Data analysis methods, apparatus, media and equipment for feedstock hydrotreating units

Abstract

The present disclosure relates to a data analysis method, device, medium and equipment for a crude oil hydrogenation unit, the method comprising: obtaining target condition data, the target condition data comprising crude oil composition, operating time of the hydrogenation unit and process parameters adopted by the hydrogenation unit during the operating time; processing multiple data in the target condition data through a target neural network to predict the physical property data of the product generated by the hydrogenation unit under the conditions of the crude oil composition, the operating time and the process parameters, and outputting the predicted physical property data. This solution can predict the physical property data of the product in advance before the product is actually generated, so as to find out in advance the products that do not meet the production requirements, or can also provide feedback to the input condition data based on the predicted physical property data, thereby guiding the optimization of the hydrogenation process.

Description

Technical Field

[0001] The present disclosure relates to the technical field of crude oil hydrogenation, and in particular, to a data analysis method, device, medium and equipment for a crude oil hydrogenation device. Background Art

[0002] At present, my country's economy maintains rapid development, and the concept of sustainable development of green economy has gradually gained attention. Therefore, the restrictions on sulfur content in crude oil are becoming more and more stringent. For example, according to relevant standards, the sulfur content in diesel must be limited to below 10μg / g. Hydrogenation refining of crude oil using a hydrogenation unit is currently the main desulfurization method. In the industrial production process, as the operation time of the hydrogenation unit increases, the activity of the hydrorefining catalyst will be lost. The change in catalyst activity directly affects whether the quality of the product meets the production requirements.

[0003] In the related art, after the hydrofining is completed, it is checked whether the sulfur content of the hydrofining product meets the requirements. If it does not meet the requirements, desulfurization is performed again. This method requires the inspection of the generated product after the hydrofining is completed, which is very inconvenient in actual application. Summary of the invention

[0004] The purpose of the present disclosure is to provide a data analysis method, device, medium and equipment for a crude oil hydrogenation unit to solve the above-mentioned technical problems.

[0005] In order to achieve the above-mentioned object, the present disclosure provides, in a first aspect, a data analysis method for a crude oil hydrogenation unit, comprising:

[0006] Acquiring target condition data, the target condition data including the composition of the feedstock oil, the operating time of the hydrogenation unit, and the process parameters adopted by the hydrogenation unit during the operating time;

[0007] Through the target neural network, multiple data in the target condition data are processed to predict the physical property data of the product generated by the hydrogenation device under the conditions of the raw oil composition, the operating time and the process parameters, and the predicted physical property data is output.

[0008] Optionally, the method further comprises:

[0009] Acquire multiple sets of training data, each set of training data includes historical condition data and real physical property data corresponding to the historical condition data;

[0010] Determine input layer nodes, hidden layer nodes and output layer nodes to obtain a first neural network;

[0011] Using a genetic algorithm to optimize the network parameters of the first neural network to obtain target network parameters, and obtaining a second neural network based on the target network parameters;

[0012] Training the second neural network according to the multiple sets of training data;

[0013] After the training is completed, the trained second neural network is determined as the target neural network.

[0014] Optionally, determining the input layer nodes, the hidden layer nodes, and the output layer nodes to obtain the first neural network includes:

[0015] Determine the number of input layer nodes and the number of output layer nodes;

[0016] According to the number of nodes in the input layer and the number of nodes in the output layer, a search range of the number of nodes in the hidden layer is estimated, and values ​​are sequentially taken from the search range;

[0017] According to the number of input layer nodes, the number of output layer nodes and the number of hidden layer nodes taken out this time, the input layer nodes, the hidden layer nodes and the output layer nodes are determined to obtain the candidate neural network this time;

[0018] Inputting multiple sets of historical condition data into the candidate neural network in sequence, obtaining predicted physical property data corresponding to the input historical condition data, and calculating the error value of the candidate neural network according to the predicted physical property data and the corresponding real physical property data;

[0019] The candidate neural network with the smallest error value is determined from multiple candidate neural networks as the first neural network.

[0020] Optionally, the optimizing the network parameters of the first neural network by using a genetic algorithm to obtain target network parameters includes:

[0021] According to the set population size k, k groups of network parameters are randomly generated; k is a positive integer greater than 1;

[0022] For each set of network parameters, determining an error value of a first neural network applying the set of network parameters, and using the error value as the fitness corresponding to the set of network parameters;

[0023] According to the fitness corresponding to each group of network parameters, selection, crossover and mutation are performed in the current k groups of network parameters to form new k groups of network parameters, and returning to execute the step of determining the error value of the first neural network applying the group of network parameters for each group of network parameters;

[0024] When the termination condition is reached, a set of optimal network parameters is determined from the current k sets of network parameters as the target network parameters.

[0025] Optionally, the training the second neural network according to the multiple groups of historical data includes:

[0026] The second neural network is subjected to the following iterative process multiple times according to the multiple groups of historical data:

[0027] In this iteration, multiple groups of historical condition data corresponding to this iteration are sequentially input into the second neural network to obtain predicted physical property data corresponding to the input historical condition data;

[0028] Calculating an error value based on the predicted physical property data and the corresponding actual physical property data;

[0029] According to the relationship between the error value of this iteration and the error value of the previous iteration, the learning rate of the previous iteration is adjusted accordingly to obtain the learning rate of this iteration;

[0030] The network parameters in the second neural network are updated according to the learning rate of this iteration.

