Multi-parameter fusion based reactor adaptive pressure and temperature coordinated control system

Abstract

The application discloses a kind of reaction kettle adaptive pressure and temperature collaborative control system based on multi-parameter fusion, it is related to reaction kettle control technical field, to solve the problem that existing technology cannot dynamically adapt the complex working condition of high viscosity material;Including signal acquisition module, data fusion processing module, intelligent prediction and decision module and execution module;The signal acquisition module is used to collect the pressure of the kettle top position and the pressure of the kettle bottom position, the temperature of the kettle top position, the temperature of the kettle position, the temperature of the kettle bottom position and the temperature of the radial kettle wall position, the temperature of the center position in the kettle, the signal of stirring rate and the upper material area material viscosity and the lower material area material viscosity;The data fusion processing module is used to pre-process and fusion extraction to the collected signal, to obtain characteristic parameter.The application has the advantages of improving the stability and safety of reaction kettle operation process.

Description

Technical Field

[0001] This invention relates to the field of reactor control technology, and more specifically, to an adaptive pressure and temperature coordinated control system for reactors based on multi-parameter fusion. Background Technology

[0002] As core equipment in industries such as chemical engineering, polymer materials, and pharmaceuticals, reaction vessels are widely used to realize process reactions such as polymerization, esterification, and mixing. However, when handling high-viscosity materials (such as polymer melts, paste adhesives, and high-concentration slurries), two major technical challenges are commonly encountered:

[0003] Firstly, there is the problem of uneven viscosity distribution. High-viscosity materials have poor flowability and high internal friction, making it difficult to achieve uniform mixing throughout the entire area under stirring. Significant spatial viscosity gradients are very likely to occur. The upper low-viscosity region may lead to product degradation due to over-reaction, while the lower high-viscosity region may form by-products due to incomplete reaction, which ultimately seriously affects the uniformity of the product.

[0004] Secondly, there is the problem of coking and sticking to the vessel walls. High-viscosity materials have strong intermolecular forces and are prone to adhering to the vessel walls due to insufficient stirring and shearing or excessively high local temperatures, forming a layer that is difficult to remove. As the reaction time increases, the layer will further carbonize and coke under sustained high temperatures. This not only leads to a sharp increase in the thermal resistance of the vessel wall but also triggers a "heat accumulation" effect: the heat in the area covered by the coking layer cannot be transferred to the material or cooling system in time, resulting in local overheating. Local overheating not only accelerates the degradation of the material and produces toxic and harmful byproducts but may also trigger violent local reactions of the material, causing a sudden increase in pressure inside the vessel and even causing safety accidents such as explosions.

[0005] Existing technologies mostly employ "single-loop temperature control" or "independent dual-loop temperature-pressure control," which adjusts the pressure solely by monitoring the average temperature or overall pressure within the reactor, failing to dynamically adapt to the complex operating conditions of high-viscosity materials. Therefore, we propose an adaptive pressure and temperature co-control system for reactors based on multi-parameter fusion. Summary of the Invention

[0006] The purpose of this invention is to provide an adaptive pressure and temperature co-control system for reactors based on multi-parameter fusion, which aims to solve the problem that existing technologies cannot dynamically adapt to complex working conditions of high-viscosity materials.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an adaptive pressure and temperature coordinated control system for a reactor based on multi-parameter fusion, comprising a signal acquisition module, a data fusion processing module, an intelligent prediction and decision-making module, and an execution module;

[0008] The signal acquisition module is used to acquire in real time the pressure at the top and bottom of the reactor, the temperature at the top, the temperature in the middle, the temperature at the bottom and radial temperatures of the reactor wall, the temperature at the center of the reactor, the stirring rate, and the viscosity of the upper and lower material regions.

[0009] The data fusion processing module is used to preprocess and fuse the collected signals to obtain feature parameters;

[0010] The intelligent prediction and decision-making module outputs the viscosity gradient prediction value, coking risk level and local overheating probability based on the fused feature parameters through a neural network prediction model, and generates collaborative control commands accordingly.

[0011] The execution module is used to respond to coordinated control commands and dynamically adjust the heating and cooling status, pressure, and stirring rate of the reactor.

[0012] Preferably, the signal acquisition module includes multiple pressure sensors, multiple temperature sensors, a speed sensor, and multiple viscosity sensors;

[0013] The pressure sensor is used to collect the pressure at the top and bottom of the reactor, respectively.

