mixtures. The Double Logarithmic model cannot be applied to non-Newtonian mixed oils and mixed oils with large differences in component oil viscosity index.

[0042] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method for predicting oil blending properties, a system for predicting oil blending properties, and a computer-readable storage medium, which can better predict the nonlinear properties in oil blending and realize self-correction of the model under a variety of complex working conditions.

[0043] In some non-limiting embodiments, the above-mentioned method for predicting the blending properties of oil products provided in the first aspect of the present invention may be implemented via the above-mentioned system for predicting the blending properties of oil products provided in the second aspect of the present invention.

[0044] Please refer to Figure 1 , Figure 1 A schematic diagram of a system for predicting oil blending properties provided according to some embodiments of the present invention is shown.

[0045] like Figure 1 As shown, the prediction system 100 for oil blending properties may be configured with a memory 110 and a processor 120. The memory 110 may include but is not limited to the computer-readable storage medium 111 provided in the third aspect of the present invention, on which computer instructions are stored. The processor 120 is connected to the memory 110 and is configured to execute the computer instructions stored in the memory 110 to implement the prediction method for oil blending properties provided in the first aspect of the present invention.

[0046] The following will first describe the working principle of the above-mentioned prediction system 100 for oil blending properties in conjunction with some examples of prediction methods for oil blending properties. Those skilled in the art will understand that these examples of prediction methods for oil blending properties are only some non-restrictive implementation methods provided by the present invention, which are intended to clearly demonstrate the main concept of the present invention and provide some specific solutions that are convenient for the public to implement, rather than to limit all functions or all working methods of the prediction system 100 for oil blending properties. Similarly, the prediction system 100 for oil blending properties is also only a non-restrictive implementation method provided by the present invention, and does not constitute a limitation on the execution subject and execution order of each step in these prediction methods for oil blending properties.

[0047] Please refer to Figure 2 , Figure 2 A flow chart of a method for predicting oil blending properties provided according to some embodiments of the present invention is shown.

[0048] like Figure 2 As shown, the method 200 for predicting oil blending properties includes step S210: obtaining a data set including key property values ​​in the oil blending process and an actual blending oil formula, wherein the key property values ​​include property values ​​of nonlinear properties.

[0049] The prediction system 100 for oil blending properties first acquires a data set including key attribute values ​​in the oil blending process and actual blending oil formulas, and the data set may also include other feature values ​​that may affect the key attribute values. Here, the key attribute values ​​include attribute values ​​of nonlinear attributes, especially strong nonlinear attributes. For example, the prediction system 100 for oil blending properties may select octane number as a key attribute, and the collected data set may also include feature points that affect octane number, such as reformed gasoline octane number, MTBE octane number, and non-aromatic gasoline octane number.

[0050] The oil blending property prediction method 200 also includes step S220: inputting the data set into the oil blending property prediction model to predict the blended oil property value, the oil blending property prediction model is established based on an artificial neural network, and the artificial neural network includes an activation function.

[0051] Here, the oil blending property prediction model established based on the artificial neural network can be established by a residual network or by a multi-layer perceptron regression model. When the oil blending property prediction model is established by a residual network, the residual network may include a convolution layer, a pooling layer, and a residual block. The size and network structure of the convolution layer, the pooling layer, and the residual block included in the residual network can be adjusted according to actual conditions. When the oil blending property prediction model is constructed by a multi-layer perceptron regression model, the multi-layer perceptron regression model may include multiple hidden layers. The number of hidden layers and the number of neurons in each hidden layer can be determined according to actual conditions.

[0052] Furthermore, the artificial neural network for establishing the oil blending property prediction model may also include an activation function, which is used to increase the ability of the oil blending property prediction model to learn nonlinear relationships. Here, the activation function may use a rectified linear unit (ReLU) and / or a sigmoid activation function.

[0053] The working principle of the above-mentioned prediction system 100 for oil blending properties will be described below in conjunction with some examples of the online training stage. Those skilled in the art will understand that these examples of the online training stage are only some non-limiting implementation methods provided by the present invention, which are intended to clearly demonstrate the main concept of the present invention and provide some specific solutions that are convenient for the public to implement, rather than to limit all functions or all working methods of the prediction system for oil blending properties. Similarly, the prediction system for oil blending properties is also only a non-limiting implementation method provided by the present invention, and does not constitute a limitation on the execution subject and execution order of each step in these online training stages.

