Optimized production control method and system for absolute ethyl alcohol

Abstract

The invention provides an absolute ethyl alcohol optimized production control method and system, and relates to the technical field of absolute ethyl alcohol production. The method comprises the following steps: determining first moisture content data, inlet pressure intensity and outlet pressure intensity through a first moisture analyzer and a pressure intensity meter which are arranged on a molecular sieve; determining whether to close a molecular sieve channel; closing a molecular sieve channel to obtain regenerated tail gas; ethanol saturated steam is obtained; opening a channel of the membrane filter to obtain first absolute ethyl alcohol steam; after purging, opening a molecular sieve channel; determining whether the membrane filter is cleaned; carrying out chemical cleaning; and opening a molecular sieve channel to obtain second absolute ethyl alcohol steam. According to the invention, when absolute ethyl alcohol is produced, the moisture content, the inlet pressure and the outlet pressure of the molecular sieve can be analyzed to determine whether to close the molecular sieve channel, so that regenerated tail gas is recycled, and absolute ethyl alcohol is alternately produced through the molecular sieve and the membrane filter. The gas consumption can be reduced, the equipment utilization rate and the product yield are improved, and the operation stability is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of anhydrous ethanol production, and particularly relates to an anhydrous ethanol optimized production control method and system. BACKGROUND

[0002] In the production of anhydrous ethanol, the saturation of the molecular sieve needs to be detected, and the molecular sieve needs to be regenerated when the molecular sieve is saturated, so as to ensure the dehydration efficiency of the molecular sieve. In the related art, the production is stopped when the molecular sieve is regenerated, or the production is maintained by using two molecular sieves in an alternating manner, that is, when one molecular sieve is regenerated, the other molecular sieve is used for production. However, the utilization rate of the ethanol-containing steam generated by the blowing process during the regeneration of the molecular sieve is low, the product yield is insufficient, and it is difficult to accurately determine whether the molecular sieve is saturated during the production process.

[0003] The information disclosed in the background section of this application is only intended to deepen the understanding of the general background of the application and should not be considered as an acknowledgement or any form of suggestion that this information constitutes prior art known to those skilled in the art. SUMMARY

[0004] The present application provides an anhydrous ethanol optimized production control method and system, which can solve the technical problems of the related art that it is difficult to accurately determine whether the molecular sieve is saturated and the utilization rate of the steam is low.

[0005] According to a first aspect of the present application, an anhydrous ethanol optimized production control method is provided, comprising: At the current time, the first moisture content data at the outlet of the molecular sieve, the inlet pressure and the outlet pressure are determined by the first moisture analyzer arranged at the outlet of the molecular sieve and the pressure gauges arranged at the inlet and outlet of the molecular sieve; According to the first moisture content data, the inlet pressure and the outlet pressure at multiple times, it is determined whether to close the molecular sieve channel; If it is determined to close the molecular sieve channel, the superheater channel is opened, and superheated steam is generated to blow the molecular sieve, and the regeneration tail gas is obtained; The regeneration tail gas and the preheated raw material ethanol are jointly input into the recovery tower to obtain ethanol saturated steam; The membrane filter channel is opened, the ethanol saturated steam is transported to the membrane filter, the first anhydrous ethanol steam is obtained, and after cooling, it is stored in the warehouse; At the next time after blowing, it is determined to open the molecular sieve channel; The second moisture content data obtained by the second moisture analyzer arranged at the outlet of the membrane filter at multiple times, and the steam flux obtained by the flux detection device arranged at the outlet of the membrane filter at multiple times, are used to determine whether the membrane filter needs to be cleaned; If cleaning is needed, chemical cleaning is performed; If it is determined to open the molecular sieve channel, the membrane filter channel and the superheater channel are closed, the ethanol saturated steam generated after the recycling tower processes the preheated raw material ethanol is input into the molecular sieve, second anhydrous ethanol steam is obtained, and after cooling, it is stored in the warehouse.

[0006] According to the present application, determining whether to close the molecular sieve channel comprises: Obtaining first moisture content data, inlet pressure and outlet pressure at multiple time points between the first time point after the last time the molecular sieve is purged and the current time point; The first moisture content data, the inlet pressure and the outlet pressure at each time point form a first judgment input vector at each time point; The first judgment input vector at the i-th time point is input into the first fully connected layer for processing to obtain the first judgment feature vector at the i-th time point; The first judgment feature vector at the i-th time point and the first hidden state vector at the i-1-th time point are input into the molecular sieve saturation judgment model to obtain the first hidden state vector at the i-th time point, wherein when i=1, the first hidden state vector at the i-1-th time point is a zero vector, and i is a positive integer; The process of obtaining the first hidden state vector is iteratively performed until the first hidden state vector at the current time point is obtained, and the first hidden state vector at the current time point is input into the second fully connected layer and the first activation layer for processing to obtain the molecular sieve saturation probability information; According to the molecular sieve saturation probability information, it is determined whether to close the molecular sieve channel.

[0007] According to the present application, the training step of the molecular sieve saturation judgment model comprises: Obtaining first training moisture content data, training inlet pressure and training outlet pressure at multiple training time points after purging the molecular sieve; Obtaining first sampling moisture content data, sampling inlet pressure and sampling outlet pressure at multiple sub-sampling time points between each training time point; According to the first training moisture content data, the training inlet pressure and the training outlet pressure at the j-th training time point, and the first sampling moisture content data, the sampling inlet pressure and the sampling outlet pressure at the multiple sub-sampling time points between the j-th training time point and the j+1-th training time point, the molecular sieve saturation labeling information at the j-th training time point is determined; According to the first training moisture content data, the training inlet pressure and the training outlet pressure through the j-th training time point and the training time points before it, the training judgment input vector of the j-th training time point and the training time points before it is obtained, and the training judgment feature vector of the j-th training time point and the training time points before it is obtained by processing through the first fully connected layer; processing the training judgment feature vectors of the jth training moment and the training moments before the jth training moment by the molecular sieve saturation judgment model to obtain a training hidden state vector of the jth training moment; inputting the training hidden state vector of the jth training moment into the second full connection layer and the first activation layer to obtain molecular sieve saturation prediction probability information of the jth training moment; obtaining a loss function of the molecular sieve saturation judgment model according to the molecular sieve saturation annotation information and the molecular sieve saturation prediction probability information of the jth training moment, and the first sampling moisture content data, the sampling inlet pressure and the sampling outlet pressure of the plurality of subdivided sampling moments between the jth training moment and the j+1th training moment; training the molecular sieve saturation judgment model according to the loss function of the molecular sieve saturation judgment model to obtain the trained molecular sieve saturation judgment model.

[0008] According to the present application, the molecular sieve saturation annotation information of the jth training moment is determined, comprising: if the first training moisture content data of the jth training moment is higher than the first moisture content threshold value, or the pressure difference between the training inlet pressure and the training outlet pressure is greater than the first pressure difference threshold value, the molecular sieve saturation annotation information of the jth training moment is determined as molecular sieve saturation; or, if there is a subdivided sampling moment in which the first sampling moisture content data is higher than the first moisture content threshold value, or the pressure difference between the sampling inlet pressure and the sampling outlet pressure is greater than the first pressure difference threshold value, the molecular sieve saturation annotation information of the jth training moment is determined as molecular sieve saturation; otherwise, the molecular sieve saturation annotation information of the jth training moment is determined as molecular sieve unsaturation.

