CN115454009A - Model Predictive Control Method of Component Formulation for Chemical Production - Google Patents

Abstract

The invention relates to the technical field of chemical control, in particular to a component distribution model predictive control method for chemical production. The method obtains the reaction characteristics of each distillation range point, and matches the distillation range points in the distillation process under different control parameters according to the reaction characteristics to obtain a plurality of matched pairs. And further obtaining a switchable index according to the reaction characteristic difference between the distillation range points in the matching pair, selecting the matching pair based on the switchable index to construct an initial training set, and amplifying the initial training set according to the matching data to obtain a prediction model for training the long-short term memory production group to prepare, so as to predict real-time data and control container parameters according to the prediction parameters. The invention predicts the parameters to be adjusted of the future distillation range point by constructing the component distribution model, controls the parameters in time and ensures the production efficiency.

Description

Model Predictive Control Method of Component Formulation for Chemical Production

technical field

The invention relates to the technical field of chemical control, in particular to a predictive control method for component allocation models used in chemical production.

Background technique

High-boiling aromatic solvents are developed and produced based on reformed aromatics as raw materials, according to international special aromatic solvent standards. They have strong solvency, low toxicity, low odor, high boiling point, slow volatilization, no water and olefins, no chlorine and heavy metals , stable chemical and physical properties and good leveling. Heavy aromatics are a kind of crude product with different properties in the early stage. Light fractions at 250 ° C can be extracted, combined with tail oil with a higher distillation range (distillation range about 260 ~ 380 ° C), rational utilization of C10 aromatics, production of high boiling point Aromatics. In the general production process, catalytic hydrogenation is required, followed by rectification and separation. However, at present, this preparation parameter is adjusted by manual experience, which is prone to unstable quality at first, and human monitoring is always required in the later stage. Therefore, in the assembly In the process of dispensing, a system is needed to intelligently judge whether the process needs to be improved and to intervene in hydrogenation to improve the efficiency of distillation.

Contents of the invention

In order to solve the above-mentioned technical problems, the object of the present invention is to provide a component formulation model predictive control method for chemical production, and the adopted technical scheme is as follows:

The present invention proposes a method for predictive control of component formulation model for chemical production, said method comprising:

Collect the set temperature value, steam temperature and heating element temperature in the container at each distillation point in the distillation process; obtain the mixed code of the appearance characteristics of the reactants in the container; according to the steam temperature, heating element temperature and container Volumetric space velocity constructs short-term response descriptors;

Match different distillation range points according to the mixed coding difference of appearance characteristics and short-term response descriptor difference in the distillation process under different control parameters, and obtain multiple matching pairs;

The switchable index is obtained according to the set temperature value difference, steam temperature difference and heating element temperature difference between two distillation range points in the matching pair; select the data in the preset number of matching pairs with the largest switchable index as the initial training set , augment the initial training set according to the matching distillation point data in the initial training set to obtain a training set, and train the long-short-term memory production group to prepare a prediction model according to the training set;

Input the real-time steam temperature and real-time heating element temperature at the real-time distillation point in the real-time distillation process into the long-short-term memory production component allocation prediction model to obtain the set temperature value and volume space velocity corresponding to the next distillation point, and according to the following The set temperature value and volume space velocity corresponding to a distillation range point control the vessel parameters.

Further, said obtaining the mixed coding of the appearance characteristics of the reactants in the container includes:

Obtain the description words of the appearance characteristics of the reactants in the container, and construct the TF-IDF appearance code according to the description words; collect the color information of the container reactants, and construct the appearance color code with the component values in the RGB color space; Appearance coding and appearance color coding are combined to obtain a mixed coding of appearance features.

Further, said obtaining appearance feature mixed coding includes:

Expand the dimension of the appearance color code to obtain the first extended code; merge the first extended code with the TF-IDF appearance code to obtain the second extended code; reduce the dimensionality of the second extended code to the preset dimension to obtain the appearance features Mixed encoding.