[0054] The online training phase of the prediction system 100 for oil blending properties may include the steps of: obtaining a sample set including key attribute values, actual blending oil formulas, and blending oil attribute values, wherein the key attribute values ​​include attribute values ​​of nonlinear attributes. The sample set obtained by the prediction system 100 for oil blending properties may be historical data of the gasoline blending process, and each sample may include the key attribute value of each component oil, the actual formula at the corresponding moment, the blending oil attribute value of the key attribute at the corresponding moment, and features that may affect the key attribute based on experience.

[0055] Preferably, the system 100 for predicting oil blending properties can perform preprocessing and standardization on the acquired sample set.

[0056] Preprocessing includes eliminating interference data and incomplete data and removing abnormal data. The prediction system 100 for oil blending properties can delete the entire sample data of incomplete data and interference data with large fluctuations contained in the original data. For the key attribute values ​​of the component oils collected in the same time period, the abnormal data can be removed according to the Laida criterion.

[0057] The standardization processing method includes Min-Max standardization or Z-score standardization. The prediction system 100 of the oil blending property can respectively use Min-Max standardization and Z-score standardization to process the preprocessed sample set, and can select the Min-Max standardization method or the Z-score standardization method according to the data distribution of the sample set and the evaluation result of the oil blending property prediction model.

[0058] Afterwards, the oil blending property prediction system 100 can determine the hyperparameter combination of the oil blending property prediction model according to the grid search and cross validation method. Here, the hyperparameter combination can include the learning rate α of the oil blending property prediction model.

[0059] Specifically, the prediction system 100 for oil blending properties may first divide the sample set into multiple subsets, for example, k subsets. Then, subsets in the k subsets are used as validation subsets in turn, and subsets in the k subsets are used as validation subsets in turn. According to the validation set hyperparameter combination corresponding to the validation subsets, the remaining subsets are used to train the oil blending property prediction model and the trained oil blending property prediction model is evaluated to form an evaluation index.

[0060] For example, the prediction system 100 for oil blending properties first uses a subset k1 among the k subsets as a validation subset, and the remaining subsets among the k subsets except the validation subset k1 as training sets. Based on the validation set hyperparameter combination corresponding to the validation subset k1, the remaining subsets are used to train the oil blending property prediction model, and the trained oil blending property prediction model is evaluated to form an evaluation index. Then, the prediction system 100 for oil blending properties uses the next subset k2 as a validation subset, and the remaining subsets among the k subsets except the validation subset k2 as training sets. Based on the validation set hyperparameter combination corresponding to the validation subset k2, the oil blending property prediction model is trained and evaluated. By analogy, the prediction system 100 for oil blending properties sequentially uses the k subsets as validation subsets, and based on the validation set hyperparameter combination corresponding to the validation subsets, the remaining subsets are used to train and evaluate the oil blending property prediction model.

[0061] Afterwards, the prediction system 100 for oil blending properties can use the average value of the k evaluation indicators as the final evaluation indicator. Based on the final evaluation indicator, the validation set hyperparameter combination with the highest accuracy on the corresponding validation subset is found from all validation set hyperparameter combinations as the optimal hyperparameter combination for the oil blending property prediction model.

[0062] Furthermore, when determining the validation set hyperparameter combination corresponding to the validation subset, the prediction system 100 for oil blending properties can form a grid with the commonly used values ​​of each hyperparameter. If some hyperparameters do not have commonly used values, they can be uniformly valued at an appropriate step size within the specified hyperparameter range. Then, exhaustively search all possible hyperparameter combinations as validation set hyperparameter combinations. Afterwards, each validation set hyperparameter combination is evaluated on the validation subset for the prediction model of oil blending properties, thereby selecting the validation set hyperparameter combination with the best performance of the oil blending property prediction model under the validation subset as the validation set hyperparameter combination corresponding to the validation subset.

[0063] Then, the prediction system 100 for oil blending properties can train the oil blending property prediction model based on the sample set, the activation function and the determined hyperparameter combination. The sample set after preprocessing and standardization can be divided into a training set and a test set, and the key attribute value of each component oil in each sample, the actual formula at the corresponding time, and the features that may have an impact on the key attribute based on experience are used as input data, and the blended oil attribute value after blending of the key attribute at the corresponding time is used as output data to train the established oil blending property prediction model.

