Production process and reaction time real-time prediction method of citric acid type polycarbonate polyether polyol

Abstract

The invention relates to the technical field of chemistry and chemical engineering, and particularly discloses a production process of citric acid type polycarbonate polyether polyol and a real-time prediction method of reaction time, the production process is carried out in a tube nest reaction device, citric acid is prevented from passivating an activation center of a catalyst in a partitioned premixing mode, and the reaction time of the citric acid type polycarbonate polyether polyol is shortened. And performing real-time monitoring and fine-grained time sequence analysis on the density of the material by adopting an artificial intelligence technology in the reaction process, so that the residual reaction time is intelligently predicted based on the time sequence dynamic correlation change characteristic of the density of the material. In this way, the reaction endpoint can be predicted in advance, and over-reaction or incomplete reaction can be effectively avoided, so that the production efficiency and the product quality are improved, and the method is suitable for industrial-scale production.

Description

Technical Field

[0001] The present application relates to the field of chemical engineering technology, and more specifically, to a production process of citric acid type polycarbonate polyether polyol and a real-time prediction method for reaction time. Background Art

[0002] Polycarbonate polyether polyols are a type of polyols with carbonate groups in the molecule and hydroxyl groups at the ends of the molecular chain. Double metal cyanide (abbreviated as DMC) catalysts are widely used in the synthesis process of polycarbonate polyether polyols because of their simple preparation, low cost, and residual heavy metals in the product below 30ppm. Carbon dioxide, one of the raw materials, is cheap and easy to obtain, non-toxic and non-flammable, has a clear carbon dioxide chemical fixation effect, and the temperature required for the reaction is not high, and the energy consumption required is low, which makes the carbon dioxide and epoxy compound adjustment copolymerization method have broad application prospects and high industrial value.

[0003] When citric acid is used as an initiator to prepare polycarbonate polyether polyols, the properties of citric acid will cause the DMC catalyst to be deactivated, thereby slowing down the polymerization reaction rate or even preventing it from proceeding, resulting in the product's functionality being far below the theoretical value, a decrease in molecular weight, and an increase in by-products. Specifically, citric acid contains three primary hydroxyl groups and one tertiary hydroxyl group. The primary hydroxyl group has strong reactivity but is unable to react completely due to catalyst deactivation; the tertiary hydroxyl group has weak reactivity and is only partially involved in the late stage of the reaction. Due to catalyst deactivation, the catalytic activity is low, resulting in the final product functionality being only 3 or less than 3 at most.

[0004] In addition, in the production process of citric acid-based polycarbonate polyether polyol, as the reaction proceeds, the conversion rate of propylene oxide gradually increases, and the molecular weight of the generated polycarbonate polyether polyol also increases, which will cause the overall density of the material to change. When the reaction approaches the end point, the generation rate of new polycarbonate polyether polyol slows down and the density of the material tends to be stable. Therefore, by monitoring the change in material density, the progress of the reaction can be indirectly understood, and it can be determined when the predetermined conversion rate and molecular weight are reached, so as to accurately judge whether the reaction has been completed.

[0005] However, traditional methods mainly rely on a fixed density threshold to determine the reaction endpoint. However, in the actual production process, due to the influence of various factors such as fluctuations in reaction conditions and changes in equipment performance, the reaction rate and the molecular weight distribution of the product will fluctuate, causing the density of the final material to change accordingly, resulting in a decrease in the accuracy of relying on a fixed density threshold to determine the reaction endpoint.

[0006] Therefore, an optimized production process of citric acid-based polycarbonate polyether polyols and a real-time prediction method for reaction time are desired. Summary of the invention

[0007] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide a novel production process of citric acid type polycarbonate polyether polyol and a real-time prediction method for reaction time. The production process is carried out in a tube-in-tube reaction device, which uses a partitioned premixing form to prevent citric acid from passivating the catalyst activation center, and uses artificial intelligence technology to monitor the density of the material in real time and perform fine-grained time series analysis during the reaction process, thereby intelligently predicting the remaining reaction time based on the time series dynamic correlation change characteristics of the material density. In this way, the reaction endpoint can be predicted in advance, effectively avoiding the occurrence of overreaction or incomplete reaction, thereby improving production efficiency and product quality, and being suitable for industrial-scale production.

[0008] Accordingly, according to one aspect of the present application, a production process of citric acid type polycarbonate polyether polyol is provided, wherein the production process of citric acid type polycarbonate polyether polyol is carried out in a tubular reactor, wherein the tubular reactor comprises a reaction pipe having a gas-liquid mass transfer zone, a static mixer and a circulation device arranged in the reaction pipe, wherein the circulation device comprises a circulation pump and a compression pump; wherein the production process comprises: step (1): fully premixing propylene oxide, citric acid and DMC catalyst in a premixing kettle to obtain a premix; step (2): the premix enters the tubular reactor for reaction, the tubular reactor adopts a gas-liquid dual circulation method for material circulation reaction, the reaction temperature is 80°C±5°C, and one end of the feed inlet of the tubular reactor is provided with one or more dual-flow atomizing nozzles for mixing and atomizing the circulating gas and liquid materials and dispersing them in the gas-liquid mass transfer zone for mass transfer. and heat transfer, the material flow rate is 5-10min to complete one cycle, and the reaction endpoint is determined by the density of the material; step (3): after the reaction is completed, the material is discharged and sequentially passed through a primary separation device to remove carbon dioxide and propylene oxide, a catalyst separation device to separate the catalyst, and a secondary separation device to remove by-products to obtain the citric acid type polycarbonate polyether polyol; wherein the step (2) also includes: obtaining a time queue of the density of the material collected by a densitometer; performing a time series feature extraction based on a local time scale on the time queue of the density of the material to obtain a sequence of local time series implicit fluctuation feature vectors of the material density; performing a feature propagation aggregation based on spatiotemporal significance on the sequence of local time series implicit fluctuation feature vectors of the material density to obtain a spatiotemporal significant aggregation propagation representation vector of the material density; and generating a predicted decoding value of the remaining reaction time based on the spatiotemporal significant aggregation propagation representation vector of the material density.

[0009] In the production process of the citric acid type polycarbonate polyether polyol, a time series feature extraction based on a local time scale is performed on the time queue of the density of the material to obtain a sequence of local time series implicit fluctuation feature vectors of the material density, including: arranging the time queue of the density of the material into a material density time series data set according to the time dimension; performing data set segmentation on the material density time series data set to obtain a sequence of material density local data sets; and inputting each material density local data set in the sequence of the material density local data set into a material density time series feature extractor based on a Bi-LSTM model to obtain a sequence of local time series implicit fluctuation feature vectors of the material density.