[0014] The temperature sensors are used to collect the temperature at the top of the reactor, the middle of the reactor, the bottom of the reactor, the radial temperature at the reactor wall, and the temperature at the center of the reactor.

[0015] The speed sensor is used to collect the stirring rate of the reactor;

[0016] The viscosity sensor is used to collect the viscosity of the material in the upper layer and the viscosity of the material in the lower layer of the reactor, respectively.

[0017] Preferably, when the data fusion processing module preprocesses the acquired signals, it includes using Kalman filtering to remove high-frequency noise from the signals acquired by the signal acquisition module, and then using 3D filtering to remove high-frequency noise from the signals acquired by the signal acquisition module. The criteria are to remove outliers and normalize the signal to the [0,1] interval.

[0018] Preferably, the feature parameters extracted by the data fusion processing module include viscosity gradient, temperature deviation rate, and pressure change rate.

[0019] Preferably, the neural network prediction model in the intelligent prediction and decision-making module is a multi-input multi-output structure, including an input layer, a hidden layer, and an output layer;

[0020] The number of neurons in the input layer is consistent with the dimension of the fused feature parameters. The hidden layer contains 16 neurons and uses the ReLU activation function. The output layer contains 3 neurons, which correspond to the viscosity gradient prediction value, coking risk level and local overheating probability, respectively.

[0021] The coking risk level is 1-5.

[0022] Preferably, the training process of the neural network prediction model is as follows:

[0023] Historical operating data of the reactor was collected as training samples. This historical operating data included pressure, temperature, stirring rate, viscosity signal and corresponding actual viscosity gradient, coking risk and local overheating state under normal operating conditions, viscosity uneven operating conditions, coking operating conditions and local overheating conditions.

[0024] The Adam optimization algorithm is used, with mean squared error as the loss function for model training.

[0025] Preferably, during model training, training is stopped when the neural network prediction model has a prediction error of ≤5% for viscosity gradient, a prediction accuracy of ≥90% for coking risk level, and a prediction error of ≤8% for local overheating probability.

[0026] Preferably, the execution module includes a heating and cooling unit, a pressure relief unit, and a stirring drive unit;

[0027] The heating and cooling unit includes a jacketed partitioned heating structure and a water cooling structure. The jacketed partitioned heating structure includes two independent heating zones, each equipped with an electric heating tube. The water cooling structure is equipped with an electric regulating valve.

[0028] The pressure relief unit uses a proportional pressure relief valve;

[0029] The stirring drive unit uses a variable frequency motor.

[0030] Preferably, the cooperative control instructions include:

[0031] When the viscosity gradient prediction value is greater than 10%, an adjustment command for the stirring drive unit and an adjustment command for the jacket partition heating structure are generated, including increasing the speed of the stirring drive unit by 5%-20%, lowering the temperature of the upper independent heating zone of the jacket partition heating structure by 2-5℃, and raising the temperature of the lower independent heating zone of the jacket partition heating structure by 1-3℃.

[0032] When the coking risk level is ≥3, adjustment commands are generated for the pressure relief unit and the stirring drive unit, including increasing the pressure relief unit's pressure relief amplitude by 0.02-0.05MPa and increasing the stirring drive unit's rotation speed by 10%-15%.

[0033] When the probability of local overheating is greater than 60%, an adjustment command for the water cooling structure and a pressure relief unit are generated, including increasing the flow rate of the water cooling structure by 10%-30% and increasing the pressure relief unit by 0.05-0.1 MPa.

[0034] Preferably, among the feature parameters extracted by the data fusion processing module, the viscosity gradient is calculated using the formula... The calculation is performed using the formula for temperature deviation rate. The calculation is performed using the formula for the rate of change of pressure. Perform calculations;

[0035] In the formula, For viscosity gradient, This refers to the viscosity measurement value of the material in the upper layer of the reactor. This refers to the viscosity measurement value of the material in the lower layer of the reactor. This refers to the temperature deviation rate. For the first Real-time measurement values ​​from a temperature sensor. The preset target reaction temperature, The rate of change of pressure, For the first A pressure sensor at the current moment The measured value, For the first A pressure sensor in the previous moment The measured value, Sampling time interval.