[0064] In addition, the prediction system 100 for oil blending properties can evaluate the oil blending property prediction model according to the mean square error and the determination coefficient. The smaller the mean square error of the oil blending property prediction model, the better the performance of the oil blending property prediction model; the closer the determination coefficient is to 1, the better the performance of the oil blending property prediction model.

[0065] Mean-Square Error (MSE) is an indicator used to measure the prediction error of a regression model. The mean-square error can calculate the average of the squared differences between the predicted values ​​of the oil blending property prediction model and the actual observed values. The specific formula is as follows:

[0066]

[0067] Among them, y i is the actual observed value in the sample, y iPre is the predicted value determined by the oil blending property prediction model, and N is the number of samples.

[0068] Coefficient of determination R 2It is an indicator used to measure the regression model fitting data. The coefficient of determination R 2 The value is between 0 and 1. The specific formula is as follows:

[0069]

[0070] Among them, y i is the actual observed value in the sample, is the mean of the actual observed values, y iPre is the predicted value determined by the oil blending property prediction model, and N is the number of samples.

[0071] Therefore, the prediction system 100 for oil blending properties can utilize a large amount of existing historical data and quickly establish a prediction model for nonlinear properties in the oil blending process based on an artificial neural network structure.

[0072] In addition, the oil blending property prediction model in the oil blending property prediction system 100 also includes an updating stage.

[0073] When the amount of data in the acquired data set exceeds the set acquisition quantity, the prediction system 100 for oil blending properties can evaluate the prediction model for oil blending properties based on the data set and the actual blended oil property value corresponding to the data set. Specifically, the prediction system 100 for oil blending properties can evaluate the prediction model for oil blending properties based on the mean square error and the determination coefficient, and when the mean square error is higher than the first threshold or the determination coefficient is lower than the second threshold, the evaluation result is determined to be the update parameter. The first threshold and the second threshold can be set by those skilled in the art according to actual conditions.

[0074] When the mean square error is lower than the first threshold or the determination coefficient is higher than the second threshold, it means that the parameters of the oil blending property prediction model do not need to be updated, and the oil blending property prediction model can well predict the new operating conditions.

[0075] When the evaluation result is to update the parameters, the prediction system 100 for oil blending properties can perform transfer learning on the oil blending property prediction model based on the data set and the actual blending oil property values ​​corresponding to the data set, and update the internal parameters of the oil blending property prediction model online. Before performing transfer learning, the prediction system 100 for oil blending properties can also perform data preprocessing and data standardization on the collected data set and the actual blending oil property values ​​corresponding to the data set.

[0076] Therefore, the oil blending property prediction system 100 can reduce the retraining cost, quickly train to obtain new parameters of the oil blending property prediction model, reduce the deviation of the oil blending property prediction model under the current working conditions, and enable the oil blending property prediction model to better adapt to different working conditions.

[0077] Preferably, the prediction system 100 for oil blending properties can adopt fine-tuning as a transfer learning strategy, that is, freezing part of the structure of the oil blending property prediction model when updating internal parameters to retain the features learned by the oil blending property prediction model during the training phase. Here, the specific number of frozen layers or structures can be appropriately adjusted according to the actual situation. Thus, the prediction system 100 for oil blending properties can prevent these frozen layers from updating parameters such as weights during the update process, and train the oil blending property prediction model through the data set and the real blending oil property values ​​corresponding to the data set, and use the oil blending property prediction model after the updated parameters as a new oil property prediction model for the prediction of blending oil property values.

[0078] The following is a more specific non-limiting preferred embodiment, based on which the training stage of the oil blending property prediction model proposed in the present invention is further explained.

[0079] The prediction system 100 for oil blending properties first obtains historical data of a gasoline blending process of a large-scale refining enterprise as a sample set, and the key attribute is the nonlinear attribute of octane number. The sample set may include 2000 feature points such as reformed gasoline octane number, MTBE (methyl tertiary butyl ether) octane number, non-aromatic gasoline octane number, etc. in certain time periods, and the prediction system 100 for oil blending properties may use these 2000 sample data as a training set in the training phase. In addition, the sample set may also include 1000 feature points such as octane number of different component oils in another time period as another working condition environment.

[0080] The prediction system 100 for oil blending properties can preprocess and standardize the 3,000 sample data included in the sample set. By preprocessing the sample data, 47 incomplete sample data are deleted, and 18 abnormal data that do not meet the Laida criteria are eliminated. After that, the characteristic values ​​of the sample data in the sample set are standardized by the Z-score standardization method.

