Reaction condition optimization control method and system for biological aviation kerosene production

Abstract

The invention discloses a reaction condition optimization control method and system for biological aviation kerosene production, and particularly relates to the field of biological aviation kerosene production. The method comprises the following steps: firstly, acquiring parameters such as temperature, pressure, flow, catalyst activity and raw material components of biological aviation kerosene in real time through a sensor module consisting of a temperature sensor, a pressure sensor, a flow sensor and the like, and then carrying out pretreatment operations such as median filtering and wavelet denoising on acquired data; constructing an LSTM-Transformer hybrid model based on the preprocessed data to predict the second-by-second value of each parameter in the future 10 minutes, correcting the predicted value according to a catalyst activity index formula, further optimizing the prediction model by using a shadow model, comparing the predicted value with a preset interval and a threshold value, judging the state of the predicted value, and determining the second-by-second value of each parameter in the future 10 minutes. And finally, generating a fixed fine-tuning, strengthening or emergency control strategy according to a judgment result. According to the method, the biological aviation kerosene production reaction conditions can be accurately predicted, regulated and controlled, and the production stability, safety and product quality are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of bio - jet fuel production. More specifically, the present invention relates to a method and system for optimizing the control of reaction conditions for bio - jet fuel production. Background Art

[0002] With the continuous growth of the global demand for clean energy, bio - jet fuel, as a sustainable aviation fuel, has attracted much attention in its production technology. In the production process of bio - jet fuel, the precise control of reaction conditions plays a decisive role in product quality, production efficiency, and production cost. In the traditional bio - jet fuel production process, the control of reaction conditions mainly relies on the experience of operators and simple monitoring equipment. Operators set reaction parameters such as temperature, pressure, flow rate, etc. based on past production experience, and at the same time use basic sensors to monitor these parameters.

[0003] However, in actual use, there are still some disadvantages. For example, the model structure adopted in the prior art is simple, unable to effectively capture the complex coupling relationships among various parameters in the bio - jet fuel production process, and also difficult to handle the long - term dependence relationships in time - series data, resulting in a low prediction accuracy of future reaction parameters and unable to meet the requirements of precise control of reaction conditions in actual production. The attenuation of catalyst activity will directly affect the reaction rate and product yield. The prior art cannot monitor the catalyst activity in real - time and accurately, nor establish an effective mechanism to adjust the reaction conditions in a timely manner according to the change of catalyst activity. The model update mechanism of the prior art is imperfect and cannot optimize and update the model in a timely manner according to newly collected data. As time goes by, the prediction error of the model gradually increases, unable to adapt to the actual production, affecting the accuracy and effectiveness of reaction condition control. The control strategy of the prior art is relatively single, lacking the ability to classify and control according to different reaction parameter states. When the reaction parameters are in different intervals, precise and effective control measures cannot be taken, unable to ensure that the reaction conditions are always in the optimal state, and affecting the stability and safety of production. Summary of the Invention

[0004] In order to overcome the above - mentioned defects of the prior art, the embodiments of the present invention provide a method and system for optimizing the control of reaction conditions for bio - jet fuel production, through the following solutions, to solve the problems raised in the above - mentioned background art.

[0005] To achieve the above object, the present invention provides the following technical solution: A method for optimizing the control of reaction conditions for bio - jet fuel production, including, S1: Bio - jet fuel data acquisition: Real - time acquisition of bio - jet fuel parameters through a sensor module; The sensor module includes a temperature sensor, a pressure sensor, a flow sensor, a component analyzer, and a catalyst activity detector; The parameters of the bio-jet fuel include temperature, pressure, flow rate, catalyst activity, and raw material components; S2: Data preprocessing: Perform preprocessing operations on the bio-jet fuel parameters obtained in step S1 to obtain a preprocessed text of the bio-jet fuel; S3: Construction of the LSTM-Transformer hybrid model: Construct multi-dimensional vectors based on the preprocessed text of the bio-jet fuel, and generate time series data based on the multi-dimensional vectors and input them into the LSTM model and the Transformer model to obtain the predicted values of each bio-jet fuel parameter per second within the next 10 minutes; S4: Catalyst activity decay correction: Calculate the catalyst activity index through the catalyst activity index formula, compare the catalyst activity index with a preset catalyst activity index threshold. If the value of the catalyst activity index is lower than the preset proportion of the threshold, then add the catalyst activity change information as a vector to step S3 to obtain new predicted values; S5: Shadow model correction: Based on the bio-jet fuel parameters newly collected within 1 hour, train the shadow model. After the training is completed, compare the predicted values output by the main model with the actual values. When the deviation exceeds the update threshold, update the parameters of the shadow model to the main model; S6: Analysis and judgment of predicted values: Calculate the deviation between the predicted values obtained in step S4 and a preset optimal parameter interval and compare it with a safety threshold. According to the comparison result, judge the state of the predicted values; S7: Generation of control strategies: Generate control strategies based on the comparison result of step S6.