Further, said constructing a short-term reaction descriptor according to the steam temperature under the sampling process, the temperature of the heating element and the volume space velocity of the container includes:

The short-term response descriptor consists of the median value of the steam temperature sequence, the median value of the heating element temperature sequence, the average value of the steam temperature sequence, the average value of the heating element temperature sequence, the initial value of the steam temperature sequence, and the temperature of the heating element at each distillation point in the distillation process. Sequence initial value composition.

Further, the method for obtaining multiple matching pairs includes:

The matching distance was constructed according to the mixed coding difference of appearance features and the short-term reaction descriptor difference between different distillation processes under different reaction parameters; according to the matching distance, the KM matching algorithm was used to match different distillation processes to obtain multiple matching pairs.

Further, the construction of the matching distance based on the mixed encoding difference and short-term reaction descriptor difference between different distillation processes under different reaction parameters includes:

The matching distance is obtained according to the matching distance formula, and the matching distance formula includes:

为馏程点X和馏程点Y之间的匹配距离，

为馏程点X和馏程点Y之间的短期反应描述子相似度，

为馏程点X和馏程点Y之间的外观特征混合编码相似度，

为馏程点X和馏程点Y在整个蒸馏过程中对应的馏程点序号的差异绝对值；若

，则

为无穷大。in,

is the matching distance between distillation point X and distillation point Y,

is the short-term response descriptor similarity between distillation point X and distillation point Y,

Coding similarity for the mixture of appearance features between distillation point X and distillation point Y,

is the absolute value of the difference between the distillation point numbers corresponding to the distillation point X and the distillation point Y in the whole distillation process; if

,but

for infinity.

Further, said obtaining the switchable index according to the set temperature value difference, the steam temperature difference and the heating element temperature difference in the two distillation processes in the matching pair includes:

The switchable index is obtained according to the switchable index formula, which includes:

为馏程点X和馏程点Y之间的可切换指数，

为馏程点X和馏程点Y之间的设定温度值的差值绝对值，

为馏程点X和馏程点Y之间的匹配距离，

为馏程点X和馏程点Y之间蒸汽温度序列的动态时间规整距离，

为馏程点X和馏程点Y之间加热元件温度序列的动态时间规整距离。in,

is the switchable index between distillation point X and distillation point Y,

is the absolute value of the difference between the set temperature values between the distillation point X and the distillation point Y,

is the matching distance between distillation point X and distillation point Y,

is the dynamic time-regularized distance of the steam temperature sequence between distillation point X and distillation point Y,

is the dynamic time warping distance of the temperature sequence of the heating element between the distillation point X and the distillation point Y.

Further, said augmenting the initial training set according to the matching distillation process data in the initial training set includes:

Replace the volume space velocity, set temperature, vapor temperature, and heating element temperature of the target distillation point with the volume space velocity, set temperature, vapor temperature, and heating element temperature of the corresponding distillation point in the distillation process to which the matching distillation point belongs , to obtain the first augmented training data;

Replace the volume space velocity, set temperature, steam temperature and heating element temperature of the distillation point next to the target distillation point with the volume space velocity, set temperature, steam temperature of the corresponding distillation point in the distillation process to which the matching distillation point belongs temperature and heating element temperature to obtain the second augmented training data;

The volume space velocity, set temperature, steam temperature and heating element temperature of the target distillation point and the distillation point next to the target distillation point are respectively replaced with the volume space of the corresponding distillation point in the distillation process to which the matching distillation point belongs. Speed, set temperature, steam temperature and heating element temperature in the container to obtain the third augmented training data;

The first augmented training data, the second augmented training data and the third augmented training data of each distillation point are used to obtain a training set.

The present invention has following beneficial effect:

In the embodiment of the present invention, a short-term reaction descriptor and a hybrid code of appearance characteristics are constructed to characterize the reaction characteristics in the distillation process. According to the reaction characteristics, the distillation range points of the distillation process under different parameters can be matched, and then by selecting the matching pair with a large switchable index as the training data, and augmenting the data according to the data between the matching pairs, the subsequent long-term The training process of the short-term memory production composition allocation prediction model has sufficient training basis, and the long-term short-term memory production composition allocation prediction model can output an optimal result according to the augmented data, ensuring the referenceability of the prediction data, and then through the long-term The short-term memory production group allocation forecast model predicts the set temperature value and volume space velocity in the future, and controls the parameters in the container. Realize timely control of container parameters and ensure production efficiency.