[0009] According to the present application, the loss function of the molecular sieve saturation judgment model is obtained, comprising: if the molecular sieve saturation annotation information of the jth training moment is determined as molecular sieve saturation, and the first training moisture content data of the jth training moment is lower than or equal to the first moisture content threshold value, or the pressure difference between the training inlet pressure and the training outlet pressure is less than or equal to the first pressure difference threshold value, the serial number of the first subdivided sampling moment in which the first sampling moisture content data is higher than the first moisture content threshold value, or the pressure difference between the sampling inlet pressure and the sampling outlet pressure is greater than the first pressure difference threshold value is determined; determining the loss function of the molecular sieve saturation judgment model according to the molecular sieve saturation annotation information and the molecular sieve saturation prediction probability information of the jth training moment, and the serial number of the first subdivided sampling moment.

[0010] According to the present application, the loss function of the molecular sieve saturation judgment model is determined, comprising: according to the formula Loss function for determining molecular sieve saturation judgment model , wherein, is the molecular sieve saturation prediction probability information of the jth training moment, is the probability value of the molecular sieve saturation determined according to the molecular sieve saturation label information of the jth training moment, is the number of subdivided sampling moments between the jth training moment and the j+1th training moment, is the serial number of the first subdivided sampling moment between the jth training moment and the j+1th training moment, which is higher than the first moisture content threshold value or the pressure difference between the sampling inlet pressure and the sampling outlet pressure is greater than the first pressure difference threshold value, N is the number of training moments, j≤N, and j, N, , are all positive integers.

[0011] According to the application, whether the membrane filter needs to be cleaned is determined, comprising: obtaining the second moisture content data and the steam flux obtained at the next moment of the moment when the membrane filter channel is opened after the last chemical cleaning, and determining whether the membrane filter needs to be cleaned.

[0012] According to the second aspect of the application, an anhydrous ethanol optimal production control system is provided, comprising: a data acquisition module, which determines the first moisture content data at the molecular sieve outlet, the inlet pressure and the outlet pressure at the molecular sieve inlet and outlet by the first moisture analyzer arranged at the molecular sieve outlet and the pressure gauge arranged at the molecular sieve inlet and outlet at the current moment; a first judgment module, which determines whether to close the molecular sieve channel according to the first moisture content data, the inlet pressure and the outlet pressure at multiple moments; a regeneration tail gas module, which opens the superheater channel and generates superheated steam to purge the molecular sieve and obtain regeneration tail gas if it is determined to close the molecular sieve channel; an ethanol saturated steam module, which inputs the regeneration tail gas and the preheated raw material ethanol into the recovery tower together to obtain ethanol saturated steam; a first anhydrous ethanol steam module, which opens the membrane filter channel, transports the ethanol saturated steam to the membrane filter, obtains first anhydrous ethanol steam, and stores in the warehouse after cooling; a molecular sieve channel opening module, which determines to open the molecular sieve channel at the next moment after the purge; a second judgment module, which determines whether the membrane filter needs to be cleaned by the second moisture content data obtained by the second moisture analyzer arranged at the membrane filter outlet at multiple moments and the steam flux obtained by the flux detection device arranged at the membrane filter outlet at multiple moments; Chemical cleaning module, if necessary, chemical cleaning is carried out; The second anhydrous ethanol vapor module, if it is determined to open the molecular sieve channel, the membrane filter channel and the superheater channel are closed, the ethanol saturated steam generated after the raw material ethanol is preheated and treated by the recovery tower is input into the molecular sieve, the second anhydrous ethanol vapor is obtained, and after cooling, it is stored in the warehouse.

[0013] By adopting the technical scheme, the following technical effects can be achieved: According to the present application, the first moisture content data, the inlet pressure and the outlet pressure can be determined through the first moisture analyzer arranged at the outlet of the molecular sieve, and the pressure gauges arranged at the inlet and the outlet of the molecular sieve, the saturation of the molecular sieve can be determined, the second moisture content data can be obtained through the second moisture analyzer arranged at the outlet of the membrane filter, and the steam flux can be obtained through the flux detection device arranged at the outlet of the membrane filter, whether the membrane filter needs to be cleaned can be determined, so that whether the molecular sieve is saturated and whether the membrane filter needs to be cleaned can be accurately determined, the regenerated tail gas can be recovered, the anhydrous ethanol vapor can be obtained through the membrane filter and the molecular sieve, the steam consumption is reduced, the utilization rate of the equipment and the steam and the product yield are improved, and the operation stability is improved. When collecting data, the first moisture content data at the outlet of the molecular sieve, the inlet pressure and the outlet pressure can be obtained based on the first moisture analyzer and the pressure gauges arranged at the inlet and the outlet of the molecular sieve, which provides basic data for determining whether to close the molecular sieve channel. When training the molecular sieve saturation judgment model, considering that the molecular sieve may reach the saturated state between adjacent training time points, a plurality of subdivided sampling time points are arranged between each training time point, so that the time point when the molecular sieve reaches saturation can be more accurately determined, and the situation that the larger the serial number of the first subdivided sampling time point when the molecular sieve is saturated, the more difficult the training and the higher the importance of the training data is considered, the weight value of the cross-entropy loss function corresponding to the molecular sieve saturation prediction probability information of each training time point is set, the loss function of the molecular sieve saturation judgment model is obtained, and the trained molecular sieve saturation judgment model is obtained, so that the accuracy and pertinence of the training are improved, and the performance of the molecular sieve saturation judgment model is improved. After it is determined to close the molecular sieve channel, the membrane filter channel can be opened, the regenerated tail gas obtained by purging the molecular sieve can be input into the membrane filter together with the raw material ethanol to obtain ethanol saturated steam, the utilization rate of the raw material ethanol can be improved, the anhydrous ethanol can be continuously produced during the regeneration of the molecular sieve, and the utilization rate of the equipment is improved. Before the molecular sieve channel is opened again, whether the membrane filter needs to be cleaned can be determined, and if necessary, chemical cleaning is carried out, so that the membrane filter can still be used to produce anhydrous ethanol when the molecular sieve is saturated next time, the production does not stop, and the utilization rate of the equipment and the steam is improved. The molecular sieve and the membrane filter work alternately, the anhydrous ethanol is continuously produced, the gas consumption is reduced, the utilization rate of the equipment and the product yield are improved, and the operation stability is improved.