[0006] Preferably, a reaction condition optimization control system for bio-jet fuel production includes: Sensor module: Composed of temperature, pressure, flow rate sensors, component analyzers, and catalyst activity detectors, responsible for real-time collection of bio-jet fuel parameters during the bio-jet fuel production process; Data preprocessing module: Includes median filtering, moving average filtering, wavelet denoising, and Min-Max normalization units, used to denoise and normalize the data collected by the sensors to obtain a preprocessed text of the bio-jet fuel; Model construction module: Includes a multi-dimensional vector construction unit, an LSTM model unit, and a Transformer model unit. The multi-dimensional vector construction unit combines the preprocessed data into multi-dimensional vectors and generates time series data. The LSTM model unit and the Transformer model unit work together to obtain the predicted values of each bio-jet fuel parameter per second within the next 10 minutes; Catalyst activity correction module: Used to calculate the catalyst activity index, and compare the catalyst activity index with a preset threshold. If it is lower than the preset proportion of the threshold, then feedback the catalyst activity change information to the model to obtain new predicted values; Shadow model update module: It includes a shadow model training unit and a comparison and update unit. The shadow model training unit trains the shadow model based on the newly collected 1-hour data using the stochastic gradient descent algorithm. The comparison and update unit compares the predicted value of the main model with the actual value, and when the deviation exceeds the threshold, updates the parameters of the main model; Predicted value analysis module: It includes a deviation calculation unit and a comparison and judgment unit. The deviation calculation unit is used to calculate the deviation between the predicted value and the boundary value of the optimal parameter interval, and the comparison and judgment unit judges the state of the predicted value in combination with the safety threshold; Control strategy execution module: It includes a fine-tuning control unit, a reinforcement control unit, and an emergency control unit, which are used to generate control strategies according to the results of the predicted value analysis module.

[0007] Technical effects and advantages of the present invention: 1. The present invention constructs a multi-dimensional vector based on the bio-jet fuel pretreatment text and uses an LSTM-Transformer hybrid model for prediction. The LSTM model can effectively capture the long-term dependence relationship of reaction parameters in the time series, and the multi-head attention mechanism of the Transformer model can capture the coupling relationship between bio-jet fuel parameters from different angles. The two work together to improve the prediction accuracy of each bio-jet fuel parameter every second within the next 10 minutes; 2. The present invention calculates the catalyst activity index through the catalyst activity index formula and compares it with a preset threshold. When the catalyst activity index is lower than the preset ratio of the threshold, the catalyst activity change information is fed back to the model, so as to obtain a more accurate predicted value and adjust the reaction conditions in a timely manner according to this, ensuring that the catalyst is always in the best working state and improving the reaction rate and product yield; 3. The present invention uses the shadow model to train and update every 1 hour, adopts the stochastic gradient descent algorithm and the mini-batch gradient descent method, uses the mean square error as the loss function, and optimizes the model according to the bio-jet fuel parameters collected within 1 hour. When the deviation between the predicted value of the main model and the actual value exceeds the update threshold, the parameters of the shadow model are updated to the main model to ensure that the model can adapt to the changes in the production process in a timely manner and maintain a high prediction accuracy; 4. According to the comparison results of the predicted value with the preset optimal parameter interval and safety threshold, the present invention classifies the control strategies into three cases: the predicted value is within the optimal interval, exceeds the optimal interval but does not reach the safety threshold, and exceeds the safety threshold. For different situations, fixed fine-tuning rules, preset reinforcement control strategies, and emergency control strategies are respectively adopted to achieve precise control of the reaction conditions and ensure the stability and safety of the production process. Description of the Drawings

[0008] Figure 1 It is a schematic diagram of the overall structure of the present invention; Figure 2 Schematic diagram of the system structure of the present invention; Figure 3 Temperature prediction accuracy graph of the present invention; Figure 4 Loss graph of the shadow model of the present invention. Specific embodiments