[0031] Optionally, the learning rate of the previous iteration is adjusted accordingly based on the relationship between the error value of the current iteration and the error value of the previous iteration to obtain the learning rate of the current iteration, including:

[0032] If the error value of this iteration is less than the error value of the previous iteration, the learning rate is increased based on the learning rate of the previous iteration to obtain the learning rate of this iteration;

[0033] If the error value of this iteration is greater than or equal to the error value of the previous iteration, the learning rate is reduced based on the learning rate of the previous iteration to obtain the learning rate of this iteration.

[0034] Optionally, after acquiring the target condition data, the method further includes:

[0035] Each item of the target condition data is normalized respectively, and the normalized data is used as the input of the target neural network.

[0036] A second aspect of the present disclosure provides a data analysis device for a crude oil hydrogenation device, comprising:

[0037] A condition data acquisition module, used to acquire target condition data, wherein the target condition data includes the composition of the feedstock oil, the operating time of the hydrogenation unit, and the process parameters adopted by the hydrogenation unit during the operating time;

[0038] The physical property data prediction module is used to process multiple data in the target condition data through a target neural network to predict the physical property data of the product generated by the hydrogenation unit under the conditions of the raw oil composition, the operating time and the process parameters, and output the predicted physical property data.

[0039] A third aspect of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0040] A fourth aspect of the present disclosure provides an electronic device, including:

[0041] a memory having a computer program stored thereon;

[0042] A processor is used to execute the computer program in the memory to implement the steps of the method described in the first aspect.

[0043] In the above scheme, a target neural network is pre-constructed. Based on the constructed target neural network, the physical property data of the product generated by the hydrogenation device under the corresponding conditions can be predicted by simply inputting the corresponding condition data into the target neural network. Therefore, based on this scheme, physical property data such as product sulfur content, product density and product distillation range can be predicted in advance before the actual product is generated, so as to find products that do not meet the production requirements in advance, or the input condition data can be fed back according to the predicted physical property data, thereby guiding the optimization of the hydrogenation process.

[0044] Other features and advantages of the present disclosure will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings are used to provide a further understanding of the present disclosure and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present disclosure but do not constitute a limitation of the present disclosure. In the accompanying drawings:

[0046] Figure 1 A schematic diagram of a diesel hydrogenation process provided by an exemplary embodiment is shown;

[0047] Figure 2 A flow chart showing a data analysis method for a feedstock oil hydrogenation unit provided by an exemplary embodiment;

[0048] Figure 3 Another flow chart showing a data analysis method for a feedstock oil hydrogenation unit provided by an exemplary embodiment is shown;

[0049] Figure 4 A flowchart showing a specific implementation of step S220 provided by an exemplary embodiment;

[0050] Figure 5 A schematic diagram of the topological structure of the first neural network is shown;

[0051] Figure 6 A flowchart showing a specific implementation of step S230 provided by an exemplary embodiment;

[0052] Figure 7 A flowchart showing a specific implementation of step S240 provided by an exemplary embodiment;

[0053] Figure 8 to Figure 11 A schematic diagram showing the relationship between the predicted physical property data of the constructed target neural network and the corresponding real physical property data;

[0054] Fig.12 A block diagram of a data analysis device for a feedstock oil hydrogenation device provided by an exemplary embodiment is shown;

[0055] Fig.13 A block diagram of an electronic device provided by an exemplary embodiment is shown.

[0056] Description of reference numerals:

[0057] 1-Raw material tank; 2-Raw material oil pump; 3-Hydrogen flow meter; 4-Buffer tank; 5-First reactor; 6-Second reactor; 7-High pressure separator; 8-Stabilization tower; 9-Product tank; 10-Water washing tower. DETAILED DESCRIPTION

[0058] The specific implementation of the present disclosure is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the present disclosure, and is not used to limit the present disclosure.

[0059] It should be noted that all actions of acquiring signals, information or data in the present disclosure are carried out in compliance with the relevant data protection laws and policies of the country where the device is located and with the authorization given by the owner of the corresponding device.

[0060] As mentioned above, in the industrial production process, as the operating time of the hydrogenation unit increases, the activity of the hydrotreating catalyst will be lost. Studies have found that as the desulfurization depth increases, the catalyst deactivation rate will increase accordingly. In addition, due to the differences in the type of catalyst used, the properties of the raw oil and the process parameters, the catalyst deactivation rate is also different. The change in catalyst activity directly affects whether the quality of the product meets the production requirements.

[0061] In this regard, the disclosed embodiment provides a data analysis method for a crude oil hydrogenation unit. Based on the industrial operation of the hydrogenation unit, a target neural network is constructed by data modeling. After the target neural network is constructed, it is only necessary to input the corresponding condition data into the target neural network to predict the physical property data of the product generated by the hydrogenation unit under the corresponding conditions. The physical property data can characterize the properties of the generated product. For example, the physical property data includes sulfur content, density and distillation range data. Based on this data analysis method, it is possible to predict the sulfur content, product density and product distillation range of the product in advance before the product is actually generated, so as to find out in advance the products that do not meet the production requirements, and it is also beneficial to feedback the input condition data according to the predicted physical property data, thereby optimizing the condition data. This has practical guiding significance for the stable production operation of the hydrogenation unit and the optimization of the hydrogenation process.