[0081] Afterwards, in this preferred embodiment, the oil blending property prediction system 100 uses a residual network to establish an oil blending property prediction model, and adjusts the relevant sizes of the convolutional layer, pooling layer and residual block of the residual network according to actual conditions, uses ReLU as the activation function of the oil blending property prediction model, and confirms that the initial learning rate in the hyperparameter combination of the oil blending property prediction model is 0.02 through grid search and cross-validation methods.

[0082] Please refer to Figure 3 , Figure 3 A network structure diagram of an oil blending property prediction model provided according to a preferred embodiment of the present invention is shown.

[0083] The network structure of the oil blending property prediction model constructed by the oil blending property prediction system 100 based on the residual network can be as follows: Figure 3 As shown, the oil blending property prediction model can include an input layer Input, a convolution layer C1, a pooling layer S1, a residual block R1-1, a residual block R1-2, a residual block R2-1, a residual block R2-2, a residual block R3-1, a residual block R3-2, a residual block R4-1, a residual block R4-2, a pooling layer S2, a fully connected layer F1 and an output layer Output in sequence. The prediction system 100 for oil blending properties inputs the processed sample set or data set into the input layer Input; the convolution kernel size of the convolution layer C1 can be 7, the step size is 2, and the padding is 3; the pooling layer S1 can be a maximum pooling layer with a size of 3 and a step size of 2; the residual block R1-1 can include 2 convolution layers with a convolution kernel size of 3 and 64 channels; the residual block R1-2 can include 2 convolution layers with a convolution kernel size of 3, one of which has a step size of 2, and the residual block R1-2 has 64 channels; the residual block R2-1 can include 2 convolution layers with a convolution kernel size of 3 and 128 channels; the residual block R2-2 can include 2 convolution layers with a convolution kernel size of 3, one of which has a step size of 2, and the residual block The channel of R2-2 is 128; the residual block R3-1 can include 2 convolution layers with a convolution kernel size of 3 and a channel of 256; the residual block R3-2 can include 2 convolution layers with a convolution kernel size of 3, one of which has a stride of 2, and the channel of the residual block R3-2 is 256; the residual block R4-1 can include 2 convolution layers with a convolution kernel size of 3 and a channel of 512; the residual block R4-2 can include 2 convolution layers with a convolution kernel size of 3, one of which has a stride of 2, and the channel of the residual block R4-2 is 512; the pooling layer S2 can be an average pooling layer with a size of 7; the fully connected layer F1 can be used to expand the vector of the pooling layer S2 into a vector; the relevant data is output through the output layer Output.

[0084] Afterwards, the prediction system 100 for oil blending properties can use the first 2000 sample data in the sample set as a training set to train the system. Figure 3 The oil blending property prediction model shown.

[0085] Please refer to Figure 4 , Figure 5 and Figure 6 , Figure 4 The figure shows the change of the mean square error of the oil blending property prediction model provided by a preferred embodiment of the present invention during the training phase. Figure 5 A comparison diagram between the predicted value and the true value of the oil blending property prediction model provided according to a preferred embodiment of the present invention is shown. Figure 6A comparison chart between the predicted value and the actual value of the oil blending property prediction model provided according to a preferred embodiment of the present invention under another working condition is shown.

[0086] The change of mean square error MSE in each training round during training is as follows: Figure 4 As shown, as the number of training rounds increases, the mean square error MSE decreases; according to the parameters updated during training, the mean square error MSE of each round fluctuates based on the change of parameters. The prediction system 100 for oil blending properties can use the parameters of the oil blending property prediction model with the smallest MSE during the training process as the parameters of the optimal oil blending property prediction model.

[0087] Afterwards, 100 sample data were randomly selected from the 2000 sample data in the training set as the validation set. Based on the optimal oil blending property prediction model, the performance of the validation set was evaluated to obtain the determination coefficient R of the optimal oil blending property prediction model. 2 Here, the comparison chart of the octane number prediction value and the true value of the optimal oil blending property prediction model is as follows: Figure 5 In addition, the prediction system 100 for oil blending properties can randomly select 100 sample data from 1000 sample data in another working condition as a validation set, and perform a performance evaluation on the validation set based on the optimal oil blending property prediction model to obtain the determination coefficient R of the optimal oil blending property prediction model in another new working condition. 2 Here, the comparison between the predicted value and the actual value of the octane number of the optimal oil blending property prediction model under another new working condition is shown in the figure below: Figure 6 shown.