[0009] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0010] As shown in the attached Figure 1 A reaction condition optimization control method for bio - jet fuel production, including: S1: Bio - jet fuel data collection: Real - time acquisition of bio - jet fuel parameters through a sensor module; The sensor module includes a temperature sensor, a pressure sensor, a flow sensor, a component analyzer, and a catalyst activity detector; It should be further noted that the installation method of the temperature sensor is as follows: A total of 5 Omega OS316K type thermocouple sensors are arranged at the top, middle, bottom of the reactor and near the catalyst loading area. The top sensor is used to monitor the temperature of the reaction gas phase space, the middle and bottom sensors respectively monitor the temperatures of different heights of the reaction liquid, and the sensor near the catalyst area can accurately reflect the temperature near the active center of the catalyst. Each sensor collects data at an interval of 0.1 second, with an accuracy of ±0.5 °C, and can quickly capture the subtle changes in the reaction temperature, such as the sharp rise in temperature at the initial stage of the reaction or the local temperature fluctuations during the reaction process. These temperature data are crucial for understanding the reaction process and judging whether the reaction is proceeding normally, because the change in temperature directly affects the reaction rate and the quality of the product; The installation method of the pressure sensor is as follows: One Honeywell STG series pressure sensor is installed on the inlet pipeline, outlet pipeline of the reactor and at the 1 / 3 and 2 / 3 heights from the bottom of the side wall of the reaction kettle. The pressure sensor on the inlet pipeline monitors the pressure when the raw material enters the reactor, the pressure sensor on the outlet pipeline detects the pressure of the product discharge, and the two sensors on the side wall of the reaction kettle can obtain the pressure distribution at different heights in the reaction kettle, so as to more comprehensively understand the reaction pressure state. As one of the key parameters in the reaction process, the stability of pressure is directly related to the safety of the reaction and the yield of the product. Through the arrangement of these sensors, the pressure change can be monitored in real time, providing an important basis for adjusting the reaction conditions; The installation method of the flow sensor is as follows: Yokogawa AXF series electromagnetic flowmeters are installed in the raw material feed pipeline and the product discharge pipeline respectively. The flowmeter in the feed pipeline monitors the input flow of the raw materials for bio-jet fuel production in real time to ensure that the raw materials enter the reactor according to the predetermined ratio. The flowmeter in the discharge pipeline is used to monitor the product output flow, based on which the reaction progress and production efficiency can be evaluated. The precise control of the flow is crucial for ensuring the stability and consistency of the reaction. By accurately measuring the flow, problems that may occur during the raw material supply or product discharge process can be detected in a timely manner, so as to take corresponding measures for adjustment; The installation method of the component analyzer is as follows: An Agilent 7890B gas chromatograph is connected to the outlet pipeline of the raw material storage tank, and raw material samples are automatically extracted for detection every 5 minutes. This instrument can accurately analyze the component contents such as fatty acid methyl ester, glycerol, moisture, and impurities in the raw materials to help understand the characteristics of the raw materials. The components of the raw materials directly affect the progress of the reaction and the quality of the products. For example, the quality of the raw materials can be judged by monitoring the fatty acid methyl ester content. Excessive glycerol content may affect the reaction effect, and the moisture and impurity contents are related to the reaction safety and product quality. Understanding the changes in the raw material components in a timely manner can optimize the reaction conditions in advance to ensure that the reaction proceeds under the best conditions; The installation method of the catalyst activity detector is as follows: A special catalyst sampling port is set inside the reactor, and a small amount of catalyst samples are obtained through the sampling device every hour and sent to the on-line detector. The detector evaluates the catalyst activity by detecting parameters such as the specific surface area, pore size distribution, and catalytic conversion rate of the catalyst. The catalyst plays a key role in the production process of bio-jet fuel, and its activity level directly affects the reaction rate and product yield. The specific surface area and pore size distribution reflect the physical structure of the catalyst, directly affecting its contact area with the reactants and the reaction rate, and the catalytic conversion rate intuitively reflects the promotion effect of the catalyst on the bio-jet fuel production reaction. The changes in these parameters can timely reflect the activity state of the catalyst, providing a reference for adjusting the reaction conditions and replacing the catalyst to ensure that the catalyst is always in the best working state; The bio-jet fuel parameters include temperature, pressure, flow rate, catalyst activity, and raw material components; S2: Data preprocessing: Perform preprocessing operations on the bio-jet fuel parameters obtained in step S1 to obtain the bio-jet fuel preprocessing text; The specific preprocessing operations are as follows: The median filtering algorithm is used to remove the pulse noise in the data, and then the moving average filtering algorithm is used to smooth the fluctuations of the flow rate data. Finally, the wavelet denoising algorithm is used to further eliminate the high-frequency noise. At the same time, the Min-Max normalization method is used to map the parameters with different dimensions to the [0,1] interval to make them suitable for the input requirements of the subsequent model.