[0010] In the production process of the citric acid type polycarbonate polyether polyol, the sequence of the local time series implicit fluctuation feature vectors of the material density is subjected to feature propagation aggregation based on spatiotemporal significance to obtain the material density spatiotemporal significant aggregation propagation representation vector, including: extracting the current material density local time series implicit fluctuation feature vector from the sequence of the local time series implicit fluctuation feature vectors of the material density, and defining other material density local time series implicit fluctuation feature vectors as historical material density local time series implicit fluctuation feature vectors to obtain a sequence of historical material density local time series implicit fluctuation feature vectors; based on the historical material density local time series implicit fluctuation feature vectors The spatial span and time span between each historical material density local time series implicit fluctuation characteristic vector in the sequence of characteristic vectors and the current material density local time series implicit fluctuation characteristic vector are used to modulate the static energy value of each historical material density local time series implicit fluctuation characteristic vector to obtain the dynamic energy spatiotemporal transfer value of each historical material density local time series implicit fluctuation characteristic vector; based on the dynamic energy spatiotemporal transfer value of each historical material density local time series implicit fluctuation characteristic vector, the sequence of material density local time series implicit fluctuation characteristic vectors is aggregated to obtain the material density spatiotemporal significant aggregation propagation representation vector.

[0011] In the production process of the citric acid type polycarbonate polyether polyol, based on the spatial span and time span between each historical material density local time series implicit fluctuation characteristic vector in the sequence of the historical material density local time series implicit fluctuation characteristic vector and the current material density local time series implicit fluctuation characteristic vector, the static energy value of each historical material density local time series implicit fluctuation characteristic vector is modulated to obtain the dynamic energy space-time transfer value of each historical material density local time series implicit fluctuation characteristic vector, including: calculating each historical material density local time series implicit fluctuation characteristic vector in the sequence The static energy value of each historical material density local time series implicit fluctuation characteristic vector; based on the static energy value of each historical material density local time series implicit fluctuation characteristic vector, calculate the energy space attenuation factor and energy time attenuation factor of each historical material density local time series implicit fluctuation characteristic vector relative to the current material density local time series implicit fluctuation characteristic vector; calculate the weighted sum of the energy space attenuation factor and energy time attenuation factor of each historical material density local time series implicit fluctuation characteristic vector to obtain the dynamic energy space-time transfer value of each historical material density local time series implicit fluctuation characteristic vector.

[0012] In the production process of the above-mentioned citric acid type polycarbonate polyether polyol, the static energy value of each historical material density local time series implicit fluctuation characteristic vector in the sequence of the historical material density local time series implicit fluctuation characteristic vector is calculated, including: calculating the expected value of the fourth power of the difference between each eigenvalue in the historical material density local time series implicit fluctuation characteristic vector and its eigenmean, and dividing the expected value by the square of the characteristic variance of the historical material density local time series implicit fluctuation characteristic vector to obtain the kurtosis of the historical material density local time series implicit fluctuation characteristic vector; calculating the kurtosis of the historical material density local time series implicit fluctuation characteristic vector and the sum of its maximum eigenvalue to obtain the static energy value.

[0013] In the production process of the citric acid type polycarbonate polyether polyol, based on the static energy value of each historical material density local time series implicit fluctuation characteristic vector, the energy space attenuation factor and energy time attenuation factor of each historical material density local time series implicit fluctuation characteristic vector relative to the current material density local time series implicit fluctuation characteristic vector are calculated, including: taking the square value of the number of characteristic vectors between the historical material density local time series implicit fluctuation characteristic vector and the current material density local time series implicit fluctuation characteristic vector as the space span coefficient, calculating the static energy value of the historical material density local time series implicit fluctuation characteristic vector and dividing it by the space span coefficient to obtain the energy space attenuation factor; calculating the timestamp difference between the historical material density local time series implicit fluctuation characteristic vector and the current material density local time series implicit fluctuation characteristic vector, rounding down the timestamp difference and dividing it by the preset time attenuation period to obtain the time span coefficient, and dividing the static energy value of the historical material density local time series implicit fluctuation characteristic vector by the natural exponential function value of the time span coefficient to obtain the energy time attenuation factor.

[0014] In the production process of the above-mentioned citric acid type polycarbonate polyether polyol, based on the dynamic energy spatiotemporal transfer value of the local time series implicit fluctuation characteristic vector of each historical material density, the sequence of the local time series implicit fluctuation characteristic vector of the material density is aggregated to obtain the material density spatiotemporal significant aggregation propagation representation vector, including: inputting the dynamic energy spatiotemporal transfer value of the local time series implicit fluctuation characteristic vector of the historical material density into the energy transfer gating module based on the mask function to obtain a sequence of dynamic energy spatiotemporal transfer weights; based on the sequence of dynamic energy spatiotemporal transfer weights, calculating the weighted sum of the sequence of the local time series implicit fluctuation characteristic vector of the historical material density to obtain the historical material density local time series feature transfer aggregation modulation vector; calculating the positional sum of the local time series feature transfer aggregation modulation vector of the historical material density and the local time series implicit fluctuation characteristic vector of the current material density to obtain the material density spatiotemporal significant aggregation propagation representation vector.

[0015] In the production process of the above-mentioned citric acid type polycarbonate polyether polyol, a predicted decoding value of the remaining reaction time is generated based on the material density spatiotemporal significant aggregation propagation representation vector, including: inputting the material density spatiotemporal significant aggregation propagation representation vector into a decoder-based reaction predictor to obtain the predicted decoding value of the remaining reaction time.

[0016] In the above-mentioned production process of citric acid type polycarbonate polyether polyol, the double-flow atomizing nozzle has a liquid material atomizing feed port, a carbon dioxide circulation air inlet, a mixing chamber air injection hole, a mixing chamber and a double-flow atomizing nozzle, wherein the liquid material enters from the liquid material atomizing feed port, CO2 enters from the carbon dioxide circulation air inlet, the liquid material and CO2 enter the mixing chamber through the mixing chamber air injection hole to be fully mixed and then are sprayed out by the double-flow atomizing nozzle.

[0017] According to another aspect of the present application, a real-time prediction method for reaction time is provided, which includes: obtaining a time queue of material density collected by a densitometer; performing time series feature extraction based on a local time scale on the time queue of material density to obtain a sequence of local time series implicit fluctuation feature vectors of material density; performing feature propagation aggregation based on spatiotemporal significance on the sequence of local time series implicit fluctuation feature vectors of material density to obtain a spatiotemporal significant aggregation propagation representation vector of material density; and generating a predicted decoding value of the remaining reaction time based on the spatiotemporal significant aggregation propagation representation vector of material density.

[0018] Compared with the prior art, the production process and real-time prediction method of reaction time of citric acid type polycarbonate polyether polyol described in this application are carried out in a tube-and-tube reaction device, which uses a partitioned premixing form to prevent citric acid from passivating the catalyst activation center, and uses artificial intelligence technology to monitor the density of the material in real time and perform fine-grained time series analysis during the reaction process, thereby intelligently predicting the remaining reaction time based on the time series dynamic correlation change characteristics of the material density. In this way, the reaction endpoint can be predicted in advance, effectively avoiding the occurrence of over-reaction or incomplete reaction, thereby improving production efficiency and product quality, and is suitable for industrial-scale production. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0020] Figure 1 Schematic diagram of the production process of citric acid-based polycarbonate polyether polyol according to an embodiment of the present application.