[0088] According to the change of the determination coefficient, it can be known that under another new working condition, the prediction performance of the oil blending property prediction model decreases. The prediction performance can be improved by updating the oil blending property prediction model online. The prediction system 100 for oil blending properties can update the internal parameters of the oil blending property prediction model according to transfer learning. For example, the fine-tuning strategy of transfer learning can be used to first update the oil blending property prediction model. Figure 3 The convolution layer C1 in the network structure of the oil blending property prediction model shown is frozen, and then 1000 sample data under another working condition are used to update the weight parameters of other layers such as the fully connected layer F1, thereby retaining the features learned by the oil blending property prediction model from the 2000 training sets of the previous working condition.

[0089] Please refer to Figure 7 , Figure 7 A comparison chart between the predicted value and the true value of the oil blending property prediction model after parameter update provided by a preferred embodiment of the present invention under another working condition is shown.

[0090] The oil blending property prediction model is updated through transfer learning, and the performance of the updated oil blending property prediction model is evaluated under the new working conditions. The determination coefficient R 2 Here, the comparison between the predicted value and the true value of the oil blending property prediction model after updating the parameters under another working condition is shown in the figure below. Figure 7 As shown, through Figure 7 It can be found that online updating of the oil blending property prediction model under new operating conditions can effectively improve the performance of the oil blending property prediction model, realize self-correction of the oil blending property prediction model, and continuously optimize the oil blending property prediction model.

[0091] Therefore, in response to new operating conditions, the oil blending property prediction system 100 provided by the present invention can update the model weights online through the fine-tuning strategy of transfer learning, thereby effectively reducing the cost of retraining the oil blending property prediction model and improving the efficiency of retraining. The updated oil blending property prediction model can have better performance for the new operating conditions and better adapt to the new operating conditions.

[0092] In summary, the present invention provides a method for predicting oil blending properties, which can quickly and efficiently establish an oil blending property prediction model that can predict the key property values ​​of blending oil in the oil blending process without relying on the physical and chemical principles of the nonlinear properties themselves, and increase the ability of the oil blending property prediction model to learn nonlinear relationships through activation functions, thereby better realizing the prediction of nonlinear properties in oil blending.

[0093] Although the above methods are illustrated and described as a series of actions for simplicity of explanation, it should be understood and appreciated that these methods are not limited by the order of the actions, because according to one or more embodiments, some actions may occur in a different order and / or concurrently with other actions from those illustrated and described herein or not illustrated and described herein but understandable to those skilled in the art.

[0094] Those skilled in the art will appreciate that information, signals, and data may be represented using any of a variety of different technologies and techniques. For example, data, instructions, commands, information, signals, bits, symbols, and chips cited throughout the above description may be represented by voltage, current, electromagnetic waves, magnetic fields or magnetic particles, optical fields or optical particles, or any combination thereof.

[0095] Those skilled in the art will further appreciate that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or a combination of the two. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps are generally described above in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. The technician may implement the described functionality in different ways for each specific application, but such implementation decisions should not be interpreted as resulting in a departure from the scope of the present invention.

[0096] The various illustrative logic modules and circuits described in conjunction with the embodiments disclosed herein may be implemented or executed with a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in cooperation with a DSP core, or any other such configuration.

[0097] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. The software module may reside in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor so that the processor can read and write information from / to the storage medium. In an alternative, a storage medium may be integrated into a processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and the storage medium may reside in a user terminal as discrete components.

[0098] In one or more exemplary embodiments, the functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented as a computer program product in software, each function may be stored on or transmitted by a computer-readable medium as one or more instructions or codes. Computer-readable media include both computer storage media and communication media, including any medium that facilitates the transfer of a computer program from one place to another. Storage media may be any available medium that can be accessed by a computer. As an example and not limitation, such a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, disk storage or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of an instruction or data structure and can be accessed by a computer. Any connection is also properly referred to as a computer-readable medium. For example, if the software is transmitted from a website, a server, or other remote source using a coaxial cable, a fiber optic cable, a twisted pair, a digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwaves, the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwaves are included in the definition of the medium. Disk and disc as used herein include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray disc, wherein disk often reproduces data magnetically, while disc reproduces data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

[0099] The previous description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the disclosure is not intended to be limited to the examples and designs described herein, but should be granted the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting oil blending properties, characterized in that: Includes steps: Acquire a data set including key attribute values ​​in the oil blending process and an actual blended oil formula, wherein the key attribute values ​​include attribute values ​​of nonlinear attributes; as well as The data set is input into an oil blending property prediction model to predict the blended oil property value. The oil blending property prediction model is established based on an artificial neural network, and the artificial neural network includes an activation function.