[0011] S3: Construction of LSTM-Transformer Hybrid Model: Construct multi-dimensional vectors based on the bio-jet fuel pretreatment text, and generate time series data based on the multi-dimensional vectors and input them into the LSTM model and the Transformer model to obtain the predicted values of each bio-jet fuel parameter per second within the next 10 minutes; The construction method of the multi-dimensional vector is as follows: Combine the preprocessed feature data in sequence to form a multi-dimensional feature vector. Assume that n features are selected, and each feature obtains a value after preprocessing. Arrange these n values in sequence to form an n-dimensional vector. For example, if 5 features such as temperature, pressure, flow rate, fatty acid methyl ester content in the raw material components, and catalyst activity specific surface area are selected, and the values x1, x2, x3, x4, x5 are obtained after standardization processing, the constructed multi-dimensional vector is [x1, x2, x3, x4, x5]. At the same time, generate 5-minute sliding window time series data at a sampling frequency of 10Hz. For each time step, arrange the multi-dimensional feature vectors of the current moment and the time steps within the past 5 minutes in sequence to form a three-dimensional tensor. For example, assume that the multi-dimensional vector of each time step is n-dimensional, and there are 5×60×10 = 3000 time steps within 5 minutes, then the dimension of the generated time series data is 3000×n. During the movement of the sliding window, update the data within the window each time to ensure that the model is always trained and predicted based on the latest time series information, so as to effectively process the dynamic change data in the bio-jet fuel production process.

[0012] The LSTM model consists of 3 stacked LSTM units. Each LSTM unit contains 128 memory units. The forget gate, input gate, and output gate in the memory unit are all composed of fully connected layers. Their weight matrices are set by random initialization and updated through the backpropagation algorithm during the training process. When processing data, the multi-dimensional feature vectors of each time step are sequentially input into the LSTM unit. The forget gate determines which information in the memory unit of the previous moment needs to be retained according to the current input and the hidden state of the previous moment; the input gate screens the current input and combines the useful information with the state of the memory unit of the previous moment to update the memory unit; the output gate outputs the hidden state of the current moment according to the updated memory unit state. Through this mechanism, the LSTM layer can effectively capture the long-term dependence relationship of the reaction parameters in the time series, such as the cumulative change trend of temperature over time during the bio-jet fuel reaction process; The Transformer model adopts a multi - head attention mechanism, which is set to 8 heads. Each head independently calculates the attention weights between input feature vectors, then concatenates the results and integrates them through a fully - connected layer. When calculating the attention weights, the input feature vectors are first mapped into three vector spaces of query, key, and value respectively. By calculating the dot - product of the query vector and the key vector and scaling it, attention scores are obtained. After being normalized by the Softmax function, they are multiplied by the value vector and summed to obtain the output of each head. The multi - head attention mechanism can capture the coupling relationships between bio - jet fuel parameters from different perspectives. For example, when analyzing the change in the content of fatty acid methyl ester in the raw material, different heads can respectively focus on its impact on reaction temperature, pressure, and catalyst activity. Then, through the subsequent fully - connected layer, this information is fused to more comprehensively understand the interaction between parameters; It should be further noted that the output of the LSTM layer is used as the input of the Transformer layer. After being processed by the Transformer layer, through a fully - connected layer, the feature vectors are mapped to the same dimension as the number of reaction parameters to be predicted, and finally, the predicted values of each reaction parameter per second within the next 10 minutes are output. For example, as shown in the attached Figure 3 temperature prediction value.