[0014] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Other features and aspects of the invention will become clearer from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort. Figure 1 An exemplary flowchart of an anhydrous ethanol optimized production control method according to an embodiment of the present invention is shown. Figure 2 An exemplary flowchart illustrating the determination of whether to close the molecular sieve channel according to an embodiment of the present invention is shown; Figure 3 A block diagram of an anhydrous ethanol optimized production control system according to an embodiment of the present invention is shown as an example. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0018] Figure 1 An exemplary flowchart illustrates a method for optimizing the production control of anhydrous ethanol according to an embodiment of the present invention, the method comprising: Step S1: At the current moment, the first moisture content data, inlet pressure and outlet pressure at the molecular sieve outlet are determined by the first moisture analyzer installed at the molecular sieve outlet and the pressure meters installed at the molecular sieve inlet and outlet. Step S2: Based on the first moisture content data, inlet pressure, and outlet pressure at multiple time points, determine whether to close the molecular sieve channel; Step S3: If it is determined that the molecular sieve channel is closed, the superheater channel is opened and superheated steam is generated to purge the molecular sieve and obtain regenerated tail gas. Step S4: The regeneration tail gas and the preheated raw material ethanol are fed together into the recovery tower to obtain saturated ethanol vapor. Step S5: Open the membrane filter channel and deliver saturated ethanol vapor to the membrane filter to obtain the first anhydrous ethanol vapor, which is then cooled and stored in the warehouse. Step S6: At the next moment after purging, determine to open the molecular sieve channel; Step S7: Determine whether the membrane filter needs cleaning by using the second moisture content data obtained at multiple times by the second moisture analyzer installed at the membrane filter outlet and the steam flux obtained at multiple times by the flux detection device installed at the membrane filter outlet. Step S8: If cleaning is required, perform chemical cleaning. Step S9: If it is determined that the molecular sieve channel is to be opened, the membrane filter channel and the superheater channel are closed. The saturated ethanol vapor generated after the recovery tower processes the preheated raw ethanol is input into the molecular sieve to obtain the second anhydrous ethanol vapor, which is then cooled and stored in the warehouse.

[0019] According to an embodiment of the present invention, the anhydrous ethanol optimized production control method can determine the first moisture content data, inlet pressure, and outlet pressure by using a first moisture analyzer installed at the outlet of the molecular sieve and pressure meters installed at the inlet and outlet of the molecular sieve to determine whether the molecular sieve is saturated. Furthermore, it can determine whether the membrane filter needs cleaning by using a second moisture analyzer installed at the outlet of the membrane filter to obtain the second moisture content data and by using a steam flux detection device installed at the outlet of the membrane filter to obtain the steam flux. This allows for accurate judgment of whether the molecular sieve is saturated and whether the membrane filter needs cleaning. It also allows for the recovery and regeneration of tail gas, and the generation of anhydrous ethanol vapor through the membrane filter and molecular sieve, reducing steam consumption, improving equipment and steam utilization and product yield, and enhancing operational stability.

[0020] According to an embodiment of the present invention, in step S1, at the current moment, the first moisture content data, inlet pressure, and outlet pressure at the molecular sieve outlet are determined by a first moisture analyzer installed at the molecular sieve outlet and pressure meters installed at the molecular sieve inlet and outlet. At the current moment, a moisture analyzer (e.g., a Karl Fischer moisture meter, a near-infrared moisture meter) is installed at the molecular sieve outlet, which is the first moisture analyzer, and detects the moisture content at the molecular sieve outlet, which is the first moisture content data. Pressure meters (e.g., mechanical pressure meters, electronic pressure meters, etc.) are installed at the molecular sieve inlet and outlet, and measure the pressure at the molecular sieve inlet and outlet, which are the inlet pressure and outlet pressure.

[0021] In this way, based on the set first moisture analyzer and pressure meter, the first moisture content data, inlet pressure and outlet pressure at the molecular sieve outlet can be obtained, providing basic data for determining whether to close the molecular sieve channel.

[0022] Figure 2 An exemplary flowchart illustrating the determination of whether to close the molecular sieve channel according to an embodiment of the present invention is shown.

[0023] According to an embodiment of the present invention, in step S2, determining whether to close the molecular sieve channel based on first moisture content data, inlet pressure, and outlet pressure at multiple time points includes: step S21, acquiring first moisture content data, inlet pressure, and outlet pressure at multiple time points from the first time point after the previous purging of the molecular sieve to the current time point; step S22, assembling the first moisture content data, inlet pressure, and outlet pressure at each time point into a first judgment input vector for each time point; step S23, inputting the first judgment input vector at the i-th time point into a first fully connected layer for processing to obtain the first judgment feature vector at the i-th time point; S24, input the first judgment feature vector at time i and the first hidden state vector at time i-1 into the molecular sieve saturation judgment model to obtain the first hidden state vector at time i, where the first hidden state vector at time i-1 is a zero vector when i=1, and i is a positive integer; Step S25, iteratively execute the process of obtaining the first hidden state vector until the first hidden state vector at the current time is obtained, and input the first hidden state vector at the current time into the second fully connected layer and the first activation layer for processing to obtain the molecular sieve saturation probability information; Step S26, determine whether to close the molecular sieve channel based on the molecular sieve saturation probability information.

[0024] According to an embodiment of the present invention, in step S21, first moisture content data, inlet pressure, and outlet pressure are acquired at multiple times between the first moment after the previous purging of the molecular sieve and the current moment. The time interval between adjacent moments can be 1 hour. The time range required for the molecular sieve to reach saturation (e.g., 8 to 24 hours) is large and uncertain. For example, if the current moment is the first moment after the first purging of the molecular sieve, only one set of first moisture content data and pressure data can be obtained. If the time interval between the first moment after the previous purging of the molecular sieve and the current moment is 12 hours, and the molecular sieve has not yet reached saturation at the current moment, then 12 sets of first moisture content data and pressure data can be obtained. The present invention does not limit this.

[0025] According to an embodiment of the present invention, in step S22, the first moisture content data, inlet pressure, and outlet pressure at each time moment are combined to form a first judgment input vector at each time moment. The first moisture content data, inlet pressure, and outlet pressure at each time moment can all form a three-dimensional vector, that is, the first judgment input vector.

[0026] According to an embodiment of the present invention, in step S23, the first judgment input vector at time i is input into the first fully connected layer for processing to obtain the first judgment feature vector at time i. Since a three-dimensional vector expresses limited information and cannot fully reflect the saturation level of the molecular sieve, making it difficult to determine whether to close the molecular sieve channel, the first judgment input vector at time i can be input into the first fully connected layer to obtain a high-dimensional vector (e.g., a 128-dimensional vector), which is the first judgment feature vector at time i. This first judgment feature vector can describe the saturation level of the molecular sieve from more perspectives, thereby more accurately determining whether to close the molecular sieve channel.

[0027] According to an embodiment of the present invention, in step S24, the first judgment feature vector at time i and the first hidden state vector at time i-1 are input into the molecular sieve saturation judgment model to obtain the first hidden state vector at time i. Here, when i=1, the first hidden state vector at time i-1 is a zero vector, and i is a positive integer. Inputting the first judgment feature vector at time i and the first hidden state vector at time i-1 into the molecular sieve saturation judgment model yields the first hidden state vector at time i. The molecular sieve saturation judgment model can be an LSTM recurrent neural network model. The first hidden state vector at time i-1 is the vector output after inputting the first hidden state vector at time i-2 and the first judgment feature vector at time i-1 into the molecular sieve saturation judgment model. When i=1 (i.e., the i-th time is the first time after the previous purging of the molecular sieve), the first hidden state vector at time i-1 is a vector of all zeros.