[0062] Figure 1 A schematic diagram of a diesel hydrogenation process provided by an exemplary embodiment is shown, please refer to Figure 1 The raw oil in the raw material tank 1 is preheated and transported by the raw material pump 2. The hydrogen input from the outside is mixed with the raw oil transported by the raw material pump 2 through the buffer tank 4 (wherein the hydrogen flowmeter 3 is used to measure the flow of hydrogen). The mixed raw oil and hydrogen enter the first reactor 5 (referred to as the first reactor), and react under the action of the catalyst in the first reactor 5 at a certain temperature and a certain pressure. The distillate enters the second reactor 6 (referred to as the second reactor), and reacts under the action of the catalyst in the second reactor 6 at a certain temperature and a certain pressure. The distillate enters the high-pressure separator 7 for separation. At the bottom of the high-pressure separator 7, the liquid distillate enters the stabilization tower 8, and after the hydrogen component is removed, it enters the product tank 9 as the obtained product. At the top of the high-pressure separator 7, the gaseous product enters the water washing tower 10 to remove harmful gases therein, and the gas component after water washing is discharged through the pipeline.

[0063] In an optional embodiment, the feedstock oil may be a diesel fraction produced by atmospheric and vacuum distillation or by other processes, or a mixture of diesel fractions produced by different processes, for example, the feedstock oil is a mixed feedstock oil composed of straight-run diesel, catalytic cracking diesel, coker diesel, coker gasoline, etc. The feedstock oil hydrogenation unit may be a diesel hydrogenation unit, which may adopt a fixed bed hydrogenation process or a continuous liquid phase hydrogenation process, etc.

[0064] In an optional embodiment, the hydrogenation process conditions are generally as follows: reaction temperature is 270-380°C, reaction pressure is 6.8-8.0 MPa, liquid hourly volume space velocity is 0.19-1.8 h -1 , the hydrogen-oil ratio is 400-3000. RS-2100 catalyst is selected, which is a NiMo system diesel hydrofining catalyst. The loading amount in the first reactor and the second reactor is 137.9m 3and 65m 3 .

[0065] In an optional embodiment, the hydrogen comes from the refinery hydrogen pipeline network or catalytic reforming unit, and can also come from the outlet of a new hydrogen machine of other hydrotreating units or hydrocracking units.

[0066] Figure 2 A flow chart of a data analysis method for a crude oil hydrogenation unit provided by an exemplary embodiment is shown. Figure 2 As shown, the method includes:

[0067] S110, obtaining target condition data, wherein the target condition data includes the composition of the feedstock oil, the operating time of the hydrogenation unit, and the process parameters adopted by the hydrogenation unit during the operating time.

[0068] Considering that the composition of the crude oil, the operating time of the hydrogenation unit, and the process parameters adopted by the hydrogenation unit during the operating time may all affect the properties of the product, the three types of data, namely, the composition of the crude oil, the operating time, and the process parameters, are used as the input data of the target neural network. Therefore, the target condition data includes the composition of the crude oil, the operating time of the hydrogenation unit, and the process parameters adopted by the hydrogenation unit during the operating time.

[0069] Exemplarily, the raw oil is a mixed raw oil composed of straight-run diesel, catalytic cracking diesel, coker diesel and coker gasoline, then the raw oil composition includes the straight-run diesel ratio, catalytic cracking diesel ratio, coke diesel ratio and coke gasoline ratio, and the process parameters include but are not limited to the average temperature of the first reactor, the temperature rise of the first reactor, the average temperature of the second reactor, the temperature rise of the second reactor, the purity of circulating hydrogen, the circulating gas volume, the amount of new hydrogen, the inlet pressure of the first reactor and the air velocity. In actual applications, the process parameters can be less or more than the parameters listed above.

[0070] S120, through the target neural network, multiple data in the target condition data are processed to predict the physical property data of the product generated by the hydrogenation unit under the conditions of the feedstock oil composition, the operating time and the process parameters, and the predicted physical property data is output.

[0071] After obtaining the target condition data, multiple data in the target condition data are input into the target neural network, and the target neural network performs fitting processing on the input data to predict the physical property data of the product generated by the hydrogenation device under the conditions given by the target condition data, and the physical property data is the predicted physical property data, and then the predicted physical property data is output. Among them, the physical property data include but are not limited to sulfur content, density and distillation range data, etc. The distillation range data include but are not limited to the initial distillation point of the product, the 50% distillation point of the product, the 95% distillation point of the product and the final distillation point of the product.

[0072] From the above process, it can be seen that the target condition data required by this scheme is relatively easy to obtain, and the crude oil composition, device operating time and device process parameters are correlated with the product sulfur content, density, distillation range and other properties. It is only necessary to input the crude oil composition, device operating time and the process parameters taken by the device at the corresponding operating time into the target neural network to quickly predict the product properties.

[0073] In some embodiments, the target condition data for each day in the next month can be input into the target neural network in advance to predict the physical property data of the products generated every day in the next month, so as to know in advance whether the sulfur content of the products generated in the next month meets the production requirements, and readjust the production process for products that may not meet the production requirements in the future until the sulfur content of the products generated meets the production requirements, thereby ensuring the quality of product production. Alternatively, it can also be used to guide the adjustment of existing processes. Specifically, the target condition data after preliminary adjustment is input into the target neural network to obtain the predicted physical property data corresponding to the adjusted target condition data. According to the predicted physical property data, the refinery technicians can be guided to make actual adjustments to the process. For example, the average temperature of a counter is increased by 10°C, and the physical property data of the products generated after the average temperature of a counter is increased by 10°C is predicted. According to the predicted physical property data, the actual process adjustment of the average temperature of a counter is guided.

[0074] Therefore, this scheme can ensure the product quality of the hydrogenation unit and has practical guiding significance for the optimization of the hydrogenation process.

[0075] Figure 3 Another flow chart of a data analysis method for a crude oil hydrogenation unit provided by an exemplary embodiment is shown to construct a target neural network. Figure 3 As shown, the method includes:

[0076] S210, obtaining multiple groups of training data, each group of training data includes historical condition data and real physical property data corresponding to the historical condition data.