[0013] S4: Catalyst activity decay correction: Calculate the catalyst activity index through the catalyst activity index formula. Compare the catalyst activity index with the preset catalyst activity index threshold. If the value of the catalyst activity index is lower than the preset ratio of the threshold, add the catalyst activity change information as a vector to step S3 to obtain a new predicted value; The catalyst activity index formula is as follows: K = exp(−E a / (R A ·T))A·(P / P0) n , where R A is the gas constant, K refers to the catalyst activity index, T is the current reaction temperature, P is the current pressure, E a refers to the activation energy of the catalyst, A is the pre - exponential factor, and P0 is one standard atmosphere; It should be further noted that the activation energy and pre - exponential factor of the catalyst are determined through laboratory experiments and the accumulation of actual production data, and are not specifically limited in this embodiment; S5: Shadow model correction: Based on the newly collected bio - jet fuel parameters within 1 hour, train the shadow model. After training, compare the predicted value output by the main model with the actual value. When the deviation exceeds the update threshold, update the parameters of the shadow model to the main model; It should be specifically noted that the training method of the shadow model is as follows: The shadow model is trained and updated every hour. The random gradient descent algorithm is used in the training process, and the mean square error is used as the loss function. The calculation formula is , where N is the number of training samples, y i is the actual reaction parameter value, is the model prediction value. As shown in the appendix Figure 4 , during the training process, according to the calculation result of the loss function, the gradients of each parameter are calculated through the backpropagation algorithm, and the weights and bias parameters of the model are updated. To improve the training efficiency, the mini-batch gradient descent method is adopted. Each time, 64 samples are selected to form a mini-batch for parameter update. After the training is completed, the parameters of the shadow model are compared with the parameters of the main model. When the deviation between the actual data and the prediction value of the main model exceeds the threshold of ±2%, the parameters of the shadow model are updated to the main model to ensure that the main model can adapt to the changes in the production process in a timely manner.

[0014] S6: Prediction value analysis and judgment: Calculate the deviation between the prediction value obtained in step S4 and the preset optimal parameter interval and compare it with the safety threshold. According to the comparison result, judge the status of the prediction value; It should be specifically noted that the optimal parameter interval is obtained through statistical analysis of historical data, and the safety threshold is set according to the equipment design parameters and industry safety standards; The deviation calculation refers to calculating the deviation between the prediction value at each time step and the boundary value of the optimal parameter interval in real time. Taking temperature as an example, calculate the temperature prediction value T pred and the lower limit and the upper limit of the optimal temperature interval, that is, Δ = - , Δ = - . Similarly, calculate the deviations of pressure and flow; The pre-judgment of the prediction value status refers to judging the status of the prediction value of each reaction parameter according to the deviation and the safety threshold. If ≤ ≤ , then the temperature prediction value is within the optimal interval; if < or > and ≤ ≤ , then the temperature prediction value exceeds the optimal interval but does not reach the safety threshold; if < or > , then the temperature prediction value exceeds the safety threshold. And so on, determine the status of the pressure and flow prediction values.

[0015] S7: Control strategy generation: Generate a control strategy based on the comparison result in step S6; Specifically, the control strategy includes: a. The predicted value is within the optimal range: When the predicted values of all reaction parameters are within the optimal parameter range, a fixed fine-tuning rule is adopted to maintain the stability of the reaction conditions. The specific rules are as follows: If the predicted temperature value increases by 0.5 °C, the cooling power increases by 3%; if it decreases by 0.5 °C, the heating power increases by 3%. When the predicted pressure value decreases by 0.1 MPa, the feed flow rate increases by 2%; when it increases by 0.1 MPa, the discharge flow rate increases by 2%. When the predicted raw material feed flow rate decreases by 5 L / min, the rotational speed of the feed pump increases by 3%; when it increases by 5 L / min, the rotational speed of the feed pump decreases by 3%. Through these fine-tuning operations, the reaction conditions are maintained in an ideal state to ensure the smooth progress of the production process.

[0016] b. The predicted value exceeds the optimal range but does not reach the safety threshold: When there is a predicted value of a reaction parameter that exceeds the optimal parameter range but does not reach the safety threshold, a preset enhanced control strategy is activated. The preset enhanced control strategy is as follows: When the predicted temperature value is higher than the optimal range, directly increase the power of the cooling equipment by 20%, and at the same time reduce the feed flow rate by 15% to slow down the reaction rate and reduce heat generation, so that the temperature can quickly return to the optimal range. If the predicted temperature value is lower than the optimal range, increase the power of the heating equipment by 20% and increase the feed flow rate by 10% to accelerate the reaction rate and increase the temperature. When the predicted pressure value is lower than the optimal range, increase the feed flow rate by 10%, and at the same time start the auxiliary pressure boosting equipment to increase the pressure to the optimal range. When the predicted pressure value is higher than the optimal range, increase the discharge flow rate by 15% to release part of the pressure and make the pressure return to normal. During the implementation of the enhanced control strategy, continuously monitor the change of the predicted value and adjust the control intensity in real time according to the actual situation to ensure the stability of the reaction parameters.