Description of drawings

In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or in the prior art, the following will briefly introduce the accompanying drawings that need to be used in the description of the embodiments or the prior art. Apparently, the appended The drawings are only some embodiments of the present invention, and those skilled in the art can also obtain other drawings based on these drawings without creative work.

Fig. 1 is a flow chart of a component formulation model predictive control method for chemical production provided by an embodiment of the present invention;

Fig. 2 is a schematic diagram of a color information collection process provided by an embodiment of the present invention.

detailed description

In order to further explain the technical means and effects of the present invention to achieve the intended purpose of the invention, the following in conjunction with the accompanying drawings and preferred embodiments, for a component allocation model predictive control method for chemical production proposed according to the present invention, Its specific implementation, structure, feature and effect thereof are described in detail as follows. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, the particular features, structures or characteristics of one or more embodiments may be combined in any suitable manner.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field of the invention.

A specific scheme of a component formulation model predictive control method for chemical production provided by the present invention will be described below in conjunction with the accompanying drawings.

Please refer to Fig. 1, which shows a flow chart of a component allocation model predictive control method for chemical production provided by an embodiment of the present invention, the method includes:

Step S1: Collect the set temperature value, steam temperature and heating element temperature of each distillation point in the distillation process; obtain the mixed code of the appearance characteristics of the reactants in the container; according to the steam temperature and heating element temperature in the sampling process and the volumetric space velocity of the container to construct a short-term reaction descriptor.

In the embodiment of the present invention, CIO heavy aromatics and CIO tail oil after cutting and extracting light components are mixed according to the volume ratio of 2:1. Take the hydrogenation and lightening component preparation production as an example, for the component preparation, that is, the distribution hydrogenation process. In the embodiment of the present invention, its hydrogen pressure is 5MPa. Taking a certain catalytic process as an example, for the entire distillation process, the temperature of the steam that is evaporated must be controlled in real time, and the nodes to be adjusted are respectively 0% of the total volume (initial distillation range), 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 95% and final boiling point (dry point), that is, in the embodiment of the present invention, a distillation process exists 12 distillation range points. In the embodiment of the present invention, for the initial distillation range, the set temperature value is preset to 98 degrees Celsius, and so on, generally the temperature is getting higher and higher, which is not limited here, that is, each distillation range point corresponds to a set temperature value.

The heating and control methods of the distillation process under different reaction parameters are different, and the degree of hydrogenation is different. Therefore, it is necessary to measure the temperature information in the container under the distillation process in real time. The specific temperature information in the container includes the steam temperature and the temperature of the heating element. Because there is a certain numerical hysteresis between the steam temperature, that is, there is a joint effect of heat absorption and specific heat capacity in the data between the heating element temperature of the reaction vessel and the steam temperature, and there is a certain hysteresis, that is, temperature damping. This feature is related to the characteristics of heavy aromatics raw materials, The rate of the hydrogenation reaction is related to the internal characteristics of the two-phase substance, so when collecting temperature information, two temperature information, the steam temperature and the heating element temperature, need to be collected.

In the embodiment of the present invention, the steam temperature T_A and the heating element temperature T_B in the reaction vessel were recorded at a rate of 1 Hz.

In the embodiment of the present invention, for each distillation range point, there are several recorded steam temperature values, and nearest neighbor resampling is performed on these values to ensure that there are 400 data between each distillation range point. That is, the data of each distillation point is processed as a fixed-length temperature record value, and this data is two data fixed-length sequences of T_A and T_B. That is, each temperature sequence corresponding to each distillation point is a sequence composed of 400 data point information.