[0021] Figure 2 This is a flow chart of step (2) in the production process of citric acid-based polycarbonate polyether polyol according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0023] Figure 1 The present invention is a schematic diagram of the production process of citric acid type polycarbonate polyether polyol according to an embodiment of the present application. The present application proposes a production process of citric acid type polycarbonate polyether polyol, wherein the production process comprises: step (1): fully premixing propylene oxide, citric acid and DMC catalyst in a premixing kettle to obtain a premix; step (2): the premix enters the tubular reactor for reaction, the tubular reactor adopts a gas-liquid dual circulation method to carry out a material circulation reaction, the reaction temperature is 80°C±5°C, one end of the feed inlet of the tubular reactor is provided with one or more double-flow atomizing nozzles for mixing and atomizing the circulating gas and liquid materials and dispersing them in the gas-liquid mass transfer zone for mass transfer and heat transfer, the material flow rate is 5-10min to complete a cycle, and the reaction end point is judged by the density of the material; step (3): after the reaction is completed, the material is discharged, and the carbon dioxide and propylene oxide are removed by a primary separation device, the catalyst is separated by a catalyst separation device, and the by-products are removed by a secondary separation device to obtain the citric acid type polycarbonate polyether polyol.

[0024] Specifically, the production process is carried out in a tube-in-tube reaction device, such as Figure 1 As shown, the purified raw materials are premixed to obtain a premix, the premix temperature is 0-60°C, the CO2 pressure in the premixing kettle is made 0.3Mpa through the carbon dioxide inlet, and the premix time is 1-4h; the DMC catalyst is prepared by the method described in Chinese patent CN115785435B.

[0025] In the production process of the present application, the premixing kettle is only for physical mixing to ensure sufficient dissolution and mixing of the reactants, and no polymerization reaction or catalyst activation reaction occurs. Furthermore, in the production process described in the present application, a step of raw material purification is also included before the raw material premixing.

[0026] The tubular reactor comprises a reaction pipeline with a gas-liquid mass transfer zone, a static mixer and a circulation device arranged in the reaction pipeline, the circulation device comprises a circulation pump and a compression pump, and one end of the feed port of the tubular reactor is provided with one or more double-flow atomizing nozzles. The premix enters the reaction pipeline with a temperature of 40°C and a carbon dioxide pressure of 0.5MPa through the feed port, and the circulation pump is turned on after 10% of the mass of the material enters. The flow rate of the circulation device is based on 5-10min to complete a cycle. At the same time, the compression pump is turned on to circulate the CO2 gas and liquid material to the top and pass through the double-flow atomizing nozzle together. The liquid material enters from the liquid material atomization feed port, and the CO2 enters from the carbon dioxide circulation air inlet, and then enters the mixing chamber through the mixing chamber injection hole in the double-flow atomizing nozzle to be fully mixed, and then the double-flow atomizing nozzle sprays droplets, the material is atomized and fully dispersed, the liquid surface area is increased, and the material is in the form of droplets in the gas-liquid mass transfer zone for gas-liquid mass transfer and heat conduction, thereby reducing the temperature of citric acid in the system to prevent thermal decomposition.

[0027] In one embodiment, in step (2), the feed volume should be lower than 70% of the pipeline volume, preferably 50%-70%; after the feeding is completed, the feed port is closed, and after the reaction starts, 20°C CO2 is slowly and continuously added through the carbon dioxide gas supply port according to the pressure sensor data, and the reaction pressure is maintained at 2-6Mpa. By adjusting the temperature of the heating medium in the insulation layer of the reaction pipeline, the temperature of the reactants in the reaction pipeline is kept at 80°C±5°C.

[0028] The reaction pipeline is provided with a gas-liquid mass transfer zone, and the raw materials are circulated and reacted in the tubular reactor in a gas-liquid dual circulation mode. The circulating pump allows the materials to complete a cycle in 5-10 minutes to ensure that the reaction proceeds while dissipating heat in time. At the same time, an SL static mixer is provided in the reaction pipeline to evenly disperse the solid catalyst into the epoxy compound, and at the same time disperse the generated reaction heat, reduce the points where the temperature is concentrated to generate high temperature, and reduce the temperature difference between the center temperature in the pipeline and the tube wall.

[0029] The reaction pipeline is provided with a liquid distributor, a connecting pipeline, a compression pump, and a double-flow atomizing nozzle at the feed port end, so that the material with a higher temperature and consumed carbon dioxide returns to the gas-liquid mass transfer zone to dissolve CO2, so that there is enough CO2 to participate in the reaction continuously, and heat exchange is performed in the gas-liquid mass transfer zone at the same time, ensuring that the material after generating heat in the synthesis process can release heat in time, and furthermore, the material in the reaction will not have a situation where more polyether products are generated due to the lower carbon dioxide content, and the gas-liquid mass transfer zone and the static mixer are fully transferred. The mass transfer and heat transfer, so that there will be no local heat center in the reaction system, resulting in the decomposition of citric acid, and the risk of the reaction system implosion is reduced. In addition, the reaction process will consume carbon dioxide all the time. After consuming carbon dioxide, carbon dioxide is continuously supplemented through the carbon dioxide gas replenishment port to maintain the reaction pressure. The addition of low-temperature carbon dioxide allows the heat of reaction to be effectively released here, thereby making the reaction system more stable. At the same time, during the reaction process, it is necessary to observe the data of the temperature sensor and dynamically adjust the temperature to maintain it at 80℃±5℃ to maintain the stability of the reaction. For example, when the material temperature exceeds 85℃, adjust the temperature of the heating medium in the insulation layer to 70℃. When the material temperature is lower than 83℃, gradually restore it to 80℃, and judge the reaction endpoint based on the density of the material.

[0030] In one embodiment, in step (2), the reaction is continued until the flow / density meter data reaches 1.1 g / cm 3 The reaction is considered to be complete and the product flows out through the discharge port.

[0031] In a preferred embodiment, in the step (2), considering that the conversion rate of propylene oxide gradually increases with the progress of the reaction, the molecular weight of the generated polycarbonate polyether polyol will also increase, which will cause the overall density of the material to change. When the reaction approaches the end point, the generation rate of new polycarbonate polyether polyol slows down and the density of the material tends to be stable. Therefore, by monitoring the change in the density of the material, the progress of the reaction can be indirectly understood, and it can be determined when the predetermined conversion rate and molecular weight are reached, so as to accurately judge whether the reaction has been completed. However, the traditional method mainly relies on a fixed density threshold to judge the reaction end point, but in the actual production process, due to the influence of various factors such as fluctuations in reaction conditions and changes in equipment performance, the reaction rate and the molecular weight distribution of the product will fluctuate, so that the density of the final material will also change accordingly, resulting in a decrease in the accuracy of relying on a fixed density threshold to judge the reaction end point. In order to improve the accuracy of the reaction end point judgment, the present application proposes a reaction end point prediction method based on density change trend, which uses artificial intelligence technology based on deep learning to perform real-time monitoring and time series analysis on the density of the material, and intelligently predicts the remaining reaction time based on the time series dynamic correlation change characteristics of the material density. In this way, the reaction endpoint can be effectively predicted in advance, thereby improving production efficiency and product quality.