[0017] c. The predicted value exceeds the safety threshold: When the predicted value of the reaction parameter exceeds the safety threshold, an emergency control strategy is immediately triggered. The emergency control strategy is as follows: Send an alarm to the operator through an audible and visual alarm device, and at the same time prominently display the abnormal parameter and its predicted trend on the monitoring interface. Automatically cut off the heating source, close 50% of the feed valves, stop the main reaction process to prevent the further expansion of danger, start the emergency cooling system, adjust the cooling power to the maximum, and quickly reduce the reaction temperature; open the pressure relief device to reduce the pressure below the safety threshold. After the safety risk is effectively controlled, combined with the prediction result of the LSTM-Transformer model, adjust the reaction parameters step by step at a fixed rate, increase the feed flow rate by 5% and reduce the cooling power by 10% every 10 minutes until the reaction parameters return to the normal production state, and re-evaluate the production conditions to ensure the safe progress of subsequent production.

[0018] As attached Figure 2 A reaction condition optimization control system for bio-jet fuel production, comprising: Sensor module: Composed of temperature, pressure, flow sensors, component analyzers, and catalyst activity detectors, responsible for real-time collection of bio-jet fuel parameters during the bio-jet fuel production process; Data preprocessing module: Includes median filtering, moving average filtering, wavelet denoising, and Min-Max normalization units, used to denoise and normalize the data collected by the sensors to obtain preprocessed bio-jet fuel text; Model construction module: Includes a multi-dimensional vector construction unit, an LSTM model unit, and a Transformer model unit. The multi-dimensional vector construction unit combines the preprocessed data into multi-dimensional vectors and generates time series data. The LSTM model unit and the Transformer model unit work together to obtain the predicted values of each bio-jet fuel parameter per second within the next 10 minutes; Catalyst activity correction module: Used to calculate the catalyst activity index, compare the catalyst activity index with a preset threshold. If it is lower than the preset threshold ratio, feedback the catalyst activity change information to the model to obtain new predicted values; Shadow model update module: Includes a shadow model training unit and a comparison update unit. The shadow model training unit trains the shadow model based on the newly collected 1-hour data using the stochastic gradient descent algorithm. The comparison update unit compares the predicted values of the main model with the actual values, and when the deviation exceeds the threshold, updates the parameters of the main model; Predicted value analysis module: Includes a deviation calculation unit and a comparison judgment unit. The deviation calculation unit is used to calculate the deviation between the predicted value and the boundary value of the optimal parameter interval, and the comparison judgment unit combines the safety threshold to judge the state of the predicted value; Control strategy execution module: Includes a fine-tuning control unit, a reinforcement control unit, and an emergency control unit, used to generate control strategies according to the results of the predicted value analysis module; The fine-tuning control unit maintains the stability of the reaction conditions when the predicted value is within the optimal interval; the reinforcement control unit adjusts the reaction conditions when the predicted value exceeds the optimal interval but does not reach the safety threshold; the emergency control unit performs emergency operations when the predicted value exceeds the safety threshold to ensure production safety.

[0019] Secondly: In the attached drawings of the disclosed embodiments of the present invention, only the structures related to the disclosed embodiments are involved. Other structures can refer to the usual designs. Without conflict, the same embodiment and different embodiments of the present invention can be combined with each other; Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for optimizing and controlling reaction conditions for the production of bio-jet fuel, characterized in that, Including: S1: Bio-jet fuel data acquisition: Obtain bio-jet fuel parameters in real time through the sensor module; The sensor module includes a temperature sensor, a pressure sensor, a flow sensor, a component analyzer, and a catalyst activity detector; The bio-jet fuel parameters include temperature, pressure, flow rate, catalyst activity, and raw material components; S2: Data preprocessing: Perform preprocessing operations on the bio-jet fuel parameters obtained in step S1 to obtain a bio-jet fuel preprocessing text; S3: LSTM-Transformer hybrid model construction: Construct a multi-dimensional vector based on the bio-jet fuel preprocessing text, and generate time series data based on the multi-dimensional vector and input it into the LSTM model and the Transformer model to obtain the predicted value per second of each bio-jet fuel parameter within the next 10 minutes; S4: Catalyst activity decay correction: Calculate the catalyst activity index through the catalyst activity index formula, compare the catalyst activity index with a preset catalyst activity index threshold. If the catalyst activity index value is lower than the preset ratio of the threshold, add the catalyst activity change information as a vector to step S3 to obtain a new predicted value; S5: Shadow model correction: Based on the newly collected bio-jet fuel parameters within 1 hour, train the shadow model. After the training is completed, compare the predicted value output by the main model with the actual value. When the deviation exceeds the update threshold, update the parameters of the shadow model to the main model; S6: Prediction value analysis and judgment: Calculate the deviation between the predicted value obtained in step S4 and the preset optimal parameter interval and compare it with the safety threshold. According to the comparison result, judge the status of the predicted value; S7: Control strategy generation: Generate a control strategy based on the comparison result in step S6.