[0077] The historical condition data includes the composition of the crude oil, the operating time of the hydrogenation unit, and the process parameters adopted by the hydrogenation unit during the operating time. Exemplarily, the composition of the crude oil includes the ratio of straight diesel, the ratio of catalytic diesel, the ratio of coke diesel, and the ratio of coke steam, and the process parameters include but are not limited to the average temperature of the first reactor, the temperature rise of the first reactor, the average temperature of the second reactor, the temperature rise of the second reactor, the purity of circulating hydrogen, the circulating gas volume, the amount of new hydrogen, the inlet pressure of the first reactor, and the air velocity.

[0078] The real physical property data corresponding to the historical condition data are the physical property data of the products actually generated by the hydrogenation unit under the conditions given by the above historical condition data. The physical property data include but are not limited to sulfur content, density and distillation range data, etc. The distillation range data include but are not limited to the initial distillation point of the product, the 50% distillation point of the product, the 95% distillation point of the product and the final distillation point of the product.

[0079] In an optional embodiment, the target neural network to be constructed includes a total of 14 input data, namely, the operating time of the hydrogenation unit, the average temperature of the first reactor, the temperature rise of the first reactor, the average temperature of the second reactor, the temperature rise of the second reactor, the circulating hydrogen purity, the circulating gas volume, the new hydrogen volume, the first reactor inlet pressure, the air velocity, the straight-diesel ratio, the catalytic-diesel ratio, the coke-diesel ratio and the coke-steam ratio; the target neural network to be constructed includes a total of 6 output data, namely, the product sulfur content, the product density, the product initial distillation point, the product 50% distillation point, the product 95% distillation point and the product final distillation point.

[0080] Based on the above embodiment, the following Table 1 shows five sets of training data among the multiple sets of training data acquired in step S210.

[0081] Table 1

[0082]

[0083]

[0084] In Table 1, the average temperature of the first and second inversions is calculated by the following formula:

[0085]

[0086]

[0087] Among them, T1 is the average temperature, T 1_1 is the average temperature of the bed, T 1_2 is the average temperature of the first and second beds, V1 is the catalyst loading volume of the first bed, V 1_1 is the catalyst loading volume of one bed, V 1_2 is the catalyst loading volume of the first reactor and the second bed, T2 is the average temperature of the second reactor, T 2_in is the inlet temperature of the second reactor, T 2_out is the outlet temperature of the second reactor.

[0088] S220, determine the input layer nodes, hidden layer nodes and output layer nodes to obtain a first neural network.

[0089] S230, optimizing the network parameters of the first neural network using a genetic algorithm to obtain target network parameters, and obtaining the second neural network according to the target network parameters.

[0090] The network parameters of the first neural network include weight information and threshold information of the first neural network. Specifically, the weight information includes the weight between any two nodes of two adjacent layers in the network, and the threshold information includes the threshold of each node in the hidden layer and the threshold of each node in the output layer. The target network parameters represent a set of optimal weight information and a set of optimal threshold information found by using a genetic algorithm.

[0091] After obtaining the first neural network, a genetic algorithm is used to optimize the weight information and threshold information of the first neural network to obtain a set of optimal weight information and a set of optimal threshold information, and the set of optimal weight information and the set of optimal threshold information are applied to the first neural network to obtain the second neural network.

[0092] S240: Training the second neural network according to the multiple sets of training data.

[0093] After the second neural network is obtained, the second neural network is trained according to the acquired multiple sets of training data to update the network parameters in the second neural network, that is, to update the weight information and threshold information in the second neural network.

[0094] S250: After the training is completed, the trained second neural network is determined as the target neural network.

[0095] In the above scheme, a genetic algorithm is first used to optimize the network parameters to obtain target network parameters, and a second neural network is obtained based on the target network parameters. Then, training is started with the second neural network as the starting point. That is to say, the starting network parameters of the neural network training are not randomly initialized, but are the target network parameters obtained after optimization using a genetic algorithm. Therefore, the second neural network obtained based on the target network parameters already has a certain initial fitting ability, which is conducive to accelerating the convergence speed of the second neural network in subsequent training steps.

[0096] Figure 4 FIG. 4 is a flowchart showing a specific implementation of step S220 provided by an exemplary embodiment. Figure 4 As shown, step S220 includes:

[0097] S2201, determine the number of input layer nodes and the number of output layer nodes.

[0098] Among them, the number of nodes in the input layer is the same as the number of items in the input data, and each node in the input layer is used to receive the input of one of the items in the input data. The number of nodes in the output layer is the same as the number of items in the output data, and each node in the output layer is used to output one of the items of the predicted physical property data.

[0099] S2202, based on the number of input layer nodes and the number of output layer nodes, estimate the search range of the number of hidden layer nodes, and sequentially select values ​​from the search range.

[0100] After determining the number of input layer nodes and the number of output layer nodes, first determine the approximate range of the number of hidden layer nodes based on the number of input layer nodes and the number of output layer nodes, that is, estimate the search range of the number of hidden layer nodes.

[0101] For example, the number of hidden layer nodes can be preliminarily calculated using the following empirical formula:

[0102]

[0103] Where m is the number of hidden layer nodes, n is the number of input layer nodes, l is the number of output layer nodes, and a is a constant in the range of [1,10].

[0104] After the number of hidden layer nodes m is calculated by the above empirical formula, it is rounded, and then the corresponding range is determined with the rounded value as the center as the search range, and then values ​​are taken in sequence from within the search range.