[0032] Figure 2 FIG. 2 is a flow chart of step (2) in the production process of citric acid type polycarbonate polyether polyol according to an embodiment of the present application. Figure 2 As shown, the step (2) also includes: S210, obtaining a time queue of the density of the material collected by the densitometer; S220, performing time series feature extraction based on the local time scale on the time queue of the density of the material to obtain a sequence of local time series implicit fluctuation feature vectors of the material density; S230, performing feature propagation aggregation based on spatiotemporal significance on the sequence of local time series implicit fluctuation feature vectors of the material density to obtain a spatiotemporal significant aggregation propagation representation vector of the material density; S240, generating a predicted decoding value of the remaining reaction time based on the spatiotemporal significant aggregation propagation representation vector of the material density.

[0033] Based on this, in the technical solution of the present application, first, in the step S210, the density data of the material is collected in real time by a densitometer to obtain a time queue of the density of the material, thereby realizing continuous monitoring of the chemical reaction process and timely capturing changes in the properties of the material during the reaction process.

[0034] Then, in order to analyze the time series correlation pattern of the material density change in more detail, the present application further extracts the time series features of the material density time series based on the local time scale. Specifically, since the density of the material will change with the advancement of the chemical reaction, in order to accurately understand the trend and law of the material density change during the reaction, in the step S220, first, the time series of the material density is arranged as a material density time series data set according to the time dimension to maintain the time sequence of the data, thereby providing accurate time series information for subsequent data analysis. Then, the material density time series data set is segmented to obtain a sequence of material density local data sets; in this way, by dividing the entire reaction process into multiple time stages, the scale of the data time series analysis is refined, which helps to capture more subtle density changes during the reaction process, thereby improving the accuracy of the prediction. Next, in order to fully capture the time series fluctuation characteristics of material density in each local time domain, the material density time series feature extractor based on the Bi-LSTM model is further used to perform time series encoding on each material density local data set in the sequence of the material density local data set, so as to utilize the bidirectional memory capability of the Bi-LSTM model to fully capture the forward and backward time series context information of the material density change. Compared with the traditional LSTM (long short-term memory network), the Bi-LSTM model adds a back propagation layer in the hidden layer, so that it can simultaneously obtain time series correlation information from both the forward and reverse directions of the data set. This bidirectional information transmission mechanism helps to more comprehensively understand the dynamic correlation characteristics in the material density time series data in each local time domain, thereby extracting more accurate and rich feature representations to obtain a sequence of implicit fluctuation feature vectors of the local time series of material density.

[0035] Correspondingly, the step S220 includes: arranging the time queue of the density of the material into a material density time series data set according to the time dimension; performing data set segmentation on the material density time series data set to obtain a sequence of material density local data sets; inputting each material density local data set in the sequence of material density local data sets into a material density time series feature extractor based on the Bi-LSTM model to obtain a sequence of implicit fluctuation feature vectors of the material density local time series.

[0036] Next, the sequence of the local time series implicit fluctuation feature vectors of the material density is further subjected to feature aggregation to integrate the material density feature information in each local time domain, thereby obtaining a global material density feature representation. It is worth mentioning that in order to improve the representativeness of feature aggregation, in the step S230, the present application adopts an attention mechanism to optimize the feature aggregation process, and performs aggregation processing of local time domain information based on the spatiotemporal attenuation characteristics of information transmission. Specifically, the local time series implicit fluctuation feature vector of the material density at the end of the sequence is first used as the current local time domain feature, and other local time series implicit fluctuation feature vectors of the material density are defined as historical local time domain features, and the static energy value is calculated based on the statistical characteristics of each historical local time domain feature, which is used as the inherent attribute of the feature vector in the feature space to represent its static importance. Next, the static energy value is spatiotemporally modulated based on the time span between each historical local time domain feature and the current local time domain feature and the spatial span of the two in the feature space, so as to dynamically adjust the energy intensity of information transfer and obtain the dynamic energy spatiotemporal transfer value of each historical local time domain feature, so that the local time domain information that is closer to the current time point and has a smaller feature space span can obtain a higher weight in the information aggregation process, thereby enhancing the sensitivity to the locality and timeliness of information transfer. Then, each dynamic energy spatiotemporal transfer value is further normalized and masked to screen and strengthen important spatiotemporal transfer information. Furthermore, based on the dynamic energy spatiotemporal transfer value after masking, the local time domain information is aggregated by weighted summation to obtain the material density spatiotemporal significant aggregation propagation representation vector with global temporal context information.

[0037] In a specific example of the present application, the step S230 includes: extracting the current material density local time series implicit fluctuation feature vector from the sequence of the material density local time series implicit fluctuation feature vectors, and defining other material density local time series implicit fluctuation feature vectors as historical material density local time series implicit fluctuation feature vectors to obtain a sequence of historical material density local time series implicit fluctuation feature vectors; based on the spatial span and time span between each historical material density local time series implicit fluctuation feature vector in the sequence of historical material density local time series implicit fluctuation feature vectors and the current material density local time series implicit fluctuation feature vector, modulating the static energy value of each historical material density local time series implicit fluctuation feature vector to obtain the dynamic energy spatiotemporal transfer value of each historical material density local time series implicit fluctuation feature vector; based on the dynamic energy spatiotemporal transfer value of each historical material density local time series implicit fluctuation feature vector, aggregating the sequence of material density local time series implicit fluctuation feature vectors to obtain the material density spatiotemporal significant aggregation propagation representation vector.

[0038] Specifically, based on the spatial span and time span between each historical material density local time series implicit fluctuation characteristic vector in the sequence of the historical material density local time series implicit fluctuation characteristic vector and the current material density local time series implicit fluctuation characteristic vector, the static energy value of each historical material density local time series implicit fluctuation characteristic vector is modulated to obtain the dynamic energy spatiotemporal transfer value of each historical material density local time series implicit fluctuation characteristic vector, including: calculating the static energy value of each historical material density local time series implicit fluctuation characteristic vector in the sequence of the historical material density local time series implicit fluctuation characteristic vector; calculating the energy spatial attenuation factor and energy temporal attenuation factor of each historical material density local time series implicit fluctuation characteristic vector relative to the current material density local time series implicit fluctuation characteristic vector based on the static energy value of each historical material density local time series implicit fluctuation characteristic vector; calculating the weighted sum of the energy spatial attenuation factor and the energy temporal attenuation factor of each historical material density local time series implicit fluctuation characteristic vector to obtain the dynamic energy spatiotemporal transfer value of each historical material density local time series implicit fluctuation characteristic vector.

[0039] Specifically, the static energy value of each historical material density local time series implicit fluctuation characteristic vector in the sequence of the historical material density local time series implicit fluctuation characteristic vector is calculated, including: calculating the expected value of the fourth power of the difference between each eigenvalue in the historical material density local time series implicit fluctuation characteristic vector and its eigenmean, and dividing the expected value by the square of the characteristic variance of the historical material density local time series implicit fluctuation characteristic vector to obtain the kurtosis of the historical material density local time series implicit fluctuation characteristic vector; calculating the kurtosis of the historical material density local time series implicit fluctuation characteristic vector and the sum of its maximum eigenvalue to obtain the static energy value.