2. The reaction condition optimization control method for bio-jet fuel production according to claim 1, characterized in that: The specific preprocessing operation is as follows: Use the median filtering algorithm to remove the impulse noise in the data, then use the moving average filtering algorithm to smooth the fluctuation of the flow data, and finally further eliminate the high-frequency noise through the wavelet denoising algorithm. At the same time, use the Min-Max normalization method to map the parameters with different dimensions to the [0,1] interval.

3. The reaction condition optimization control method for bio-jet fuel production according to claim 1, characterized in that: The construction method of the multi-dimensional vector is as follows: Combine the preprocessed feature data in sequence to form a multi-dimensional feature vector. Assume that n features are selected, and each feature obtains a value after preprocessing. Arrange these n values in sequence to form an n-dimensional vector. At the same time, generate 5-minute sliding window time series data at a sampling frequency of 10Hz. For each time step, arrange the multi-dimensional feature vectors of the current time step and the time steps within the past 5 minutes in sequence to form a three-dimensional tensor.

4. A method for optimizing the control of reaction conditions for the production of bio-jet fuel according to claim 1, characterized in that: The LSTM model consists of 3 stacked LSTM units. Each LSTM unit contains 128 memory cells. The forget gate, input gate, and output gate in the memory cells are all composed of fully connected layers. Their weight matrices are set by random initialization and updated through the backpropagation algorithm during the training process. When processing data, the multi-dimensional feature vectors at each time step are sequentially input into the LSTM unit. The forget gate determines which information in the memory cell at the previous moment needs to be retained based on the current input and the hidden state at the previous moment; the input gate filters the current input and combines the useful information with the state of the memory cell at the previous moment to update the memory cell; The output gate outputs the hidden state at the current moment according to the updated memory cell state.

5. The reaction condition optimization control method for bio-jet fuel production according to claim 1, wherein: The Transformer model adopts a multi-head attention mechanism, set to 8 heads. Each head independently calculates the attention weights between the input feature vectors, and then the results are concatenated and integrated through a fully connected layer. When calculating the attention weights, first, the input feature vectors are respectively mapped into three vector spaces of query, key, and value. By calculating the dot product of the query vector and the key vector and scaling it, the attention scores are obtained. After being normalized by the Softmax function, they are multiplied by the value vector and summed to obtain the output of each head.

6. The optimized control method for reaction conditions in the production of bio-jet fuel according to claim 1, characterized in that: The formula for the catalyst activity index is as follows: K = exp(−E a / (R A ·T))A·(P / P0) n , where R A is the gas constant, K is the catalyst activity index, T is the current reaction temperature, P is the current pressure, E a is the activation energy of the catalyst, A is the pre-exponential factor, and P0 is one standard atmosphere.

7. A method for optimizing and controlling reaction conditions for the production of bio-jet fuel according to claim 1, characterized in that: The training method of the shadow model is as follows: The shadow model is trained and updated every hour. The random gradient descent algorithm is used in the training process, and the mean square error is used as the loss function. The calculation formula is , where N is the number of training samples, and y i is the actual reaction parameter value, is the model prediction value. During the training process, according to the calculation result of the loss function, the gradients of each parameter are calculated through the backpropagation algorithm, and the weights and bias parameters of the model are updated. The mini-batch gradient descent method is adopted, and 64 samples are selected each time to form a mini-batch for parameter update. After the training is completed, the parameters of the shadow model are compared with those of the main model. When the deviation between the actual data and the prediction value of the main model exceeds the threshold of ±2%, the parameters of the shadow model are updated to the main model.