2. The method for predicting oil blending properties according to claim 1, characterized in that: The artificial neural network is a residual network, which includes a convolutional layer, a pooling layer and a residual block.

3. The method for predicting oil blending properties according to claim 1, characterized in that: The artificial neural network is a multi-layer perceptron regression model.

4. The method for predicting oil blending properties according to any one of claims 2 or 3, characterized in that: The training phase of the oil blending property prediction model includes the following steps: Acquire a sample set including key attribute values, actual blended oil formulas, and blended oil attribute values, wherein the key attribute values ​​include attribute values ​​of nonlinear attributes; Determine the hyperparameter combination of the oil blending property prediction model according to the grid search and cross validation method; as well as According to the sample set, the oil blending property prediction model is trained based on the activation function and the hyperparameter combination.

5. The method for predicting oil blending properties according to claim 4, characterized in that: The step of obtaining a sample set including key attribute values, actual blended oil formula and blended oil attribute values ​​comprises: Preprocessing the sample set, wherein the preprocessing includes eliminating interfering data and incomplete data and removing abnormal data; and The sample set is standardized, and the standardization method includes Min-Max standardization or Z-score standardization.

6. The method for predicting oil blending properties according to claim 4, characterized in that: The step of determining the hyperparameter combination of the oil blending property prediction model according to the grid search and cross validation method includes: Dividing the sample set into k subsets; Sequentially taking subsets of the k subsets as validation subsets, and according to the validation set hyperparameter combination corresponding to the validation subsets, using the remaining subsets to train the oil blending property prediction model and evaluating the trained oil blending property prediction model to form an evaluation index; and The average value of the k evaluation indicators is used as the final evaluation indicator to determine the validation set hyperparameter combination with the highest accuracy in the corresponding validation subset as the hyperparameter combination of the oil blending property prediction model.

7. The method for predicting oil blending properties according to claim 6, characterized in that: The steps of sequentially using subsets of the k subsets as validation subsets, training the oil blending property prediction model with the remaining subsets according to the validation set hyperparameter combination corresponding to the validation subsets, and evaluating the trained oil blending property prediction model to form an evaluation index include: Determine the hyperparameters according to commonly used values ​​or adjust the hyperparameters in steps within a specified parameter range; Determine a plurality of validation set hyperparameter combinations by exhaustively searching the hyperparameters; and The multiple validation set hyperparameter combinations are used to evaluate the oil blending property prediction model on the validation subset to determine that the validation set hyperparameter combination with the best evaluation result is the validation set hyperparameter combination corresponding to the validation subset.

8. The method for predicting oil blending properties according to claim 4, characterized in that: The training phase also includes the steps of: The oil blending property prediction model was evaluated based on mean square error and determination coefficient.

9. The method for predicting oil blending properties according to any one of claims 2 or 3, characterized in that: The updating stage of the oil blending property prediction model includes the following steps: In response to the amount of data of the acquired data set exceeding a set acquisition quantity, the oil blending property prediction model is evaluated based on the data set and the actual blending oil property value corresponding to the data set; as well as In response to the evaluation result being an updated parameter, transfer learning is performed on the oil blending property prediction model to update internal parameters of the oil blending property prediction model online.

10. The method for predicting oil blending properties according to claim 9, characterized in that: In response to the amount of data of the acquired data set exceeding the set acquisition quantity, the step of evaluating the oil blending property prediction model based on the data set and the actual blending oil property value corresponding to the data set includes: Evaluating the oil blending property prediction model based on mean square error and coefficient of determination; and When the mean square error is higher than a first threshold or the determination coefficient is lower than a second threshold, the evaluation result is determined as an updated parameter.

11. The method for predicting oil blending properties according to claim 9, characterized in that: The step of performing transfer learning on the oil blending property prediction model to update the internal parameters of the oil blending property prediction model online in response to the evaluation result is as follows: Part of the structure of the oil product blending property prediction model is frozen to retain the features learned by the oil product blending property prediction model during the training phase.

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

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

Images