[0036] Compared with the prior art, the beneficial effects of the present invention are:

[0037] 1. This invention utilizes the collaborative work of a signal acquisition module, a data fusion processing module, an intelligent prediction and decision-making module, and an execution module. The signal acquisition module comprehensively and accurately collects pressure, temperature, stirring rate, and viscosity data of materials in different areas within the reactor, achieving high-precision prediction capabilities. It can accurately output viscosity gradient prediction values, coking risk levels, and local overheating probabilities, and generate targeted collaborative control commands accordingly. This enables adaptive collaborative control of reactor pressure and temperature, ensuring the stability and safety of reactor operation and effectively avoiding the impact of adverse conditions such as viscosity inhomogeneity, coking, and local overheating on the reaction process.

[0038] 2. This invention uses Kalman filtering to remove high-frequency noise; 3. The criteria remove outliers and normalize signals, effectively improving data quality. At the same time, by calculating characteristic parameters such as viscosity gradient, temperature deviation rate and pressure change rate, the spatiotemporal fusion of multidimensional data is achieved. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of the system architecture of the present invention;

[0040] Figure 2 This is a schematic diagram illustrating the principle and flow of the present invention. Detailed Implementation

[0041] Example 1

[0042] In this embodiment, the adaptive pressure and temperature collaborative control system for the reactor based on multi-parameter fusion includes a signal acquisition module, a data fusion processing module, an intelligent prediction and decision-making module, and an execution module.

[0043] in:

[0044] The signal acquisition module includes multiple pressure sensors, multiple temperature sensors, a speed sensor, and multiple viscosity sensors;

[0045] Pressure sensors are located at the top and bottom of the reactor, respectively, to collect the pressure at the top and bottom of the reactor.

[0046] Temperature sensors are located at the top, middle, and bottom of the reactor, as well as at the radial wall and center of the reactor, to collect the temperature at the top, middle, and bottom of the reactor, and at the radial wall and center of the reactor, respectively.

[0047] The speed sensor is used to collect the stirring rate of the reactor;

[0048] Viscosity sensors are located in the upper and lower material regions of the reactor, respectively, to collect the viscosity of the material in the upper and lower material regions of the reactor.

[0049] in:

[0050] The data fusion processing module is used to preprocess and fuse the signals acquired by the signal acquisition module to obtain feature parameters and realize the spatiotemporal synchronous fusion of multidimensional data.

[0051] Preprocessing includes using Kalman filtering to remove high-frequency noise from the signals acquired by the signal acquisition module, and passing the signals through 3... The criteria are to remove outliers and normalize the signal to the [0,1] interval;

[0052] Fusion extraction involves calculating characteristic parameters from the preprocessed signal, including viscosity gradient, temperature deviation rate, and pressure change rate. This calculation method includes:

[0053] The viscosity gradient is calculated using the following formula: ;

[0054] In the formula, For viscosity gradient, This is the real-time viscosity measurement value of the material in the upper layer of the reactor. The formula, which measures the real-time viscosity of the lower layer material in the reactor, firstly considers the viscosity difference between the upper and lower layers, as the signal acquisition module deploys viscosity sensors in both the upper and lower material areas. To eliminate the influence of either "upper layer viscosity greater than lower layer" or "lower layer viscosity greater than upper layer" on the assessment of unevenness, the absolute value of the difference is taken. Next, to avoid interference from the absolute viscosity value in assessing the degree of non-uniformity (e.g., with the same difference, the relative non-uniformity of low-viscosity materials is more severe), the arithmetic mean of the viscosities of the upper and lower layers is introduced as a benchmark (the arithmetic mean is easy to calculate and can reflect the overall viscosity level of the material), i.e. Then, the absolute value of the viscosity difference between the upper and lower layers is divided by the arithmetic mean to obtain the relative deviation, which reflects the "proportion of local differences relative to the overall level." Finally, to facilitate subsequent neural network model recognition and manual judgment, the relative deviation is multiplied by... Converted to percentage form, i.e. Finally, the viscosity gradient formula was derived, enabling the quantification of the viscosity non-uniformity between the upper and lower layers of materials in the reactor.