[0105] S2203, according to the number of input layer nodes, the number of output layer nodes and the number of hidden layer nodes taken out this time, determine the input layer nodes, hidden layer nodes and output layer nodes to obtain the candidate neural network this time.

[0106] For the values ​​taken out from the search range this time, the corresponding number of input layer nodes, the corresponding number of hidden layer nodes and the corresponding number of output layer nodes are determined according to the number of input layer nodes, the number of output layer nodes and the value of the number of hidden layer nodes taken out this time, and the candidate neural network of this time is obtained. Among them, the network parameters in the candidate neural network can be initialized and generated first.

[0107] S2204, inputting multiple groups of historical condition data into the candidate neural network of this time in sequence, and obtaining predicted physical property data corresponding to the input historical condition data.

[0108] Multiple groups of historical condition data are input into the candidate neural network in sequence. The candidate neural network performs fitting processing on multiple data in any group of historical condition data input to predict the predicted physical property data of the product generated by the hydrogenation device under the conditions corresponding to the group of historical condition data. Finally, the predicted physical property data corresponding to each group of historical condition data are obtained.

[0109] S2205, calculating the error value of the candidate neural network based on the predicted physical property data and the corresponding real physical property data.

[0110] According to the predicted physical property data and the actual physical property data corresponding to each set of historical condition data, the error value of the candidate neural network is calculated. Optionally, the error value of the candidate neural network is calculated using the mean square error (MSE):

[0111]

[0112] Among them, y i represents the real physical property data corresponding to the i-th group of historical condition data among the multiple groups of historical condition data input to the candidate neural network this time, y i ′ represents the predicted physical property data corresponding to the i-th group of historical condition data, and n represents the number of multiple groups of historical condition data input into the candidate neural network this time.

[0113] S2206, determining a candidate neural network with the smallest error value from multiple candidate neural networks as the first neural network.

[0114] In the previous steps, it is necessary to sequentially select values ​​from the search range corresponding to the number of nodes in the hidden layer, and for each value taken out, a corresponding candidate neural network will be obtained, and a total of multiple candidate neural networks will be obtained. According to the error value corresponding to each candidate neural network, a candidate neural network with the smallest error value is determined from the multiple candidate neural networks as the first neural network.

[0115] In some embodiments, the first neural network is a multi-layer feedforward neural network (BP neural network), and adopts a double hidden layer structure, which can improve the fitting effect, that is, the first neural network includes an input layer, a first hidden layer, a second hidden layer and an output layer. For the first hidden layer and the second hidden layer, firstly estimate the first search range corresponding to the first hidden layer and the second search range corresponding to the second hidden layer respectively. Then, first fix the number of nodes in one of the hidden layers, such as taking a value from the first search range corresponding to the first hidden layer as the number of nodes in the first hidden layer, and fix it. On the basis of the fixed number of nodes in the first hidden layer, take values ​​from the second search range corresponding to the second hidden layer in sequence as the number of nodes in the second hidden layer. According to the number of nodes in the input layer, the number of nodes in the first hidden layer, the number of nodes in the second hidden layer and the number of nodes in the output layer, obtain the corresponding candidate neural network, then continue to take the next value from the first search range and fix it. On the basis of the fixed number of nodes in the first hidden layer, take values ​​from the second search range in sequence, and repeat the above process. Assuming that a total of M values ​​can be taken in the first search range and a total of N values ​​can be taken in the second search range, M*N candidate neural networks can be obtained in the end. Finally, the candidate neural network with the smallest error value is determined as the first neural network from the M*N candidate neural networks.

[0116] As an example, the first neural network may have a topology of 14-16-11-6, which means that the network has 14 input layer nodes, 16 first hidden layer nodes, 11 second hidden layer nodes and 6 output layer nodes. Figure 5 A schematic diagram of a topological structure of the first neural network is shown.

[0117] In the above scheme, the search range of the number of hidden layer nodes is first determined based on the number of input layer nodes and the number of output layer nodes, and then multiple fittings are performed within the search range to determine the neural network with the smallest error value as the first neural network. Therefore, the first neural network obtained has initially possessed a certain fitting ability, which is conducive to speeding up the optimization speed and training speed of subsequent steps.

[0118] Figure 6 A flowchart showing a specific implementation of step S230 provided by an exemplary embodiment is shown. Figure 6 As shown, step S230 includes:

[0119] S2301, randomly generate k groups of network parameters according to the set population size k.

[0120] Wherein, k is a positive integer greater than 1. Exemplarily, the population size k is set to 100, and then 100 sets of network parameters are randomly generated according to the set population size k.

[0121] S2302: For each set of network parameters, determine an error value of a first neural network that applies the set of network parameters, and use the error value as the fitness corresponding to the set of network parameters.

[0122] For each set of network parameters in the k sets of network parameters, the set of network parameters is applied to the first neural network, and an error value (such as MSE) of the first neural network applying the set of network parameters is determined, and then the error value is used as the fitness corresponding to the set of network parameters.

[0123] Wherein, each group of network parameters includes a group of weight information and a group of threshold information, a group of weight information includes the weight between any two nodes of two adjacent layers in the network, and a group of threshold information includes the threshold of each node in the hidden layer and the threshold of each node in the output layer. Exemplarily, the first neural network adopts a double hidden layer structure, and each group of network parameters includes the weight from each node in the input layer to each node in the first hidden layer, the weight from each node in the first hidden layer to each node in the second hidden layer, the weight from each node in the second hidden layer to each node in the output layer, the threshold of each node in the first hidden layer, the threshold of each node in the second hidden layer, and the threshold of each node in the output layer.