[0040] Specifically, based on the static energy value of each historical material density local time series implicit fluctuation characteristic vector, the energy space attenuation factor and energy time attenuation factor of each historical material density local time series implicit fluctuation characteristic vector relative to the current material density local time series implicit fluctuation characteristic vector are calculated, including: taking the square value of the number of characteristic vectors between the historical material density local time series implicit fluctuation characteristic vector and the current material density local time series implicit fluctuation characteristic vector as the space span coefficient, calculating the static energy value of the historical material density local time series implicit fluctuation characteristic vector and dividing it by the space span coefficient to obtain the energy space attenuation factor; calculating the timestamp difference between the historical material density local time series implicit fluctuation characteristic vector and the current material density local time series implicit fluctuation characteristic vector, rounding down the timestamp difference and dividing it by a preset time attenuation period to obtain the time span coefficient, and dividing the static energy value of the historical material density local time series implicit fluctuation characteristic vector by the natural exponential function value of the time span coefficient to obtain the energy time attenuation factor.

[0041] Specifically, based on the dynamic energy spatiotemporal transfer values ​​of the local time series implicit fluctuation characteristic vectors of each historical material density, the sequence of the local time series implicit fluctuation characteristic vectors of material density is aggregated to obtain the material density spatiotemporal significant aggregation propagation representation vector, including: inputting the dynamic energy spatiotemporal transfer values ​​of the local time series implicit fluctuation characteristic vectors of each historical material density into an energy transfer gating module based on a mask function to obtain a sequence of dynamic energy spatiotemporal transfer weights; based on the sequence of dynamic energy spatiotemporal transfer weights, calculating the weighted sum of the sequence of local time series implicit fluctuation characteristic vectors of historical material density to obtain the historical material density local time series feature transfer aggregation modulation vector; calculating the positional sum of the historical material density local time series feature transfer aggregation modulation vector and the current material density local time series implicit fluctuation characteristic vector to obtain the material density spatiotemporal significant aggregation propagation representation vector.

[0042] Accordingly, the step S230 includes: processing the sequence of the local temporal implicit fluctuation feature vectors of the material density with the following feature aggregation formula to obtain the spatiotemporal significant aggregation propagation representation vector of the material density, wherein the feature aggregation formula is:

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049] in, represents the sequence of implicit fluctuation eigenvectors of the local time series of the material density, , and Represents the first, second and The implicit fluctuation characteristic vector of the local time series of historical material density, Represents the implicit fluctuation characteristic vector of the local time series of the current material density, The value of is the number of eigenvectors in the sequence of the local time series implicit fluctuation eigenvectors of the material density. Indicates the The implicit fluctuation feature vector of the local time series of historical material density The eigenvalues ​​at the positions, , and Respectively represent the The maximum eigenvalue, eigenmean and square of eigenvariance of the local time series implicit fluctuation eigenvector of the historical material density are It means to calculate the expected value of a set. Indicates the The static energy value of the implicit fluctuation characteristic vector of the local time series of historical material density, Indicates the The number of feature vectors between the feature vectors of the local time series implicit fluctuation of the historical material density and the feature vector of the local time series implicit fluctuation of the current material density, and Respectively represent the local time series implicit fluctuation feature vector of the current material density and the The timestamp of the implicit fluctuation feature vector of the local time series of historical material density, Indicates rounding down. Indicates the preset time decay period, represents the exponential function operation, , and Respectively represent the The energy space attenuation factor, energy time attenuation factor and dynamic energy space-time transfer value of the local time series implicit fluctuation characteristic vector of the historical material density relative to the local time series implicit fluctuation characteristic vector of the current material density, and are different weight hyperparameters, Indicates mask processing, represents the sigmoid activation function, Indicates the preset mask threshold, The vector representing the spatiotemporal significant aggregation propagation of the material density is represented.

[0050] Finally, in step S240, the material density spatiotemporal significant aggregation propagation representation vector is input into a decoder-based reaction predictor to obtain a predicted decoded value of the remaining reaction time. It should be understood that the decoder is a decoding regression model based on deep learning, which is mainly used to convert the encoded feature representation back to the original data space or generate a target data representation. In the technical solution of the present application, the decoder extracts the deep-level features of the material density spatiotemporal significant aggregation propagation representation vector layer by layer, fully understands the material density time series change information contained therein, and maps it to the target numerical space, thereby generating a predicted decoded value of the remaining reaction time. In this way, the remaining time of the chemical reaction can be grasped in real time to avoid low production efficiency or safety hazards caused by too long or too short reaction time in the chemical industry.

[0051] In the technical solution of the present application, each material density local time series implicit fluctuation feature vector in the sequence of material density local time series implicit fluctuation feature vectors respectively represents the implicit feature of material density local time series fluctuation. When performing feature propagation aggregation based on spatiotemporal significance, the heterogeneity of node feature expression of each material density local time series implicit fluctuation feature vector in each material density local time series implicit fluctuation feature vector in the sequence of material density local time series implicit fluctuation feature vectors will cause the calculation of spatiotemporal parameters to shift, resulting in a large area of ​​simple repetition in the material density spatiotemporal significant aggregation propagation representation vector obtained by spatiotemporal significant feature propagation aggregation, which will affect the spatiotemporal transmission aggregation semantic logical dependency and reduce the accuracy of the decoding result based on temporal semantic distribution.

[0052] Preferably, the material density spatiotemporal significant aggregation propagation representation vector is input into a decoder-based reaction predictor to obtain a predicted decoded value of the remaining reaction time, including: Based on the association value and distance value of the eigenvalue of the material density spatiotemporal significant aggregation propagation representation vector, the material density spatiotemporal significant aggregation propagation association response matrix and the material density spatiotemporal significant aggregation propagation distance response matrix are determined, that is:

[0053] in, and They represent the first The eigenvalues ​​and eigenvalues, , The vector representing the spatiotemporal significant aggregation propagation of the material density is represented, represents the set of real numbers, The length of the vector representing the spatiotemporal significant aggregation propagation of the material density is represented, Indicates the material density spatiotemporal significant aggregation propagation correlation response matrix The eigenvalues ​​of the positions, Indicates the material density spatiotemporally significant aggregation propagation distance response matrix The eigenvalue of the position; Performing matrix multiplication of the material density spatiotemporal significant aggregation propagation representation vector and the material density spatiotemporal significant aggregation propagation associated response matrix to obtain a material density spatiotemporal significant aggregation propagation associated response vector; Performing matrix multiplication of the material density spatiotemporal significant aggregation propagation distance response matrix and the transposed vector of the material density spatiotemporal significant aggregation propagation representation vector to obtain a material density spatiotemporal significant aggregation propagation distance response vector; After performing a dot multiplication on the material density spatiotemporal significant aggregation propagation association response matrix and the material density spatiotemporal significant aggregation propagation distance response matrix, perform a matrix multiplication on the transposed vector of the material density spatiotemporal significant aggregation propagation representation vector to obtain a material density spatiotemporal significant aggregation propagation logic bias vector; The material density spatiotemporal significant aggregation propagation associated response vector, the material density spatiotemporal significant aggregation propagation distance response vector and the material density spatiotemporal significant aggregation propagation logical bias vector are added to obtain an optimized material density spatiotemporal significant aggregation propagation representation vector; The optimized material density spatiotemporal significant aggregation propagation representation vector is input into a decoder-based reaction predictor to obtain a predicted decoding value of the remaining reaction time.