[0028] According to an embodiment of the present invention, in step S25, the process of obtaining the first hidden state vector is iteratively executed until the first hidden state vector at the current moment is obtained. The first hidden state vector at the current moment is then input into the second fully connected layer and the first activation layer for processing to obtain the molecular sieve saturation probability information. During the iterative execution of the above process of obtaining the first hidden state vector, the first hidden state vector obtained at each moment is combined with the first hidden state vector at the previous moment. Therefore, in the process of obtaining the first hidden state vector at the current moment, the first hidden state vectors at all previous moments are comprehensively utilized. That is, the first hidden state vector at the current moment is combined with the first hidden state vectors at all previous moments and compared with them, taking into account the overall trend of change. This allows the first hidden state vector at the current moment to accurately describe the saturation status of the molecular sieve at the current moment, thereby more accurately determining whether to close the molecular sieve channels. Inputting the first hidden state vector at the current moment into the second fully connected layer converts the first hidden state vector at the current moment into a 1-dimensional vector, which can be used to describe whether the molecular sieve is saturated at the current moment. By inputting the aforementioned 1D vector into the activation layer for processing (for example, using the sigmoid function to process the data in the 1D vector), the data in the 1D vector can be mapped to the range [0,1]. This can be considered as the probability of molecular sieve saturation and can be used as molecular sieve saturation probability information. The larger the molecular sieve saturation probability information, the greater the probability that the molecular sieve has reached saturation.

[0029] According to an embodiment of the present invention, in step S26, it is determined whether to close the molecular sieve channel based on the molecular sieve saturation probability information. For example, when the molecular sieve saturation probability information is greater than 0.5, it can be considered that the molecular sieve is saturated and the molecular sieve channel needs to be closed; conversely, when the molecular sieve saturation probability information is less than 0.5, it can be considered that the molecular sieve has not reached a saturated state and the molecular sieve channel does not need to be closed.

[0030] According to an embodiment of the present invention, the above molecular sieve saturation judgment model can be trained before use. The training steps of the molecular sieve saturation judgment model include: acquiring first training moisture content data, training inlet pressure, and training outlet pressure at multiple training moments after purging the molecular sieve; acquiring first sampling moisture content data, sampling inlet pressure, and sampling outlet pressure at subdivided sampling moments between each training moment; determining the molecular sieve saturation labeling information at the j-th training moment based on the first training moisture content data, training inlet pressure, and training outlet pressure at the j-th training moment, and the first sampling moisture content data, sampling inlet pressure, and sampling outlet pressure at multiple subdivided sampling moments between the j-th training moment and the (j+1)-th training moment; and obtaining the training judgment for the j-th training moment and the training moments before the j-th training moment based on the first training moisture content data, training inlet pressure, and training outlet pressure at the j-th training moment and the training moments before the j-th training moment. The input vector is processed through the first fully connected layer to obtain the training judgment feature vector for the j-th training time and previous training time points. The molecular sieve saturation judgment model is then used to process the training judgment feature vector for the j-th training time and previous training time points to obtain the training hidden state vector for the j-th training time. This training hidden state vector is then input into the second fully connected layer and the first activation layer for further processing to obtain the molecular sieve saturation prediction probability information for the j-th training time. Based on the molecular sieve saturation labeling information and the molecular sieve saturation prediction probability information for the j-th training time, as well as the first sampled moisture content data, sampling inlet pressure, and sampling outlet pressure for multiple subdivided sampling times between the j-th and j+1-th training times, the loss function of the molecular sieve saturation judgment model is obtained. The molecular sieve saturation judgment model is then trained using this loss function to obtain the trained molecular sieve saturation judgment model.

[0031] According to embodiments of the present invention, similar to acquiring first moisture content data, inlet pressure, and outlet pressure, first training moisture content data, training inlet pressure, and training outlet pressure can be acquired at multiple training moments after purging the molecular sieve. The time interval between these training moments is the same as the time interval between the multiple moments, for example, 1 hour for both. When the first training moisture content is high, or the pressure difference between the training inlet and outlet pressures is large, the molecular sieve can be considered saturated. However, the moment when the molecular sieve reaches saturation may not coincide with the training moment. For example, the molecular sieve may be saturated 5 minutes after the previous training moment, but it may not be discovered until 55 minutes later at the current training moment that the molecular sieve has reached saturation, making timely intervention impossible and resulting in excessive impurities being introduced into the production of anhydrous ethanol. Therefore, when only focusing on whether the molecular sieve has reached saturation at each training moment, the long time interval between each training moment will affect the accuracy of determining when the molecular sieve has reached saturation, thereby affecting the purity of the produced anhydrous ethanol and production efficiency. Therefore, by subdividing the sampling time between each training moment, the first sample moisture content data, sampling inlet pressure, and sampling outlet pressure can be obtained. The time interval between these subdivided sampling moments is relatively small (e.g., 5 minutes or 2 minutes). This allows for the detection of the moisture content data, inlet pressure, and outlet pressure of the molecular sieve at multiple subdivided sampling moments between each training moment, further determining whether the molecular sieve has reached saturation at any of the subdivided sampling moments between training moments, thus more accurately determining when the molecular sieve reaches saturation. This invention does not limit the time interval of the subdivided sampling moments.

[0032] According to an embodiment of the present invention, the molecular sieve saturation labeling information at the j-th training time is determined based on the first training moisture content data, training inlet pressure, and training outlet pressure at the j-th training time, and the first sampling moisture content data, sampling inlet pressure, and sampling outlet pressure at multiple subdivided sampling times between the j-th and j+1-th training times. This includes: if the first training moisture content data at the j-th training time is higher than a first moisture content threshold, or the pressure difference between the training inlet pressure and the training outlet pressure is greater than a first pressure difference threshold, then the molecular sieve saturation labeling information at the j-th training time is determined to be molecular sieve saturated; or, if there exists a subdivided sampling time where the first sampling moisture content data is higher than the first moisture content threshold, or the pressure difference between the sampling inlet pressure and the sampling outlet pressure is greater than the first pressure difference threshold, then the molecular sieve saturation labeling information at the j-th training time is determined to be molecular sieve saturated; otherwise, the molecular sieve saturation labeling information at the j-th training time is determined to be molecular sieve unsaturated.

[0033] According to an embodiment of the present invention, the first moisture content threshold may be 1%. The first differential pressure threshold is determined based on the design differential pressure value of the molecular sieve. For example, the first differential pressure threshold for a 3A molecular sieve may be 0.12 bar / m, and the first differential pressure threshold for a 4A molecular sieve may be 0.24 bar / m. The present invention does not limit this. When the first training moisture content data at the j-th training moment is higher than the first moisture content threshold, or the differential pressure between the training inlet pressure and the training outlet pressure is greater than the first differential pressure threshold, it can be considered that the molecular sieve is in an abnormal working state and has reached saturation. Therefore, the molecular sieve saturation labeling information at the j-th training moment can be determined as molecular sieve saturation. If there is a subdivided sampling time between the j-th and j+1-th training times where the first sampled moisture content data is higher than the first moisture content threshold, or the pressure difference between the sampling inlet pressure and the sampling outlet pressure is greater than the first pressure difference threshold, the molecular sieve will be saturated at the (j+1)-th training time and unsaturated at the j-th training time. Therefore, intervention at the (j+1)-th training time is too late and will affect the purity of the produced anhydrous ethanol. Thus, the molecular sieve saturation label information at the j-th training time can be determined as molecular sieve saturation, indicating that the molecular sieve has reached saturation at the j-th training time, and timely intervention can be implemented. Otherwise, the molecular sieve can be considered unsaturated, and the molecular sieve saturation label information at the j-th training time can be determined as molecular sieve unsaturation, allowing continued use.