[0124] S2303, according to the fitness corresponding to each group of network parameters, select, cross and mutate in the current k groups of network parameters to form new k groups of network parameters, and return to step S2302.

[0125] Repeat steps S2302 to S2303 until the termination condition is reached.

[0126] S2304: When the termination condition is reached, a set of optimal network parameters is determined from the current k sets of network parameters as target network parameters.

[0127] When the termination condition is reached, according to the error value corresponding to each group of network parameters, a group of network parameters with the smallest error value is determined from the current k groups of network parameters as the target network parameters, thereby completing the optimization of the network parameters. The target network parameters are a group of optimal network parameters optimized by the genetic algorithm. In step S230, the first neural network applying the group of target network parameters is determined as the second neural network.

[0128] Furthermore, in the training process of the neural network, the network parameters need to be updated according to the learning rate. The size of the learning rate will affect the convergence speed of the network. Generally speaking, the learning rate is a fixed value given by experience. If the learning rate is too small, the number of training iterations will increase. If the learning rate is too large, it will be impossible to find the optimal solution. Figure 7 A flowchart showing a specific implementation of step S240 provided by an exemplary embodiment is shown.

[0129] like Figure 7 As shown, in step S240, the second neural network is subjected to the following iterative process multiple times according to multiple groups of historical data:

[0130] S2401, during this iteration, multiple groups of historical condition data corresponding to this iteration are sequentially input into the second neural network to obtain predicted physical property data corresponding to the input historical condition data.

[0131] S2402, calculating an error value based on the predicted physical property data and the corresponding actual physical property data.

[0132] S2403, according to the size relationship between the error value of this iteration and the error value of the previous iteration, make corresponding adjustments based on the learning rate of the previous iteration to obtain the learning rate of this iteration.

[0133] S2404: Update the network parameters in the second neural network according to the learning rate of this iteration.

[0134] From the above process, it can be seen that in the training process of the second neural network, the learning rate is not fixed, but is obtained by making corresponding adjustments based on the learning rate of the previous iteration according to the relationship between the error value of this iteration and the error value of the previous iteration. Therefore, the learning rate can be dynamically adjusted with the error of the iterative process.

[0135] In some embodiments, the following learning rate equation is established:

[0136]

[0137] in, represents the learning rate of this iteration, represents the learning rate of the last iteration, MSE(n+1) represents the error value of this iteration, MSE(n) represents the error value of the last iteration, α and β are fitting parameters, and their value ranges are [0,1].

[0138] Based on the above learning rate equation, it can be seen that if the error value of this iteration is less than the error value of the previous iteration, the learning rate is increased on the basis of the learning rate of the previous iteration to obtain the learning rate of this iteration; if the error value of this iteration is greater than or equal to the error value of the previous iteration, the learning rate is reduced on the basis of the learning rate of the previous iteration to obtain the learning rate of this iteration.

[0139] In the above embodiment, when the error value of this iteration decreases, it means that the training of the second neural network is optimized in a good direction, and the learning rate is increased accordingly to speed up the training process and reduce the number of iterations so that the second neural network can converge faster. When the error value of this iteration increases, it means that the training of the second neural network is optimized in a bad direction, and the learning rate is reduced accordingly to facilitate finding the optimal solution.

[0140] In an optional embodiment, in order to eliminate the differences in the various data in the network input data due to different orders of magnitude and dimensions, it is necessary to normalize the various data and input the normalized data as input data.

[0141] Exemplarily, the data normalization processing formula is as follows:

[0142]

[0143] Among them, X i is the i-th item in the input data, t i is the normalized data corresponding to the i-th item, t max is the maximum value of the i-th data, t min is the minimum value of the i-th data.

[0144] After obtaining the target condition data, each item of the target condition data needs to be normalized, and then the normalized data is used as the input of the target neural network. Correspondingly, in the above steps S210 to S250, each item of the historical condition data also needs to be normalized, and then the normalized data is used as the input of each neural network, including the candidate neural network, the first neural network, and the second neural network.

[0145] After the target neural network is constructed through the above steps S210 to S250, the effect of the target neural network is tested on the test set. Figure 8 to Figure 11 The figure shows the relationship between the predicted physical property data of the constructed target neural network and the corresponding real physical property data. Figure 8 to Figure 11 It can be seen that the data points of properties such as product sulfur content, density and distillation range are all close to the diagonal line, indicating that the deviation between the predicted value and the true value of the physical property data of the target neural network is small, indicating that the obtained target neural network has good predictive ability and can more accurately predict the product properties of the hydrogenation unit under different operating times and different process parameters, with high accuracy and practical application value.

[0146] Fig.12 A block diagram of a data analysis device for a crude oil hydrogenation device provided by an exemplary embodiment is shown. Fig.12 , the device 300 comprises:

[0147] The condition data acquisition module 310 is used to acquire target condition data, wherein the target condition data includes the composition of the feedstock oil, the operating time of the hydrogenation unit, and the process parameters adopted by the hydrogenation unit during the operating time;

[0148] The physical property data prediction module 320 is used to process multiple data in the target condition data through a target neural network to predict the physical property data of the product generated by the hydrogenation unit under the conditions of the feedstock oil composition, the operating time and the process parameters, and output the predicted physical property data.