[0054] Here, the optimized material density spatiotemporal significant aggregation propagation represents the vector It is expressed as:

[0055] exist ,and In the case of:

[0056]

[0057] in, Represents the material density spatiotemporal significant aggregation propagation correlation response matrix, Represents the material density spatiotemporally significant aggregation propagation distance response matrix, The vector representing the spatiotemporal significant aggregation propagation of the material density is represented by, represents matrix multiplication, It means point multiplication by position. Indicates adding by position point. represents the transpose of a vector, Indicates the intermediate vector of material density spatiotemporal significant aggregation propagation, Represents the material density spatiotemporal significant aggregation propagation logic bias vector, The vector representing the optimized spatiotemporal aggregation propagation of material density.

[0058] Therefore, by taking the autocorrelation matrix and self-distance matrix of the material density spatiotemporal significant aggregation propagation representation vector as the statistical reference-free distribution response framework of the material density spatiotemporal significant aggregation propagation representation vector, based on the retrieval-enhanced reverse response construction of the material density spatiotemporal significant aggregation propagation representation vector, the simple repetition of feature distribution is avoided, and by ensuring the intrinsic mapping logic of the material density spatiotemporal significant aggregation propagation representation vector based on the retrieval-response context, the surface combinatorial mapping of the material density spatiotemporal significant aggregation propagation representation vector is avoided, so as to achieve the logical dependency mapping of the material density spatiotemporal significant aggregation propagation representation vector to the decoding target domain while maintaining the intuitive response correlation, thereby improving the accuracy of the predicted decoding value of the remaining reaction time obtained by the reaction predictor of the decoder based on the input of the material density spatiotemporal significant aggregation propagation representation vector.

[0059] like Figure 1 As shown, after the reaction is completed, the outflowing mixture containing polycarbonate polyether polyol, cyclic carbonate and unreacted raw material propylene oxide is passed into a primary separation device, and the primary separation device separates the propylene oxide and dissolved CO2 in the mixture to obtain a crude product, and adjusts the viscosity of the crude product at the same time; the crude product is passed into a catalyst separation device, and a refining agent is added for adsorption treatment, and a crude product with a catalyst metal content of less than 5 ppm is obtained after filtration; the refined crude product is passed into a secondary separation device to separate the main product polycarbonate polyether polyol and the by-product cyclic propylene carbonate, that is, a citric acid type polycarbonate polyether polyol product is obtained, and then filled.

[0060] In one embodiment, the feed ratio of propylene oxide and citric acid is propylene oxide:succinic acid=45:1-95:1, and the DMC catalyst is 500-1000ppm.

[0061] In one embodiment, in step (2), the volume of the material accounts for 50%-70% of the reaction pipeline; In one embodiment, in step (3), the material comprises product polycarbonate polyether polyol, by-product cyclic carbonate and unreacted propylene oxide, and the primary separation device purifies and recovers unreacted carbon dioxide and propylene oxide by vacuum distillation; the catalyst separation device adds a refining agent to the material and then separates and removes the residual catalyst using a solid-liquid separation device; the secondary separation device includes a distillation tower, which separates the product citric acid type polycarbonate polyether polyol and the by-product propylene carbonate by distillation.

[0062] Furthermore, the structural formula of the citric acid type polycarbonate polyether polyol prepared by the above production process and the corresponding device is as follows: , Wherein 1≤m1, m2, m3≤34, 1≤n1, n2, n3≤48, 0≤m4≤37, 0≤n4≤58, m and n are both positive integers; the functionality of the citric acid type polycarbonate polyether polyol is 3-4; preferably, the functionality is 3.2-4; the molecular weight range of the citric acid type polycarbonate polyether polyol is 2500-6500; the PO conversion rate of the citric acid type polycarbonate polyether polyol is above 90%.

[0063] Correspondingly, the present application also provides a real-time prediction method for reaction time, which includes: obtaining a time queue of material density collected by a densitometer; performing time series feature extraction based on the local time scale on the time queue of material density to obtain a sequence of local time series implicit fluctuation feature vectors of material density; performing feature propagation aggregation based on spatiotemporal significance on the sequence of local time series implicit fluctuation feature vectors of material density to obtain a spatiotemporal significant aggregation propagation representation vector of material density; and generating a predicted decoding value of the remaining reaction time based on the spatiotemporal significant aggregation propagation representation vector of material density.

[0064] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned real-time prediction method of reaction time have been described in detail in the description of the production process of the above-mentioned citric acid type polycarbonate polyether polyol, and therefore, its repeated description will be omitted.

[0065] In summary, the production process and real-time prediction method of reaction time of citric acid type polycarbonate polyether polyol according to the embodiment of the present application are explained, and the production process is carried out in a tube-in-tube reaction device, which uses the form of partition premixing to prevent citric acid from passivating the catalyst activation center, and uses artificial intelligence technology to monitor the density of the material in real time and fine-grained time series analysis during the reaction process, so as to intelligently predict the remaining reaction time based on the time series dynamic correlation change characteristics of the material density. In this way, the reaction endpoint can be predicted in advance, and the occurrence of over-reaction or incomplete reaction can be effectively avoided, thereby improving production efficiency and product quality, and being suitable for industrial-scale production.

[0066] The present application will be further described below in conjunction with specific embodiments.

[0067] Example 1 8L propylene oxide, 439.6g citric acid and 4g DMC catalyst were added to the premixing kettle for full premixing. After 0.3Mpa of carbon dioxide was introduced, the mixture was mixed and stirred at a temperature of 40°C. After stirring for 1h, the mixture was pressed into an 8L*2 tubular reactor with a metering pump. The diameter of a single reaction pipe was 100mm, the length of the reaction pipe was 1000mm, the temperature in the reaction pipe was 40°C, and the carbon dioxide pressure was 0.5MPa. After all the mixture was pressed in, carbon dioxide was introduced into the reaction pipe to 3Mpa and the pressure was maintained. The circulation pump was turned on, the flow rate was controlled at 800g / min, and the gas-liquid ratio of the atomizing nozzle was controlled at 5%. After reacting at a temperature of 80℃±5℃ for 3h, the density was uniformly increased to 1.102g / cm 3 The reaction is complete and the product is taken out after cooling down. Unreacted carbon dioxide and propylene oxide are recovered by vacuum distillation in the primary separation device, and then a refining agent is added to the catalyst separation device to remove the residual catalyst by solid-liquid separation equipment, and then the main and side products are separated by a distillation tower in the secondary separation device to obtain a purified citric acid type polycarbonate polyether polyol product.