[0034] According to an embodiment of the present invention, the training moisture content data, training inlet pressure, and training outlet pressure of the first training time and the training time prior to the j-th training time are used to form the training judgment input vector of the training time prior to the j-th training time. After inputting the training judgment input vector of the training time prior to the j-th training time into the first fully connected layer to increase its dimensionality, the training judgment feature vector of the training time prior to the j-th training time can be obtained. The training judgment feature vector is then input into the molecular sieve saturation judgment model to obtain the training hidden state vector of the j-th training time (the acquisition method is similar to the first hidden state vector of the current time, and will not be repeated here). Then, the training hidden state vector of the j-th training time is input into the second fully connected layer to reduce it to a 1-dimensional vector, and input into the first activation layer to map the 1-dimensional vector to the range [0,1] to obtain the molecular sieve saturation prediction probability information of the j-th training time, which can describe the probability of molecular sieve saturation at the j-th training time and can be used to train the molecular sieve saturation judgment model.

[0035] According to an embodiment of the present invention, the loss function of the molecular sieve saturation judgment model is obtained based on the molecular sieve saturation labeling information and molecular sieve saturation prediction probability information at the j-th training time, and the first sampling moisture content data, sampling inlet pressure, and sampling outlet pressure at multiple subdivided sampling times between the j-th and j+1-th training times. This includes: if the molecular sieve saturation labeling information at the j-th training time is determined to indicate molecular sieve saturation, and the first training moisture content data at the j-th training time is lower than or equal to a first moisture content threshold, or the pressure difference between the training inlet pressure and the training outlet pressure is less than or equal to a first pressure difference threshold, then the sequence number of the first subdivided sampling time where the first sampling moisture content data is higher than the first moisture content threshold, or the pressure difference between the sampling inlet pressure and the sampling outlet pressure is greater than the first pressure difference threshold, is determined; and the loss function of the molecular sieve saturation judgment model is determined based on the molecular sieve saturation labeling information and molecular sieve saturation prediction probability information at the j-th training time, and the sequence number of the first subdivided sampling time.

[0036] According to an embodiment of the present invention, if the molecular sieve saturation labeling information at the j-th training time is determined to be molecular sieve saturation, and the first training moisture content data at the j-th training time is lower than or equal to the first moisture content threshold, and the pressure difference between the training inlet pressure and the training outlet pressure is less than or equal to the first pressure difference threshold (i.e., the molecular sieve is not saturated at the j-th training time, but the labeling information at the j-th training time is molecular sieve saturation), then it can be considered that there is a sub-sampling time when the molecular sieve is saturated between the j-th training time and the (j+1)-th training time. In this case, the sequence number of the first sub-sampling time when the first sampling moisture content data is higher than the first moisture content threshold, or the pressure difference between the sampling inlet pressure and the sampling outlet pressure is greater than the first pressure difference threshold, i.e., the sequence number of the first sub-sampling time when the molecular sieve is saturated, can be determined. For example, 30 sub-sampling times are set between the j-th and j+1-th training times. The first sub-sampling time where the first sampled moisture content data is higher than the first moisture content threshold, or the pressure difference between the sampling inlet pressure and the sampling outlet pressure is greater than the first pressure difference threshold, is the 5th sub-sampling time. Therefore, the sequence number of the first sub-sampling time when the molecular sieve becomes saturated is 5. Since the molecular sieve will remain saturated after saturation without intervention, the molecular sieve will be saturated at all sub-sampling times after the first sub-sampling time. Since the saturation level of molecular sieves gradually increases with use, the closer the first sub-sampling time when the molecular sieve is saturated is to the j-th training time (i.e., the smaller the index), the closer the molecular sieve is to saturation at the j-th training time. In this case, the j-th training time is more likely to be judged as saturated when training the molecular sieve saturation judgment model. Conversely, the farther the first sub-sampling time when the molecular sieve is saturated is from the j-th training time (i.e., the larger the index), the more difficult it is to judge the j-th training time as saturated. In other words, the greater the training difficulty, the more important it is when training the molecular sieve saturation judgment model.

[0037] According to an embodiment of the present invention, the loss function of the molecular sieve saturation judgment model is determined based on the molecular sieve saturation labeling information and molecular sieve saturation prediction probability information at the j-th training time, and the sequence number of the first subdivision sampling time, including: determining the loss function of the molecular sieve saturation judgment model according to formula (1). , (1) in, This provides the molecular sieve saturation prediction probability information for the j-th training time. Let be the probability value of molecular sieve saturation determined based on the molecular sieve saturation labeling information at the j-th training time. This represents the number of subdivision sampling times between the j-th training time and the (j+1)-th training time. Let N be the sequence number of the first subdivided sampling time when the first sampled moisture content data between the j-th training time and the (j+1)-th training time is higher than the first moisture content threshold, or when the pressure difference between the sampling inlet pressure and the sampling outlet pressure is greater than the first pressure difference threshold, where N is the number of training times, j≤N, and j, N, ... , All are positive integers.

[0038] According to an embodiment of the present invention, in formula (1), the probability value of molecular sieve saturation determined based on the molecular sieve saturation labeling information at the j-th training time is 0 or 1. When labeled as molecular sieve saturation, the probability value of molecular sieve saturation is 1, and when labeled as molecular formula unsaturated, the probability value of molecular sieve saturation is 0. For based on and The cross-entropy loss function can be adjusted during training to minimize the parameters of the molecular sieve saturation judgment model, thereby achieving better performance. and The error between them is reduced.

[0039] According to an embodiment of the present invention, in formula (1), The difference between the molecular sieve saturation prediction probability information at time 1 and time j can be considered as the magnitude of the error in the molecular sieve saturation prediction probability information at time j. The larger the magnitude of the error, the greater the difficulty for the molecular sieve saturation judgment model to accurately determine the molecular sieve saturation prediction probability information at time j. Therefore, the data at time j is more important for the training of the molecular sieve saturation judgment model. These can be used as weight coefficients in the cross-entropy loss function at the j-th training time, describing the importance of the molecular sieve saturation prediction probability information at the j-th training time for training the molecular sieve saturation judgment model. Because... The value of is in the range (0,1), therefore Less than 1, therefore The smaller the value, the more important the molecular sieve saturation prediction probability information at the corresponding training time is for training the molecular sieve saturation judgment model; the larger the above weight coefficients, that is, The smaller the value, The larger the value of , the more difficult it is to accurately determine saturation at training time j, where the first sampled moisture content data between training time j and training time j+1 is higher than the first moisture content threshold, or the pressure difference between the sampling inlet pressure and the sampling outlet pressure is greater than the first pressure difference threshold. In other words, the greater the training difficulty, the more important the molecular sieve saturation prediction probability information at training time j is in training the molecular sieve saturation judgment model. Therefore, a higher weighting coefficient is set. The weighted summation of the cross-entropy loss function corresponding to the molecular sieve saturation prediction probability information at each training time can be represented as the loss function of the molecular sieve saturation judgment model.

[0040] According to an embodiment of the present invention, a molecular sieve saturation judgment model is trained based on the loss function of the molecular sieve saturation judgment model. Backpropagation can be performed using the aforementioned loss function, and the parameters of the molecular sieve saturation judgment model can be adjusted using the gradient descent method to train the model. After multiple training iterations (i.e., training using molecular sieve saturation prediction probability information from multiple training times), training is complete, resulting in a trained molecular sieve saturation judgment model.