[0149] Optionally, the device 300 further includes:

[0150] A training data acquisition module, used to acquire multiple sets of training data, each set of training data includes historical condition data and real physical property data corresponding to the historical condition data;

[0151] A network determination module, used to determine input layer nodes, hidden layer nodes and output layer nodes to obtain a first neural network;

[0152] A parameter optimization module, used to optimize the network parameters of the first neural network using a genetic algorithm to obtain target network parameters, and obtain a second neural network according to the target network parameters;

[0153] A network training module, used for training the second neural network according to the multiple sets of training data;

[0154] The target determination module is used to determine the trained second neural network as the target neural network after the training is completed.

[0155] Optionally, the network determination module includes:

[0156] A node number determination module is used to determine the number of input layer nodes and the number of output layer nodes;

[0157] A search range determination module, used to estimate the search range of the number of nodes in the hidden layer according to the number of nodes in the input layer and the number of nodes in the output layer, and sequentially take values ​​from the search range;

[0158] A candidate network determination module is used to determine the input layer nodes, the hidden layer nodes and the output layer nodes according to the number of input layer nodes, the number of output layer nodes and the number of hidden layer nodes taken out this time, so as to obtain the candidate neural network this time;

[0159] A candidate error determination module, used to sequentially input multiple groups of historical condition data into the candidate neural network of this time, obtain predicted physical property data corresponding to the input historical condition data, and calculate the error value of the candidate neural network according to the predicted physical property data and the corresponding real physical property data;

[0160] The first network determination module is used to determine the candidate neural network with the smallest error value from multiple candidate neural networks as the first neural network. Optionally, the parameter optimization module includes:

[0161] The initial population generation module is used to randomly generate k groups of network parameters according to the set population size k; k is a positive integer greater than 1;

[0162] A fitness calculation module, used to determine, for each set of network parameters, an error value of the first neural network applying the set of network parameters, and use the error value as the fitness corresponding to the set of network parameters;

[0163] A descendant population generation module is used to select, cross and mutate the current k groups of network parameters according to the fitness corresponding to each group of network parameters to form new k groups of network parameters, and jump to the fitness calculation module to execute the step of determining the error value of the first neural network applying the group of network parameters for each group of network parameters;

[0164] The optimal parameter determination module is used to determine a group of optimal network parameters from the current k groups of network parameters as the target network parameters when the termination condition is reached.

[0165] Optionally, the network training module is used to perform multiple iterations on the second neural network according to the multiple groups of historical data. The network training module includes a prediction submodule, an error calculation submodule, a learning rate calculation submodule and a parameter updating submodule. In each iteration process: the prediction submodule is used to input the multiple groups of historical condition data corresponding to this iteration into the second neural network in sequence, and obtain predicted physical property data corresponding to the input historical condition data; the error calculation submodule is used to calculate the error value according to the predicted physical property data and the corresponding real physical property data; the learning rate calculation submodule is used to make corresponding adjustments based on the learning rate of the previous iteration according to the relationship between the error value of this iteration and the error value of the previous iteration, and obtain the learning rate of this iteration; the parameter updating submodule is used to update the network parameters in the second neural network according to the learning rate of this iteration.

[0166] Optionally, the learning rate calculation submodule is used to: when the error value of this iteration is less than the error value of the previous iteration, increase the learning rate on the basis of the learning rate of the previous iteration to obtain the learning rate of this iteration; when the error value of this iteration is greater than or equal to the error value of the previous iteration, reduce the learning rate on the basis of the learning rate of the previous iteration to obtain the learning rate of this iteration.

[0167] Optionally, the device 300 also includes: a normalization processing module, which is used to normalize each item of data in the target condition data after acquiring the target condition data, and use the normalized data as input of the target neural network.

[0168] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0169] Fig.13 FIG. 4 is a block diagram of an electronic device 400 according to an exemplary embodiment. Fig.13 As shown, the electronic device 400 may include: a processor 401 and a memory 402. The electronic device 400 may also include one or more of a multimedia component 403, an input / output (I / O) interface 404, and a communication component 405.

[0170] The processor 401 is used to control the overall operation of the electronic device 400 to complete all or part of the steps in the data analysis method of the feedstock oil hydrogenation unit. The memory 402 is used to store various types of data to support the operation of the electronic device 400, and these data may include instructions for any application or method used to operate on the electronic device 400, and application-related data, such as contact data, sent and received messages, pictures, audio, video, etc. The memory 402 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read-Only Memory, referred to as EPROM), programmable read-only memory (Programmable Read-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, disk or optical disk. The multimedia component 403 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal may be further stored in the memory 402 or sent through the communication component 405. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 404 provides an interface between the processor 401 and other interface modules, and the other interface modules may be keyboards, mice, buttons, etc. These buttons may be virtual buttons or physical buttons. The communication component 405 is used for wired or wireless communication between the electronic device 400 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, 4G, NB-IOT, eMTC, or other 5G, etc., or a combination of one or more of them, is not limited here. Therefore, the corresponding communication component 405 may include: Wi-Fi module, Bluetooth module, NFC module, etc.

[0171] In an exemplary embodiment, the electronic device 400 can be implemented by one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned data analysis method for the crude oil hydrogenation unit.

[0172] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the data analysis method of the feedstock oil hydrogenation device described above are implemented. For example, the computer-readable storage medium can be the memory 402 including the program instructions, and the program instructions can be executed by the processor 401 of the electronic device 400 to complete the data analysis method of the feedstock oil hydrogenation device described above.

[0173] In another exemplary embodiment, a computer program product is also provided, which includes a computer program that can be executed by a programmable device, and the computer program has a code portion for executing the data analysis method of the crude oil hydrogenation unit mentioned above when executed by the programmable device.