[0068] Example 2 Compared with Example 1, other parameters and reaction materials in Example 2 are unchanged. The only change is that the density control method described above is adopted in Example 2.

[0069] The crude product after the catalyst is removed is sampled and analyzed. At this time, the crude product contains the main product polycarbonate polyether polyol and the by-product cyclic propylene carbonate. Specifically, the crude product is sampled and collected in a container, and the crude product logistics sample is characterized by nuclear magnetic resonance hydrogen spectrum to calculate the ratio of polymer to cyclic small molecules in the crude product. After the polymer is purified, the nuclear magnetic resonance hydrogen spectrum test is performed again to calculate the ratio of polycarbonate chain segments to polyether chain segments on the polymer main chain. There are only two structures, polycarbonate chain segments and polyether chain segments, on the polymer main chain, and the sum of the two percentages is 100%.

[0070] With the help of 1 The amount of carbon dioxide introduced in the obtained polycarbonate polyether polyol (carbonate segment content) and the ratio of propylene carbonate (cyclic carbonate) to polycarbonate polyether polyol were determined by H-NMR (Bruker, DPX400, 400 MHz; pulse program zg30, waiting time d1:10 s, 64 scans). In each case, the sample was dissolved in deuterated chloroform. 1 The relevant resonances in H-NMR (based on TMS = 0 ppm) are as follows: Among them, 5.0ppm and 4.2ppm belong to the proton peaks of the methine and methylene on the polycarbonate chain, 4.9ppm, 4.5ppm and 4.1ppm belong to the proton peaks of the methine and methylene in the five-membered ring carbonate, and 3.5-3.8ppm belong to the proton peak of the ether chain. The capital letter A plus a number subscript represents the integrated area of ​​the peak at a certain ppm on the nuclear magnetic resonance hydrogen spectrum. A is the abbreviation of Area in English, for example, A 5.0 represents the integrated area of ​​the peak at 5.0 ppm. 1 The integrated area of ​​the HNMR spectrum and its related proton peaks, the ratio of carbonate chains in the copolymerization reaction (molar ratio) (F CO2 ) and the mass fraction of cyclic carbonate content (W PC ), the amount of carbon dioxide embedded (M CO2 ) is calculated as follows: Among them, F CO2 =(A 5.0 +A 4.2 -2×A 4.6 ) / [(A 5.0 +A 4.2 -2×A 4.6 )+A 3.5 ]×100%; W PC =102×A 1.5 / [102×(A 5.0 +A 4.2 -2×A 4.6 +A1.5)+58×A 3.5 ]×100%; M CO2 =44×F CO2 / [102×F CO2 +58×(1-F CO2 )]×100%; The coefficient 44 is the molar mass of carbon dioxide, the coefficient 58 is the molar mass of PO, and the coefficient 102 is the sum of the molar masses of carbon dioxide (molar mass 44 g / mol) and the molar masses of PO (molar mass 58 g / mol).

[0071] The conversion rate of propylene oxide is calculated by cooling the product, separating the unreacted propylene oxide through distillation in a primary separation device, and calculating the conversion rate. The calculation formula is n=m×(1-M CO2 ) / (Mm×M CO2 ); wherein m is the mass of the product after removing unreacted propylene oxide, and M is the total sample containing propylene oxide.

[0072] The polymer number average molecular weight (Mn) and polymer molecular weight polydispersity index (PDI) were determined by gel permeation chromatography (GPC).

[0073] Average functionality:

[0074] in is the number average molecular weight, determined by gel permeation chromatography (GPC), and the hydroxyl value (OHV) is obtained by GB / T12008.3-2009 Plastic polyether polyols Part 3: Determination of hydroxyl value. The test results are shown in Table 1 below: Table 1

[0075] Compared with Example 1, other parameters and reaction materials of Example 2 have not been changed. The only change is that the density control method described above is adopted in Example 2. It can be seen from the data of Example 2 that the density control as described above can make the raw materials react more fully, and the conversion rate will not be low due to the premature termination of the reaction caused by the fixed judgment point of the reaction progress. The full reaction of CO2 also improves the carbonate content and increases the fixation rate of CO2; it will not prolong the reaction time due to the excessively late termination of the reaction when the reaction speed becomes faster in the early stage, resulting in a wide molecular weight distribution, and can improve the average functionality and better performance. The adaptive density control strategy of Example 2 found that the desired density was achieved at a shorter reaction time, so that the reaction can be terminated flexibly and timely, and the reaction time is shorter and the production efficiency is higher under the premise of ensuring the quality of the product.

[0076] Finally, it should be noted that the embodiments described above are only some embodiments of the present invention, rather than all embodiments. The detailed description of the embodiments of the present invention is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

Claims

1. A production process of citric acid type polycarbonate polyether polyol, It is characterized in that The production process of the citric acid type polycarbonate polyether polyol is carried out in a tubular reactor, wherein the tubular reactor comprises a reaction pipe having a gas-liquid mass transfer zone, a static mixer arranged in the reaction pipe, and a circulation device, wherein the circulation device comprises a circulation pump and a compression pump; Wherein, the production process comprises: Step (1): fully premixing propylene oxide, citric acid and DMC catalyst in a premixing kettle to obtain a premix; Step (2): the premixed material enters the tubular reactor for reaction. The tubular reactor adopts a gas-liquid dual circulation method to carry out a cyclic reaction of the material. The reaction temperature is 80°C±5°C. One end of the feed inlet of the tubular reactor is provided with one or more dual-flow atomizing nozzles for mixing and atomizing the circulated gas and liquid materials and dispersing them in the gas-liquid mass transfer zone for mass transfer and heat transfer. The flow rate of the material is 5-10 minutes to complete one cycle. The reaction endpoint is determined by the density of the material. Step (3): after the reaction is completed, the material is discharged and sequentially passed through a primary separation device to remove carbon dioxide and propylene oxide, a catalyst separation device to separate the catalyst, and a secondary separation device to remove by-products to obtain the citric acid type polycarbonate polyether polyol; Wherein, the step (2) further includes: Obtain the time sequence of the density of the material collected by the density meter; Extracting time series features based on local time scale for the time queue of the density of the material to obtain a sequence of local time series implicit fluctuation feature vectors of the material density; Performing feature propagation aggregation based on spatiotemporal significance on the sequence of local time series implicit fluctuation feature vectors of the material density to obtain a spatiotemporal significant aggregation propagation representation vector of the material density; Based on the material density spatiotemporal significant aggregation propagation representation vector, a predicted decoding value of the remaining reaction time is generated.

2. The production process of citric acid type polycarbonate polyether polyol according to claim 1, It is characterized in that Extracting time series features based on local time scale for the time series of the density of the material to obtain a sequence of local time series implicit fluctuation feature vectors of the material density includes: Arranging the time queue of the density of the material into a material density time series data set according to the time dimension; Performing data set segmentation on the material density time series data set to obtain a sequence of material density local data sets; Each material density local data set in the sequence of the material density local data sets is input into the material density time series feature extractor based on the Bi-LSTM model to obtain a sequence of the material density local time series implicit fluctuation feature vectors.