During the distillation process, due to the influence of reaction parameters at each distillation point, the reactants in the container will have different morphological characteristics, such as the color and transparency of the reactants. Therefore, in order to further describe the distillation process, it is necessary to obtain the mixed codes of the appearance characteristics of the reactants in the container at each distillation point in the distillation process, and use the mixed codes of appearance characteristics to characterize the morphological information of the reactants at each distillation point, including:

Analyze the physical property data, obtain the description words of the appearance characteristics of the reactants in the container, and express them through conventional words: the appearance words can be yellowish, clear, etc. In order to maintain the unity of words in the industry and realize the disambiguation of words, in the implementation of the present invention In the example, English words are used. For example, in Bright & Clear, Bright and Clear are segmented to obtain TF-IDF appearance codes based on word segmentation in the thesaurus. It should be noted that the dimension of the appearance coding is related to the size of the lexicon of the segmented words. The specific coding method is a technical means well known to those skilled in the art, and will not be repeated here; the method of obtaining the description words of the appearance characteristics of the reactants in the container can be obtained through Visual inspection or neural network acquisition is not limited here, and the acquisition method can be freely selected according to the specific implementation scenario.

The color information of the reactants in the container is further collected, and the appearance color coding is constructed with the component values in the RGB color space. Please refer to Fig. 2, which shows a schematic diagram of a color information acquisition process provided by an embodiment of the present invention. The embodiment of the present invention uses the ADJD-S313-QR999RGB component chromaticity sensor, under constant light source brightness The RGB value of the appearance is measured below, and the bypass can be built in the reaction pipeline or the observation window can be directly installed to measure. Wherein, this embodiment uses the CMOS digital color sensor of avago company model as adjd-s313-qr999, light source 1 provides backlight, obtains the color reference of transmission, light source 2 provides diffuse reflection color reference, after RGB chromaticity measurement, based on the sensor's maximum The range is three floating-point values normalized to 0~1 for the RGB components, that is, the appearance color coding of each distillation point is coded by the normalized data of the three color channel components.

Further merge the TF-IDF appearance coding and appearance color coding to obtain a mixed coding of appearance features, including:

a. First, dimensionally expand the multiple normalized RGB components in the appearance color code C to obtain the first extended code H_0 of a 12*3-dimensional high-dimensional vector.

b. Then merge it with the text TF-IDF appearance code to obtain the second extended code H_1 of a high-dimensional vector of 12*3+X dimensions, where X is the number of dimensions of the TF-IDF appearance code.

c. Further reduce the dimensionality of the second extended code to the preset dimension to obtain a mixed code of appearance features. In the embodiment of the present invention, a high-dimensional space is constructed based on the high-dimensional vector H_1 in each production record, and the dimension is reduced to 10 dimensions based on the PCA algorithm to obtain a mixed code of appearance features. That is, the preset dimension in the embodiment of the present invention is 10 dimensions.

i. Specifically, from an empirical point of view, 10 dimensions is the number of dimensions with acceptable accuracy. The implementer can continue to increase the dimension based on this number of dimensions, so as to ensure the accuracy of the data. The vector after dimension reduction is mainly used to represent the uncontrolled time Physical properties and appearance data of each evaporation process.

ii. So far, based on the previous data, the appearance feature hybrid code H after dimensionality reduction is obtained.

The role of the appearance feature mixed code H is to jointly express the subjective information and objective information for this distillation process, so as to avoid the subjective error of manual analysis, and at the same time improve the accuracy and objectivity of the data in the storage process.

According to each production data, short-term reaction descriptors are constructed according to the steam temperature, heating element temperature and volume space velocity of the container under the sampling process, including:

Taking the hydrogen pressure of 6MPa as an example, during the production process, the proportions of the catalyst will be fine-tuned, and the steaming temperature will also be fine-tuned. Correspondingly, taking the hydrogen-oil volume ratio of 1000 as an example, the volumetric space velocity V may be between 1.8 and 1.5 Between, the distillation temperature of each distillation range point will have different target values. In general, the higher the V, the higher the assumed catalyst activity, and the greater the processing capacity of the corresponding device. It should be noted that any reaction vessel and reaction principle have marginal effects, and V cannot be increased infinitely. For the measured container device, a relatively large V means that there are more raw materials passing through the catalyst per unit time, and the residence time of raw materials on the catalyst is short. The reaction depth is shallow; on the contrary, a small V means a long reaction time, and generally lowering V is beneficial to improving the conversion rate of the reaction. However, too low V means that the amount of catalyst required under the same treatment amount is large, and in the distillation range, it is economically unreasonable to reduce the reaction efficiency in a disguised manner. Therefore, V represents the experience of fine-tuning the catalyst when it is put into production at this time, and it is not necessarily the best parameter. Therefore, temperature parameters and volumetric space velocity V are influencing factors for the contextual characteristics of short-term reactions. As for V, since it has already been put into production, the recorded V is generally adjusted through fine-tuning and experience. Therefore, a short-term response descriptor is constructed according to the steam temperature, heating element temperature, and container volume space velocity during the sampling process, including:

，其中

为蒸汽温度序列中值，

为加热元件温度序列中值，

为蒸汽温度序列均值，

为加热元件温度序列均值，

为蒸汽温度序列初始值，

为加热元件温度序列初始值。The short-term reaction descriptor consists of the median value of the steam temperature series, the median value of the heating element temperature series, the mean value of the steam temperature series, the mean value of the heating element temperature series, the initial value of the steam temperature series and the initial value of the heating element temperature series at each distillation point in the distillation process. value composition. That is, the short-term response descriptor is a vector composed of multiple characteristic parameters, and its mathematical expression is

,in

is the median value of the steam temperature series,

is the median value of the heating element temperature sequence,

is the mean value of the steam temperature series,

is the mean value of the heating element temperature sequence,

is the initial value of the steam temperature series,

It is the initial value of the heating element temperature sequence.

Step S2: Match different distillation range points according to the mixed coding difference of appearance characteristics and the difference of short-term response descriptors between each distillation range point in the distillation process under different control parameters, and obtain multiple matching pairs.

First of all, different volume space velocity V, catalyst adjustment and temperature adjustment will produce different post-production data. Therefore, for the short-term reaction mode descriptor Q and the color reference value C of the distillation point, it can be searched based on the K-M operator according to The following strategy is used for type allocation: search for the similarities and differences and matching distances between each short-term reaction mode and the reaction mode of the next reaction distillation range, and can construct model parameters for the degree of subsequent control intervention and recommended temperature values. That is, the reaction process of each distillation point is regarded as a short-term reaction process, and the distillation points in the distillation process under different reaction parameters are matched. For example, the second distillation point in reaction mode A is used as the target distillation point, and the target distillation point is matched with all distillation points in reaction mode B, reaction mode D and other reaction modes.

The matching distance is obtained according to the matching distance formula, and the matching distance formula includes:

为馏程点X和馏程点Y之间的匹配距离，

为馏程点X和馏程点Y之间的短期反应描述子相似度，

为馏程点X和馏程点Y之间的外观特征混合编码相似度，

为馏程点X和馏程点Y在整个蒸馏过程中对应的馏程点序号的差异绝对值；若

，则

为无穷大。in,

is the matching distance between distillation point X and distillation point Y,

is the short-term response descriptor similarity between distillation point X and distillation point Y,

Coding similarity for the mixture of appearance features between distillation point X and distillation point Y,

is the absolute value of the difference between the distillation point numbers corresponding to the distillation point X and the distillation point Y in the whole distillation process; if

,but

for infinity.

表示了两种数据的整体差异，进一步引入馏程点序号的差异，使得后续匹配过程能够保证将不同馏程点序号的馏程点进行匹配的同时，还保证了匹配的两个馏程点的馏程点序号差异不会过大，增加后续预测数据的参考性。In the matching distance formula, the short-term response descriptor similarity is multiplied by the appearance feature mixed coding similarity and then subtracted by the value 1, that is

Indicates the overall difference between the two data, and further introduces the difference in the number of the distillation point, so that the subsequent matching process can ensure that the distillation points with different numbers of the distillation point are matched, and at the same time ensure the matching of the two distillation points. The difference in the serial number of the distillation procedure point will not be too large, which increases the reference of subsequent prediction data.