[0041] In this way, when training the molecular sieve saturation judgment model, considering the possibility that the molecular sieve may reach saturation between adjacent training times, multiple sub-sampling times are set between each training time to more accurately determine when the molecular sieve reaches saturation. Furthermore, considering that the larger the sequence number of the first sub-sampling time when the molecular sieve is saturated, the greater the training difficulty and the higher the importance of the training data, weights are set for the cross-entropy loss function corresponding to the molecular sieve saturation prediction probability information at each training time to obtain the loss function of the molecular sieve saturation judgment model. The trained molecular sieve saturation judgment model is then obtained, which improves the accuracy and relevance of the training and enhances the performance of the molecular sieve saturation judgment model.

[0042] According to an embodiment of the present invention, in step S3, if it is determined that the molecular sieve channel is closed, the superheater channel is opened and superheated steam is generated to purge the molecular sieve and obtain regenerated tail gas. If it is determined that the molecular sieve channel is closed, at which point the molecular sieve is saturated, the superheater channel can be opened to generate superheated steam to purge the molecular sieve, desorb the adsorbed moisture and other impurities in the molecular sieve, and obtain regenerated tail gas, which contains ethanol vapor and water vapor.

[0043] According to an embodiment of the present invention, in step S4, the regeneration tail gas and preheated raw material ethanol are fed together into a recovery tower to obtain ethanol saturated vapor. The regeneration tail gas containing ethanol vapor and the preheated raw material ethanol are then fed together into the recovery tower for distillation to obtain ethanol saturated vapor, wherein the ethanol purity of the ethanol saturated vapor is 95%.

[0044] According to an embodiment of the present invention, in step S5, the membrane filter channel is opened, and saturated ethanol vapor is delivered to the membrane filter to obtain first anhydrous ethanol vapor, which is then cooled and stored. When the molecular sieve channel is determined to be closed, the membrane filter channel can be opened to deliver saturated ethanol vapor to the membrane filter, thereby obtaining first anhydrous ethanol vapor, which is then cooled and stored. The resulting anhydrous ethanol has a purity of over 99.5%. The membrane filter channel can continue producing anhydrous ethanol during molecular sieve regeneration, ensuring uninterrupted production and improving equipment utilization and steam utilization.

[0045] According to an embodiment of the present invention, in step S6, the molecular sieve channel is determined to be opened at the next moment after purging. At the next moment after purging, it can be considered that the molecular sieve has been regenerated and can continue to be used, thereby opening the molecular sieve channel to produce anhydrous ethanol through the molecular sieve channel.

[0046] According to an embodiment of the present invention, in step S7, at the next moment after purging (the purging time is usually 30 minutes, so the purging can be completed within the time period between two adjacent moments, so the next moment after purging is the next moment after the current moment), before opening the molecular sieve channel and closing the membrane filter channel and the superheater channel, the second moisture content data obtained by the second moisture analyzer set at the membrane filter outlet at multiple moments, and the steam flux obtained by the flux detection device set at the membrane filter outlet at multiple moments, are used to determine whether the membrane filter needs to be cleaned.

[0047] According to an embodiment of the present invention, determining whether the membrane filter needs cleaning is achieved by using second moisture content data obtained at multiple times by a second moisture analyzer installed at the outlet of the membrane filter and steam flux obtained at multiple times by a flux detection device installed at the outlet of the membrane filter. This includes: obtaining second moisture content data and steam flux obtained at the next moment after the previous chemical cleaning and the moment when the membrane filter channel is opened, to determine whether the membrane filter needs cleaning.

[0048] According to an embodiment of the present invention, the next moment after the previous chemical cleaning and the opening of the membrane filter channel is obtained, that is, after the membrane filter can be used after one chemical cleaning, after each time the membrane filter channel is opened and used for a period of time (e.g., 1 hour), the second moisture content data is obtained by a second moisture analyzer installed at the membrane filter outlet, and the steam flux is obtained by a flux detection device (e.g., differential pressure flow meter, vortex flow meter, etc.) installed at the membrane filter outlet. For example, after the previous chemical cleaning, the membrane filter channel is opened three times. After each opening and use of the membrane filter channel, before closing the membrane filter channel, the second moisture content data and steam flux are obtained.

[0049] According to an embodiment of the present invention, similar to determining whether to close the molecular sieve channel, the second moisture content data and steam flux are used to form a judgment input vector, and the judgment input vector is processed using a membrane filter saturation judgment model to obtain membrane filter saturation probability information. When the membrane filter saturation probability information is greater than 0.5, the membrane filter can be considered saturated and the membrane filter channel needs to be closed for cleaning. When training the membrane filter saturation judgment model, a threshold for the second moisture content (e.g., 1%) and a threshold for steam flux (e.g., 7-8 kg / (m²·h)) can be preset in a manner similar to that used for training the molecular sieve saturation judgment model. If the moisture content data detected at the membrane filter outlet is higher than the second moisture content threshold or the steam flux is lower than the steam flux threshold, the membrane filter is judged to be saturated and needs to be cleaned. This determines the membrane filter saturation labeling information at each training time. For example, multiple subdivided sampling times can be set between adjacent training times. When the membrane filter is saturated at the j-th training time, the membrane filter saturation labeling information at the j-th training time can be determined as membrane filter saturation. Alternatively, if there are subdivided sampling times for membrane filter saturation between the j-th and j+1-th training times, the membrane filter saturation labeling information at the j-th training time can be determined as membrane filter saturation. Furthermore, based on the membrane filter saturation prediction probability information and membrane filter saturation labeling information at multiple training times, as well as the moisture content data and steam flux at the membrane filter outlet at subdivided sampling times between adjacent training times, the loss function of the membrane filter saturation judgment model is determined. The determination method is similar to that of the molecular sieve saturation judgment model, and will not be elaborated here. The membrane filter saturation judgment model is then trained based on its loss function to improve its performance.

[0050] According to an embodiment of the present invention, in step S8, if cleaning is required, chemical cleaning is performed. If cleaning is required, the membrane filter is chemically cleaned using a chemical solution (e.g., 0.1%-1% sodium hydroxide solution, 50-200 ppm sodium hypochlorite solution, etc.).

[0051] In this way, after confirming the molecular sieve channel is closed, the membrane filter channel can be opened. The regeneration exhaust gas obtained from purging the molecular sieve can be fed into the membrane filter along with the raw ethanol to produce ethanol-saturated vapor, improving the utilization rate of the ethanol raw material. Anhydrous ethanol can be continuously produced during the molecular sieve regeneration process, increasing equipment utilization. Before reopening the molecular sieve channel, it can be determined whether the membrane filter needs cleaning. If so, chemical cleaning is performed, ensuring that the membrane filter can continue to produce anhydrous ethanol when the molecular sieve becomes saturated again, preventing production interruptions and improving equipment and steam utilization. Alternating operation of the molecular sieve and membrane filter for continuous anhydrous ethanol production reduces gas consumption, increases equipment utilization and product yield, and enhances operational stability.