3. The production process of citric acid type polycarbonate polyether polyol according to claim 2, It is characterized in that The sequence of the local time series implicit fluctuation feature vectors of the material density is subjected to feature propagation aggregation based on spatiotemporal significance to obtain a spatiotemporal significant aggregation propagation representation vector of the material density, including: Extracting the current material density local time series implicit fluctuation feature vector from the sequence of the material density local time series implicit fluctuation feature vectors, and defining other material density local time series implicit fluctuation feature vectors as historical material density local time series implicit fluctuation feature vectors to obtain a sequence of historical material density local time series implicit fluctuation feature vectors; Based on the spatial span and time span between each historical material density local time series implicit fluctuation characteristic vector in the sequence of the historical material density local time series implicit fluctuation characteristic vector and the current material density local time series implicit fluctuation characteristic vector, the static energy value of each historical material density local time series implicit fluctuation characteristic vector is modulated to obtain the dynamic energy spatiotemporal transfer value of each historical material density local time series implicit fluctuation characteristic vector; Based on the dynamic energy spatiotemporal transfer value of each historical material density local time series implicit fluctuation characteristic vector, the sequence of the material density local time series implicit fluctuation characteristic vector is aggregated to obtain the material density spatiotemporal significant aggregation propagation representation vector.

4. The production process of citric acid type polycarbonate polyether polyol according to claim 3, It is characterized in that Based on the spatial span and time span between each historical material density local time series implicit fluctuation feature vector in the sequence of the historical material density local time series implicit fluctuation feature vector and the current material density local time series implicit fluctuation feature vector, the static energy value of each historical material density local time series implicit fluctuation feature vector is modulated to obtain the dynamic energy spatiotemporal transfer value of each historical material density local time series implicit fluctuation feature vector, including: Calculating the static energy value of each historical material density local time series implicit fluctuation characteristic vector in the sequence of the historical material density local time series implicit fluctuation characteristic vector; Based on the static energy value of each historical material density local time series implicit fluctuation characteristic vector, calculate the energy space attenuation factor and energy time attenuation factor of each historical material density local time series implicit fluctuation characteristic vector relative to the current material density local time series implicit fluctuation characteristic vector; The weighted sum of the energy space attenuation factor and the energy time attenuation factor of each historical material density local time series implicit fluctuation characteristic vector is calculated to obtain the dynamic energy space-time transfer value of each historical material density local time series implicit fluctuation characteristic vector.

5. The production process of citric acid type polycarbonate polyether polyol according to claim 4, It is characterized in that Calculating the static energy value of each historical material density local time series implicit fluctuation characteristic vector in the sequence of the historical material density local time series implicit fluctuation characteristic vector includes: Calculate the expected value of the fourth power of the difference between each eigenvalue and its eigenmean in the local time series implicit fluctuation eigenvector of the historical material density, and divide the expected value by the square of the eigenvariance of the local time series implicit fluctuation eigenvector of the historical material density to obtain the kurtosis of the local time series implicit fluctuation eigenvector of the historical material density; The static energy value is obtained by calculating the sum of the kurtosis of the local time series implicit fluctuation eigenvector of the historical material density and its maximum eigenvalue.

6. The production process of citric acid type polycarbonate polyether polyol according to claim 5, It is characterized in that Based on the static energy value of each historical material density local time series implicit fluctuation characteristic vector, the energy space attenuation factor and energy time attenuation factor of each historical material density local time series implicit fluctuation characteristic vector relative to the current material density local time series implicit fluctuation characteristic vector are calculated, including: Taking the square value of the number of eigenvectors between the local time series implicit fluctuation eigenvector of the historical material density and the local time series implicit fluctuation eigenvector of the current material density as the spatial span coefficient, the static energy value of the local time series implicit fluctuation eigenvector of the historical material density is calculated and divided by the spatial span coefficient to obtain the energy spatial attenuation factor; Calculate the timestamp difference between the local time series implicit fluctuation characteristic vector of the historical material density and the local time series implicit fluctuation characteristic vector of the current material density, round down the timestamp difference and divide it by the preset time decay period to obtain the time span coefficient, and divide the static energy value of the local time series implicit fluctuation characteristic vector of the historical material density by the natural exponential function value of the time span coefficient to obtain the energy time decay factor.

7. The production process of citric acid type polycarbonate polyether polyol according to claim 6, It is characterized in that Based on the dynamic energy spatiotemporal transfer value of each historical material density local time series implicit fluctuation feature vector, aggregating the sequence of the material density local time series implicit fluctuation feature vector to obtain the material density spatiotemporal significant aggregation propagation representation vector, including: Input the dynamic energy spatiotemporal transfer value of each historical material density local time series implicit fluctuation feature vector into the energy transfer gating module based on the mask function to obtain a sequence of dynamic energy spatiotemporal transfer weights; Based on the sequence of dynamic energy spatiotemporal transfer weights, a weighted sum of the sequence of local temporal implicit fluctuation feature vectors of historical material density is calculated to obtain a local temporal feature transfer aggregation modulation vector of historical material density; The position-based sum of the local time series characteristic transfer aggregation modulation vector of the historical material density and the local time series implicit fluctuation characteristic vector of the current material density is calculated to obtain the spatiotemporal significant aggregation propagation representation vector of the material density.

8. The production process of citric acid type polycarbonate polyether polyol according to claim 7, It is characterized in that Based on the material density spatiotemporal significant aggregation propagation representation vector, a predicted decoding value of the remaining reaction time is generated, including: The material density spatiotemporal significant aggregation propagation representation vector is input into a decoder-based reaction predictor to obtain a predicted decoding value of the remaining reaction time.

9. The production process of citric acid type polycarbonate polyether polyol according to claim 1, It is characterized in that The double-flow atomizing nozzle comprises a liquid material atomizing feed inlet, a carbon dioxide circulation air inlet, a mixing chamber air injection hole, a mixing chamber and a double-flow atomizing nozzle, wherein the liquid material enters from the liquid material atomizing feed inlet, and the CO 2 From the carbon dioxide circulation inlet, liquid materials and CO 2 The gas enters the mixing chamber through the mixing chamber air injection hole to be fully mixed and then is sprayed out from the double-flow atomizing nozzle.

10. A real-time prediction method for reaction time, applied to the production process of citric acid type polycarbonate polyether polyol according to claim 1, It is characterized in that include: A time queue of the density of a material collected by a densitometer is obtained, wherein the material is a mixture of a gas and a liquid material, the gas is carbon dioxide, and the liquid material is a mixture of propylene oxide, citric acid, and a DMC catalyst; Extracting time series features based on local time scale for the time queue of the density of the material to obtain a sequence of local time series implicit fluctuation feature vectors of the material density; Performing feature propagation aggregation based on spatiotemporal significance on the sequence of local time series implicit fluctuation feature vectors of the material density to obtain a spatiotemporal significant aggregation propagation representation vector of the material density; Based on the material density spatiotemporal significant aggregation propagation representation vector, a predicted decoding value of the remaining reaction time is generated.

Images