为两个短期反应描述子之间的余弦相似度，

为两个外观特征混合编码之间的余弦相似度。需要说明的是，余弦相似度为本领域技术人员熟知的现有技术，在此以短期反应描述子之间的余弦相似度为例，列出其余弦相似度的表达式：In the embodiment of the present invention,

is the cosine similarity between two short-term response descriptors,

Cosine similarity between two appearance feature mixture encodings. It should be noted that the cosine similarity is an existing technology well known to those skilled in the art. Taking the cosine similarity between short-term response descriptors as an example, the expression of the cosine similarity is listed:

为短期反应描述子

中的第

个元素，

为短期反应描述子

中的第

个元素。in,

short-term response descriptor

in the first

elements,

short-term response descriptor

in the first

elements.

Under the distance constraint of the above D formula, the K-M algorithm will search for several matching pairs, and most of the matching results between the matching pairs are 1 difference between the distillation range nodes, for example, between 10% and 20%; and In the short-term response mode, the temperature characteristics and the parameters of V are approximate, and the appearance at this time is similar.

Step S3: Obtain the switchable index according to the set temperature value difference, the steam temperature difference and the heating element temperature difference between the two distillation range points in the matching pair; select the data in the preset number of matching pairs with the largest switchable index as For the initial training set, the initial training set is augmented according to the matching distillation point data in the initial training set to obtain a training set, and the long-short-term memory production group is trained according to the training set to prepare a prediction model.

The distillation range points in a matched pair can represent the temperature values of the next distillation range that can be matched under different or as similar as possible parameters. Based on these temperature values and the similarity of the temperature information of each distillation range point, a more suitable reference matching pair is selected among the matching pairs:

Define a switchable index for each matching pair, which represents the degree of sustainable delayed distillation between the upper and lower distillation ranges. If the index is larger, it is considered that there are more fractions that can be produced in this case, that is A switchable index is obtained from the difference in set temperature values, vapor temperature difference and heating element temperature difference in two distillation range points in a matched pair, including:

The switchable index is obtained according to the switchable index formula, which includes:

为馏程点X和馏程点Y之间的可切换指数，

为馏程点X和馏程点Y之间的设定温度值的差值绝对值，

为馏程点X和馏程点Y之间的匹配距离，

为馏程点X和馏程点Y之间蒸汽温度序列的动态时间规整距离，

为馏程点X和馏程点Y之间加热元件温度序列的动态时间规整距离。in,

is the switchable index between distillation point X and distillation point Y,

is the absolute value of the difference between the set temperature values between the distillation point X and the distillation point Y,

is the matching distance between distillation point X and distillation point Y,

is the dynamic time-regularized distance of the steam temperature sequence between distillation point X and distillation point Y,

is the dynamic time warping distance of the temperature sequence of the heating element between the distillation point X and the distillation point Y.

In the switchable index formula, the denominator is the sum of the two distances, that is, the larger the distance, the greater the difference between the two distillation points, and the smaller the switchable index; the absolute value of the difference between the numerator and the set temperature value matches The reciprocal of the distance product, that is, the smaller the absolute value of the difference and the smaller the matching distance, it means that the closer the two distillation points are, the larger the switchable index is.

The data in the preset number of matching pairs with the largest switchable index is selected as the initial training set. In the embodiment of the present invention, the top 25% of the matching pairs with the switchable index are selected to construct the initial training set.

For the top 25% of the matching pairs, any matching pair has a sequence of distillation ranges, and the initial training set is augmented with the postpartum data corresponding to the previous distillation point, that is According to the matching distillation point data in the initial training set, the initial training set is augmented to obtain the training set. The specific augmentation methods include:

Replace the volume space velocity, set temperature, vapor temperature, and heating element temperature of the target distillation point with the volume space velocity, set temperature, vapor temperature, and heating element temperature of the corresponding distillation point in the distillation process to which the matching distillation point belongs , to obtain the first augmented training data; replace the volumetric space velocity, set temperature, steam temperature and heating element temperature of the distillation point next to the target distillation point with the corresponding distillation point in the distillation process to which the matching distillation point belongs The volume space velocity, set temperature, steam temperature and heating element temperature are obtained to obtain the second augmented training data; respectively, the volume space velocity, set temperature, and steam temperature of the target distillation point and the next distillation point and the temperature of the heating element are simultaneously replaced with the volume space velocity, set temperature, steam temperature and temperature of the heating element in the vessel corresponding to the distillation process to which the distillation point belongs to obtain the third augmented training data; The first augmented training data, the second augmented training data and the third augmented training data of the process point are obtained to obtain a training set.