[0052] According to an embodiment of the present invention, in step S9, if it is determined that the molecular sieve channel is to be opened, then the membrane filter channel and the superheater channel are closed. The saturated ethanol vapor generated after the recovery tower processes the preheated raw ethanol is input into the molecular sieve to obtain the second anhydrous ethanol vapor, which is then cooled and stored. If it is determined that the molecular sieve channel is to be opened, then the membrane filter channel and the superheater channel are closed. Anhydrous ethanol is produced only through the molecular sieve channel. The saturated ethanol vapor generated after the recovery tower processes the preheated raw ethanol (i.e., saturated ethanol vapor excluding regeneration tail gas) is input into the molecular sieve to obtain the second anhydrous ethanol vapor, which is then cooled and stored. The purity of the obtained anhydrous ethanol is above 99.5%.

[0053] The anhydrous ethanol optimized production control method according to an embodiment of the present invention can determine the first moisture content, inlet pressure, and outlet pressure using a first moisture analyzer installed at the molecular sieve outlet and pressure meters installed at the molecular sieve inlet and outlet to determine whether the molecular sieve is saturated. Furthermore, it can determine whether the membrane filter needs cleaning by using the second moisture content data obtained from a second moisture analyzer installed at the membrane filter outlet and the steam flux obtained from a flux detection device installed at the membrane filter outlet. This allows for accurate judgment of whether the molecular sieve is saturated and whether the membrane filter needs cleaning. It also allows for the recovery and regeneration of tail gas, and the generation of anhydrous ethanol vapor through the membrane filter and molecular sieve, reducing steam consumption, improving equipment and steam utilization, increasing product yield, and enhancing operational stability. During data collection, the first moisture content data, inlet pressure, and outlet pressure at the molecular sieve outlet can be obtained based on the installed first moisture analyzer and pressure meters, providing basic data for determining whether to close the molecular sieve channel. When training the molecular sieve saturation judgment model, considering the possibility that the molecular sieve may reach saturation between adjacent training moments, multiple subdivided sampling moments were set between each training moment to more accurately determine when the molecular sieve reaches saturation. Furthermore, considering that a larger sequence number of the first subdivided sampling moment when the molecular sieve reaches saturation indicates greater training difficulty and higher importance of the training data, weights were assigned to the cross-entropy loss function corresponding to the molecular sieve saturation prediction probability information at each training moment, resulting in the loss function of the molecular sieve saturation judgment model. This trained molecular sieve saturation judgment model improved the accuracy and relevance of the training, thus enhancing its performance. After confirming the closure of the molecular sieve channel, the membrane filter channel can be opened. The regeneration tail gas obtained from purging the molecular sieve can be input into the membrane filter along with the raw material ethanol to obtain ethanol saturated vapor, improving the utilization rate of the ethanol raw material. Anhydrous ethanol can be continuously produced during the molecular sieve regeneration process, increasing equipment utilization. Before reopening the molecular sieve channel, it can be determined whether the membrane filter needs cleaning. If so, chemical cleaning is performed, ensuring that the membrane filter can continue to produce anhydrous ethanol when the molecular sieve becomes saturated again, preventing production interruptions and improving equipment and steam utilization. Alternating operation of molecular sieves and membrane filters enables continuous production of anhydrous ethanol, which reduces gas consumption, increases equipment utilization and product yield, and improves operational stability.

[0054] Figure 3 An exemplary block diagram of an anhydrous ethanol optimized production control system according to an embodiment of the present invention is shown, the system comprising: The data acquisition module, at the current moment, determines the first moisture content data, inlet pressure and outlet pressure at the molecular sieve outlet by using the first moisture analyzer set at the molecular sieve outlet and the pressure meters set at the molecular sieve inlet and outlet. The first judgment module determines whether to close the molecular sieve channel based on the first moisture content data, inlet pressure, and outlet pressure at multiple times. If the molecular sieve channel is closed, the superheater channel is opened in the regenerated exhaust gas module, and superheated steam is generated to purge the molecular sieve and obtain regenerated exhaust gas. The ethanol saturated steam module feeds the regeneration tail gas and preheated raw ethanol into the recovery tower to obtain ethanol saturated steam. The first anhydrous ethanol vapor module opens the membrane filter channel, delivers saturated ethanol vapor to the membrane filter, obtains the first anhydrous ethanol vapor, and stores it after cooling. The molecular sieve channel opening module determines to open the molecular sieve channel at the next moment after purging. The second judgment module determines whether the membrane filter needs cleaning by using the second moisture content data obtained at multiple times by the second moisture analyzer set at the membrane filter outlet and the steam flux obtained at multiple times by the flux detection device set at the membrane filter outlet. The chemical cleaning module performs chemical cleaning if necessary. In the second anhydrous ethanol vapor module, if the molecular sieve channel is determined to be open, the membrane filter channel and the superheater channel are closed. The saturated ethanol vapor generated by the recovery tower after processing the preheated raw ethanol is input into the molecular sieve to obtain the second anhydrous ethanol vapor, which is then cooled and stored.

[0055] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0056] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments, and any variations or modifications may be made to the implementation of the present invention without departing from the stated principles.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing the production control of anhydrous ethanol, characterized in that, include: At the current moment, the first moisture content data at the molecular sieve outlet, the inlet pressure and the outlet pressure are determined by the first moisture analyzer installed at the molecular sieve outlet and the pressure meters installed at the molecular sieve inlet and outlet. Based on the initial moisture content data, inlet pressure, and outlet pressure at multiple time points, determine whether to close the molecular sieve channel; If it is determined that the molecular sieve channel is closed, the superheater channel is opened and superheated steam is generated to purge the molecular sieve and obtain regenerated tail gas. The regeneration tail gas and the preheated raw ethanol are fed together into the recovery tower to obtain saturated ethanol vapor. Open the membrane filter channel and deliver saturated ethanol vapor to the membrane filter to obtain the first anhydrous ethanol vapor, which is then cooled and stored in the warehouse. The molecular sieve channel is activated immediately after purging. The second moisture content data obtained by the second moisture analyzer installed at the outlet of the membrane filter at multiple times, and the steam flux obtained by the flux detection device installed at the outlet of the membrane filter at multiple times, are used to determine whether the membrane filter needs to be cleaned. If cleaning is required, chemical cleaning should be performed. If it is determined that the molecular sieve channel is to be opened, the membrane filter channel and the superheater channel are closed. The saturated ethanol vapor generated by the preheated raw ethanol in the recovery tower is fed into the molecular sieve to obtain the second anhydrous ethanol vapor, which is then cooled and stored.

2. The optimized production control method for anhydrous ethanol according to claim 1, characterized in that, Based on initial moisture content data, inlet pressure, and outlet pressure at multiple time points, determine whether to close the molecular sieve channel, including: Acquire the first moisture content data, inlet pressure, and outlet pressure at multiple times between the first moment after the previous purging of the molecular sieve and the current moment; The first moisture content data, inlet pressure, and outlet pressure at each time moment are combined to form the first judgment input vector at each time moment; The first judgment input vector at time i is input into the first fully connected layer for processing to obtain the first judgment feature vector at time i. Input the first judgment feature vector at time i and the first hidden state vector at time i-1 into the molecular sieve saturation judgment model to obtain the first hidden state vector at time i. When i=1, the first hidden state vector at time i-1 is a zero vector, and i is a positive integer. The process of obtaining the first hidden state vector is iteratively executed until the first hidden state vector at the current time is obtained. The first hidden state vector at the current time is then input into the second fully connected layer and the first activation layer for processing to obtain the molecular sieve saturation probability information. Based on the molecular sieve saturation probability information, determine whether to close the molecular sieve channel.