[0174] The preferred embodiments of the present disclosure are described in detail above in conjunction with the accompanying drawings; however, the present disclosure is not limited to the specific details in the above embodiments. Within the technical concept of the present disclosure, a variety of simple modifications can be made to the technical solution of the present disclosure, and these simple modifications all fall within the protection scope of the present disclosure.

[0175] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present disclosure will not further describe various possible combinations.

[0176] In addition, various embodiments of the present disclosure may be arbitrarily combined, and as long as they do not violate the concept of the present disclosure, they should also be regarded as the contents disclosed by the present disclosure.

Claims

1. A data analysis method for a crude oil hydrogenation unit, characterized in that: include: Acquiring target condition data, the target condition data including the composition of the feedstock oil, the operating time of the hydrogenation unit, and the process parameters adopted by the hydrogenation unit during the operating time; Processing multiple data in the target condition data through a target neural network to predict the physical property data of the product generated by the hydrogenation device under the conditions of the feedstock oil composition, the operating time and the process parameters, and outputting the predicted physical property data; The target neural network is determined in the following way: Acquire multiple sets of training data, each set of training data includes historical condition data and real physical property data corresponding to the historical condition data; Determine input layer nodes, hidden layer nodes and output layer nodes to obtain a first neural network; Using a genetic algorithm to optimize the network parameters of the first neural network to obtain target network parameters, and obtaining a second neural network based on the target network parameters; Training the second neural network according to the multiple sets of training data; After the training is completed, the trained second neural network is determined as the target neural network.

2. The method according to claim 1, characterized in that The step of determining the input layer nodes, the hidden layer nodes and the output layer nodes to obtain the first neural network comprises: Determine the number of input layer nodes and the number of output layer nodes; According to the number of nodes in the input layer and the number of nodes in the output layer, a search range of the number of nodes in the hidden layer is estimated, and values ​​are sequentially taken from the search range; According to the number of input layer nodes, the number of output layer nodes and the number of hidden layer nodes taken out this time, the input layer nodes, the hidden layer nodes and the output layer nodes are determined to obtain the candidate neural network this time; Inputting multiple sets of historical condition data into the candidate neural network in sequence, obtaining predicted physical property data corresponding to the input historical condition data, and calculating the error value of the candidate neural network according to the predicted physical property data and the corresponding real physical property data; The candidate neural network with the smallest error value is determined from multiple candidate neural networks as the first neural network.

3. The method according to claim 1, characterized in that The method of optimizing the network parameters of the first neural network by using a genetic algorithm to obtain target network parameters includes: According to the set population size k, k groups of network parameters are randomly generated; k is a positive integer greater than 1; For each set of network parameters, determining an error value of a first neural network applying the set of network parameters, and using the error value as the fitness corresponding to the set of network parameters; According to the fitness corresponding to each group of network parameters, selection, crossover and mutation are performed in the current k groups of network parameters to form new k groups of network parameters, and returning to execute the step of determining the error value of the first neural network applying the group of network parameters for each group of network parameters; When the termination condition is reached, a set of optimal network parameters is determined from the current k sets of network parameters as the target network parameters.

4. The method according to claim 1, characterized in that: The training of the second neural network according to the multiple sets of training data includes: The second neural network is subjected to the following iterative process multiple times according to the multiple sets of training data: In this iteration, multiple groups of historical condition data corresponding to this iteration are sequentially input into the second neural network to obtain predicted physical property data corresponding to the input historical condition data; Calculating an error value based on the predicted physical property data and the corresponding actual physical property data; According to the relationship between the error value of this iteration and the error value of the previous iteration, the learning rate of the previous iteration is adjusted accordingly to obtain the learning rate of this iteration; The network parameters in the second neural network are updated according to the learning rate of this iteration.

5. The method according to claim 4, characterized in that According to the relationship between the error value of this iteration and the error value of the previous iteration, a corresponding adjustment is made on the basis of the learning rate of the previous iteration to obtain the learning rate of this iteration, including: If the error value of this iteration is less than the error value of the previous iteration, the learning rate is increased based on the learning rate of the previous iteration to obtain the learning rate of this iteration; If the error value of this iteration is greater than or equal to the error value of the previous iteration, the learning rate is reduced based on the learning rate of the previous iteration to obtain the learning rate of this iteration.

6. The method according to claim 1, characterized in that After acquiring the target condition data, the method further includes: Each item of the target condition data is normalized respectively, and the normalized data is used as the input of the target neural network.

7. A data analysis device for a crude oil hydrogenation device, characterized in that: include: A condition data acquisition module, used to acquire target condition data, wherein the target condition data includes the composition of the feedstock oil, the operating time of the hydrogenation unit, and the process parameters adopted by the hydrogenation unit during the operating time; A physical property data prediction module, used to process multiple data in the target condition data through a target neural network to predict the physical property data of the product generated by the hydrogenation unit under the conditions of the feedstock oil composition, the operating time and the process parameters, and output the predicted physical property data; A training data acquisition module, used to acquire multiple sets of training data, each set of training data includes historical condition data and real physical property data corresponding to the historical condition data; A network determination module, used to determine input layer nodes, hidden layer nodes and output layer nodes to obtain a first neural network; A parameter optimization module, used to optimize the network parameters of the first neural network using a genetic algorithm to obtain target network parameters, and obtain a second neural network according to the target network parameters; A network training module, used for training the second neural network according to the multiple sets of training data; The target determination module is used to determine the trained second neural network as the target neural network after the training is completed.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

9. An electronic device, characterized in that: include: a memory having a computer program stored thereon; A processor, configured to execute the computer program in the memory to implement the steps of the method according to any one of claims 1 to 6.

Images