For example, for the target distillation point a, there is a distillation process A to which it belongs, and the number of the distillation point in the distillation process A is i; the target distillation point a corresponds to a matching distillation point b, and the matching distillation Point b belongs to distillation process B. Then the first augmented training data of the target distillation point a is replaced by the distillation process point number i in the distillation process B; the second augmented training data is the distillation process point number i in the distillation process A The distillation point data of i+1 is replaced by the distillation process point number i+1 in the distillation process B; the third augmented training data is the distillation process point number i in the distillation process A The point data is replaced by the distillation process point number i in distillation process B, and the distillation process point number i+1 in distillation process A is replaced by the distillation process point number i in distillation process B +1 for the distillation point data.

Based on the matching effect of K-M, based on the short-term response mode, the temperature characteristics and the parameters of V are approximate, and the appearance is similar to the postpartum data as the data source. The fraction that can be produced in this case by augmentation is more control model training data. The long-short-term memory production group allocation prediction model can be based on Batch Size, and based on several augmented data for error adjustment, so as to obtain better prediction results that match the postpartum data. For the training set, based on T_A, T_B of the previous distillation range, the label is set to T_C and space velocity V of the next distillation range.

Step S4: Input the real-time steam temperature and the real-time heating element temperature at the real-time distillation point in the real-time distillation process into the long-short-term memory production component allocation prediction model to obtain the set temperature value and volume space velocity in the next distillation process, and The vessel parameters are controlled according to the set temperature value and the volume space velocity in the next distillation process.

So far, in the real-time production process, based on the real-time steam temperature and real-time heating element temperature at the real-time distillation point in a real-time distillation process, after resampling, input the long-short-term memory production component allocation prediction model to obtain the next distillation The set temperature value and volume space velocity corresponding to the process point can be set manually by the staff or automatically set by the distillation vessel, so as to achieve the predictive optimization control effect and increase the distillate output of the distillation range.

To sum up, the embodiments of the present invention obtain the reaction characteristics of each distillation point, and match the distillation points in the distillation process under different control parameters according to the reaction characteristics to obtain multiple matching pairs. Further, the switchable index is obtained according to the reaction characteristic difference between the internal distillation points of the matching pair, and the matching pair is selected based on the switching index to construct the initial training set, and the initial training set is augmented according to the matching data to obtain the training set for training The long-short-term memory production group configures the prediction model, and then realizes the prediction of real-time data, and controls the container parameters according to the prediction parameters. In the embodiment of the present invention, the production efficiency is ensured by constructing a component allocation model to predict the parameters that need to be adjusted at future distillation range points.

It should be noted that: the order of the above embodiments of the present invention is only for description, and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. Multitasking and parallel processing are also possible or may be advantageous in certain embodiments.

Each embodiment in this specification is described in a progressive manner, the same and similar parts of each embodiment can be referred to each other, and each embodiment focuses on the differences from other embodiments.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within range.

Claims (8)

1. A method for predictive control of a component dispensing model for chemical production, the method comprising:

collecting the set temperature value, the steam temperature and the temperature of a heating element in a container of each distillation range point in the distillation process; obtaining appearance characteristic mixed codes of reactants in the container; constructing a short-term reaction descriptor according to the steam temperature, the heating element temperature and the volume space velocity of the container in the sampling process;

matching different distillation range points according to appearance characteristic mixed coding difference and short-term reaction descriptor difference between the distillation range points in the distillation process under different control parameters to obtain a plurality of matched pairs;

obtaining a switchable index based on a set temperature value difference between two boiling point values in the matched pair, a steam temperature difference, and a heating element temperature difference; selecting data in a preset number of matching pairs with the largest switchable indexes as an initial training set, amplifying the initial training set according to mutually matched distillation range data in the initial training set to obtain a training set, and training a long-term and short-term memory production component system prediction model according to the training set;

inputting the real-time steam temperature of the real-time distillation range point and the real-time heating element temperature in the real-time distillation process into the long-short term memory production component preparation prediction model to obtain the set temperature value and the volume airspeed corresponding to the next distillation range point, and controlling the container parameters according to the set temperature value and the volume airspeed corresponding to the next distillation range point.