3. The optimized production control method for anhydrous ethanol according to claim 2, characterized in that, The training steps for the molecular sieve saturation judgment model include: At multiple training moments following the purging of the molecular sieve, the first training moisture content data, training inlet pressure, and training outlet pressure were acquired. At each training moment, the first sampled moisture content data, sampling inlet pressure, and sampling outlet pressure are obtained through subdivided sampling moments. Based on the first training moisture content data, training inlet pressure and training outlet pressure at the j-th training time, and the first sampling moisture content data, sampling inlet pressure and sampling outlet pressure at multiple subdivided sampling times between the j-th and j+1-th training times, determine the molecular sieve saturation labeling information at the j-th training time. Based on the first training moisture content data, training inlet pressure and training outlet pressure at the j-th training time and the training time before it, the training judgment input vector at the j-th training time and the training time before it is obtained, and then processed by the first fully connected layer to obtain the training judgment feature vector at the j-th training time and the training time before it. The training judgment feature vectors of the j-th training time and the training time before it are processed by the molecular sieve saturation judgment model to obtain the training hidden state vector of the j-th training time. The training hidden state vector at the j-th training time is input into the second fully connected layer and the first activation layer for processing to obtain the molecular sieve saturation prediction probability information at the j-th training time. Based on the molecular sieve saturation labeling information and molecular sieve saturation prediction probability information at the j-th training time, as well as the first sampled moisture content data, sampling inlet pressure and sampling outlet pressure at multiple subdivided sampling times between the j-th and j+1-th training times, the loss function of the molecular sieve saturation judgment model is obtained. Based on the loss function of the molecular sieve saturation judgment model, the molecular sieve saturation judgment model is trained to obtain the trained molecular sieve saturation judgment model.

4. The anhydrous ethanol optimized production control method according to claim 3, characterized in that, Based on the first training moisture content data, training inlet pressure, and training outlet pressure at the j-th training time, and the first sampling moisture content data, sampling inlet pressure, and sampling outlet pressure at multiple subdivided sampling times between the j-th and j+1-th training times, the molecular sieve saturation labeling information at the j-th training time is determined, including: If the first training moisture content data at the j-th training time is higher than the first moisture content threshold, or the pressure difference between the training inlet pressure and the training outlet pressure is greater than the first pressure difference threshold, then the molecular sieve saturation labeling information at the j-th training time is determined as molecular sieve saturation. Alternatively, if there is a subdivided sampling time where the first sampled moisture content data is higher than the first moisture content threshold, or the pressure difference between the sampling inlet pressure and the sampling outlet pressure is greater than the first pressure difference threshold, then the molecular sieve saturation labeling information at the j-th training time is determined as molecular sieve saturation. Otherwise, the molecular sieve saturation labeling information at the j-th training time is determined as molecular sieve unsaturation.

5. The optimized production control method for anhydrous ethanol according to claim 4, characterized in that, Based on the molecular sieve saturation labeling information and molecular sieve saturation prediction probability information at the j-th training time, and the first sampled moisture content data, sampling inlet pressure, and sampling outlet pressure at multiple subdivided sampling times between the j-th and j+1-th training times, the loss function of the molecular sieve saturation judgment model is obtained, including: If the molecular sieve saturation labeling information at the j-th training time is determined to be molecular sieve saturation, and the first training moisture content data at the j-th training time is lower than or equal to the first moisture content threshold, and the pressure difference between the training inlet pressure and the training outlet pressure is less than or equal to the first pressure difference threshold, then the sequence number of the first subdivision sampling time is determined to be the first sampling time when the first sampling moisture content data is higher than the first moisture content threshold, or the pressure difference between the sampling inlet pressure and the sampling outlet pressure is greater than the first pressure difference threshold. Based on the molecular sieve saturation labeling information and molecular sieve saturation prediction probability information at the j-th training time, and the sequence number of the first subdivision sampling time, the loss function of the molecular sieve saturation judgment model is determined.

6. The optimized production control method for anhydrous ethanol according to claim 5, characterized in that, Based on the molecular sieve saturation labeling information and molecular sieve saturation prediction probability information at the j-th training time, and the sequence number of the first subdivision sampling time, the loss function of the molecular sieve saturation judgment model is determined, including: According to the formula Determine the loss function of the molecular sieve saturation judgment model ,in, This provides the molecular sieve saturation prediction probability information for the j-th training time. Let be the probability value of molecular sieve saturation determined based on the molecular sieve saturation labeling information at the j-th training time. This represents the number of subdivision sampling times between the j-th training time and the (j+1)-th training time. Let N be the sequence number of the first subdivided sampling time when the first sampled moisture content data between the j-th training time and the (j+1)-th training time is higher than the first moisture content threshold, or when the pressure difference between the sampling inlet pressure and the sampling outlet pressure is greater than the first pressure difference threshold, where N is the number of training times, j≤N, and j, N, ... , All are positive integers.

7. The anhydrous ethanol optimized production control method according to claim 1, characterized in that, The need for membrane filter cleaning is determined by using second moisture content data obtained at multiple times from a second moisture analyzer installed at the membrane filter outlet, and steam flux data obtained at multiple times from a flux detection device installed at the membrane filter outlet. This includes: The second moisture content data and steam flux are obtained at the moment immediately following the last chemical cleaning and the moment the membrane filter channel is opened, to determine whether the membrane filter needs cleaning.

8. An optimized production control system for anhydrous ethanol, characterized in that, include: The data acquisition module, at the current moment, determines the first moisture content data, inlet pressure and outlet pressure at the molecular sieve outlet by using the first moisture analyzer set at the molecular sieve outlet and the pressure meters set at the molecular sieve inlet and outlet. The first judgment module determines whether to close the molecular sieve channel based on the first moisture content data, inlet pressure, and outlet pressure at multiple times. If the molecular sieve channel is closed, the superheater channel is opened in the regenerated exhaust gas module, and superheated steam is generated to purge the molecular sieve and obtain regenerated exhaust gas. The ethanol saturated steam module feeds the regeneration tail gas and preheated raw ethanol into the recovery tower to obtain ethanol saturated steam. The first anhydrous ethanol vapor module opens the membrane filter channel, delivers saturated ethanol vapor to the membrane filter, obtains the first anhydrous ethanol vapor, and stores it after cooling. The molecular sieve channel opening module determines to open the molecular sieve channel at the next moment after purging. The second judgment module determines whether the membrane filter needs cleaning by using the second moisture content data obtained at multiple times by the second moisture analyzer set at the membrane filter outlet and the steam flux obtained at multiple times by the flux detection device set at the membrane filter outlet. The chemical cleaning module performs chemical cleaning if necessary. In the second anhydrous ethanol vapor module, if the molecular sieve channel is determined to be open, the membrane filter channel and the superheater channel are closed. The saturated ethanol vapor generated by the recovery tower after processing the preheated raw ethanol is input into the molecular sieve to obtain the second anhydrous ethanol vapor, which is then cooled and stored.

Images