8. The reaction condition optimization control method for bio - jet fuel production according to claim 1, characterized in that: The control strategy includes: a. The predicted values are within the optimal range: When the predicted values of each reaction parameter are all within the optimal parameter range, a fixed fine-tuning rule is adopted to maintain the stability of the reaction conditions. The specific rules are as follows: When the predicted temperature value increases by 0.5 °C, the cooling power increases by 3%; when it decreases by 0.5 °C, the heating power increases by 3%. When the predicted pressure value decreases by 0.1 MPa, the feed flow rate increases by 2%; when it increases by 0.1 MPa, the discharge flow rate increases by 2%. When the predicted value of the raw material feed flow rate decreases by 5 L / min, the rotation speed of the feed pump increases by 3%; when it increases by 5 L / min, the rotation speed of the feed pump decreases by 3%; b. The predicted values exceed the optimal range but do not reach the safety threshold: When there are predicted values of reaction parameters that exceed the optimal parameter range but do not reach the safety threshold, a preset enhanced control strategy is activated. The preset enhanced control strategy is as follows: If the predicted temperature value is higher than the optimal range, directly increase the power of the cooling equipment by 20% and at the same time reduce the feed flow rate by 15%. When the predicted temperature value is lower than the optimal range, increase the power of the heating equipment by 20% and increase the feed flow rate by 10%. When the predicted pressure value is lower than the optimal range, increase the feed flow rate by 10% and at the same time start the auxiliary pressurization equipment to increase the pressure to the optimal range. When the predicted pressure value is higher than the optimal range, increase the discharge flow rate by 15%; c. Predicted value exceeds the safety threshold: When the predicted value of the reaction parameter exceeds the safety threshold, an emergency control strategy is immediately triggered. The emergency control strategy is as follows: An alarm is sent to the operator through an audible and visual alarm device. At the same time, the abnormal parameter and its predicted trend are prominently displayed on the monitoring interface. The heating source is automatically cut off, 50% of the feed valve is closed, the main reaction process is stopped, the emergency cooling system is started, the cooling power is adjusted to the maximum, the reaction temperature is rapidly reduced, the pressure relief device is opened, and the pressure is reduced below the safety threshold. After the safety risk is effectively controlled, in combination with the prediction results of the LSTM-Transformer model, the reaction parameters are gradually adjusted at a fixed rate. The feed flow rate is increased by 5% every 10 minutes, and the cooling power is reduced by 10% until the reaction parameters return to the normal production state, and the production conditions are re-evaluated.

9. A reaction condition optimization control system for bio-jet fuel production, which is used to implement the reaction condition optimization control method for bio-jet fuel production according to any one of claims 1-7 above, characterized in that, Including: Sensor module: Composed of temperature, pressure, flow sensors, component analyzers, and catalyst activity detectors, it is responsible for real-time collection of bio-jet fuel parameters during the bio-jet fuel production process; Data preprocessing module: Includes median filtering, moving average filtering, wavelet denoising, and Min-Max normalization units, which are used to denoise and normalize the data collected by the sensors to obtain preprocessed bio-jet fuel text; Model construction module: Includes a multi-dimensional vector construction unit, an LSTM model unit, and a Transformer model unit. The multi-dimensional vector construction unit combines the preprocessed data into multi-dimensional vectors and generates time series data. The LSTM model unit and the Transformer model unit work together to obtain the predicted value of each bio-jet fuel parameter per second within the next 10 minutes; Catalyst activity correction module: Used to calculate the catalyst activity index and compare the catalyst activity index with a preset threshold. If it is lower than the preset ratio of the threshold, the catalyst activity change information is fed back to the model to obtain a new predicted value; Shadow model update module: Includes a shadow model training unit and a comparison update unit. The shadow model training unit trains the shadow model based on the newly collected 1-hour data using the stochastic gradient descent algorithm. The comparison update unit compares the predicted value of the main model with the actual value, and when the deviation exceeds the threshold, the parameters of the main model are updated; Predicted value analysis module: Includes a deviation calculation unit and a comparison judgment unit. The deviation calculation unit is used to calculate the deviation between the predicted value and the boundary value of the optimal parameter interval. The comparison judgment unit combines the safety threshold to judge the state of the predicted value; Control strategy execution module: Includes a fine-tuning control unit, a reinforcement control unit, and an emergency control unit, which are used to generate a control strategy according to the results of the predicted value analysis module.

Images