2. The method as claimed in claim 1, wherein the obtaining of the appearance feature mixture codes of the reagents in the containers comprises:

obtaining descriptive words of appearance characteristics of reactants in the container, and constructing TF-IDF appearance codes according to the descriptive words; collecting color information of the container reactant, and constructing an appearance color code according to component values in an RGB color space; and combining the TF-IDF appearance coding and the appearance color coding to obtain the appearance characteristic mixed coding.

3. The method as claimed in claim 2, wherein the obtaining of the appearance feature hybrid coding comprises:

expanding the dimension of the appearance color code to obtain a first expanded code; merging the first extended code and the TF-IDF appearance code to obtain a second extended code; and reducing the dimension of the second extended code to a preset dimension to obtain an appearance characteristic mixed code.

4. The method as claimed in claim 1, wherein the step of constructing the short-term response descriptor according to the steam temperature, the heating element temperature and the volume space velocity of the container during sampling comprises:

the short-term reaction descriptor is composed of a steam temperature sequence median value, a heating element temperature sequence median value, a steam temperature sequence mean value, a heating element temperature sequence mean value, a steam temperature sequence initial value and a heating element temperature sequence initial value at each distillation range point in the distillation process.

5. The method of claim 1, wherein the step of obtaining a plurality of matching pairs comprises:

constructing a matching distance according to appearance characteristic mixed coding difference and short-term reaction descriptor difference between different distillation processes under different reaction parameters; and matching different distillation processes by utilizing a KM matching algorithm according to the matching distance to obtain a plurality of matching pairs.

6. The method as claimed in claim 5, wherein the step of constructing the matching distance according to the appearance feature mixed coding difference and short-term response descriptor difference between different distillation processes under different response parameters comprises:

obtaining a matching distance according to a matching distance formula, wherein the matching distance formula comprises:

wherein,

is the matching distance between the boiling point X and the boiling point Y,

short term reaction descriptor similarity between boiling point X and boiling point Y,

the similarity is coded for the appearance feature mixture between the boiling point X and the boiling point Y,

the absolute value of the difference of the corresponding distillation range point serial numbers of the distillation range point X and the distillation range point Y in the whole distillation process is shown; if it is

Then, then

Is infinite.

7. The method of claim 1, wherein the obtaining the switchable index according to the set temperature value difference, the steam temperature difference and the heating element temperature difference in the two distillation processes in the matching pair comprises:

obtaining a switchable index according to a switchable index formula, the switchable index formula comprising:

wherein,

is a switchable index between the boiling point X and the boiling point Y,

is the absolute value of the difference between the set temperature values between the distillation range point X and the distillation range point Y,

is the matching distance between boiling point X and boiling point Y,

is the dynamic time-warping distance of the steam temperature sequence between the distillation range point X and the distillation range point Y,

is the dynamic time-warping distance of the heating element temperature sequence between boiling point X and boiling point Y.

8. The method of claim 1, wherein the augmenting the initial training set based on distillation process data that match each other in the initial training set comprises:

replacing the volume airspeed, the set temperature, the steam temperature and the heating element temperature of the target distillation range point with the volume airspeed, the set temperature, the steam temperature and the heating element temperature of the corresponding distillation range point in the distillation process to which the matched distillation range point belongs to obtain first augmentation training data;

replacing the volume airspeed, the set temperature, the steam temperature and the heating element temperature of the next distillation range point of the target distillation range point with the volume airspeed, the set temperature, the steam temperature and the heating element temperature of the corresponding distillation range point in the distillation process to which the matched distillation range point belongs to obtain second augmentation training data;

respectively replacing the volume airspeed, the set temperature, the steam temperature and the heating element temperature of the target distillation range point and the next distillation range point of the target distillation range point with the volume airspeed, the set temperature, the steam temperature and the heating element temperature of the corresponding distillation range point in the distillation process to which the matched distillation range point belongs, and obtaining third augmentation training data;

and combining the first augmented training data, the second augmented training data and the third augmented training data of each distillation range point to obtain a training set.

Images