[0055] The temperature deviation rate is calculated using the following formula: ;

[0056] In the formula, This refers to the temperature deviation rate. For the first Real-time measurement values ​​from a temperature sensor. For the preset target reaction temperature, the formula first, for the first... A temperature sensor, its real-time measurement value is... To eliminate the influence of "measured values ​​higher than the target temperature" or "lower than the target temperature" on the judgment of the degree of deviation, the absolute value of the difference between the two is taken, i.e. The absolute deviation is obtained. Then, to reflect the correlation between the degree of deviation and the target temperature, this absolute deviation is divided by the target reaction temperature. This yields the relative deviation; finally, to standardize the judgment criteria, the relative deviation is multiplied by... Converted to percentage form, i.e. Finally, the temperature deviation rate formula was derived, which enabled the quantification of the degree of deviation between a single temperature monitoring point and the target process temperature.

[0057] The pressure change rate is calculated using the following formula: ;

[0058] In the formula, The rate of change of pressure, For the first A pressure sensor at the current moment The measured value, For the first A pressure sensor in the previous moment The measured value, The sampling time interval, the formula first, for the first A pressure sensor determines the pressure value at two key sampling moments: the current moment... Measured values and the previous sampling time Measured values ( (where the sampling time interval is), then the pressure change at these two moments is calculated, i.e. The sign of the change is retained (positive sign indicates pressure increase, negative sign indicates pressure decrease) to reflect the pressure change trend. Finally, according to the physical definition of the rate of change (the change of a physical quantity per unit time), the pressure change is divided by the sampling time interval. ,Right now The real-time rate of pressure change is approximated, and the formula for the rate of pressure change is finally derived, thus realizing the quantification of the rate and trend of pressure change at a single pressure point inside the reactor.

[0059] in:

[0060] The intelligent prediction and decision-making module outputs the viscosity gradient prediction value, coking risk level and local overheating probability based on the feature parameters obtained after fusion through the neural network prediction model, and generates collaborative control commands accordingly.

[0061] The neural network prediction model has a multi-input multi-output structure, including an input layer, a hidden layer, and an output layer;

[0062] The number of neurons in the input layer is consistent with the dimension of the fused feature parameters. The hidden layer contains 16 neurons and uses the ReLU activation function. The output layer contains 3 neurons, which correspond to the viscosity gradient prediction value, the coking risk level (level 1-5), and the probability of local overheating, respectively.

[0063] and:

[0064] The neural network prediction model is trained through the following training process:

[0065] Historical operating data of the reactor was collected as training samples. This historical operating data included pressure, temperature, stirring rate, viscosity signal and corresponding actual viscosity gradient, coking risk and local overheating state under normal operating conditions, viscosity uneven operating conditions, coking operating conditions and local overheating conditions.

[0066] The Adam optimization algorithm was used to train the model with mean squared error as the loss function. Training was stopped when the neural network prediction model had a prediction error of ≤5% for viscosity gradient, a prediction accuracy of ≥90% for coking risk level, and a prediction error of ≤8% for local overheating probability.

[0067] in:

[0068] The execution module is used to respond to coordinated control commands and dynamically adjust the heating and cooling status, pressure, and stirring rate of the reactor.

[0069] The execution module includes a heating and cooling unit, a pressure relief unit, and a stirring drive unit;

[0070] The heating and cooling unit includes a jacketed partitioned heating structure and a water cooling structure. The jacketed partitioned heating structure includes two independent heating zones, each equipped with an electric heating tube. The water cooling structure is equipped with an electric regulating valve.

[0071] The pressure relief unit uses a proportional pressure relief valve;

[0072] The stirring drive unit uses a variable frequency motor;

[0073] Accordingly, the coordinated control instructions include:

[0074] When the viscosity gradient prediction value is greater than 10%, an adjustment command for the stirring drive unit and an adjustment command for the jacket partition heating structure are generated, including increasing the speed of the stirring drive unit by 5%-20%, lowering the temperature of the upper independent heating zone of the jacket partition heating structure by 2-5℃, and raising the temperature of the lower independent heating zone of the jacket partition heating structure by 1-3℃.

[0075] When the coking risk level is ≥3, adjustment commands are generated for the pressure relief unit and the stirring drive unit, including increasing the pressure relief unit's pressure relief amplitude by 0.02-0.05MPa and increasing the stirring drive unit's rotation speed by 10%-15%.

[0076] When the probability of local overheating is greater than 60%, an adjustment command for the water cooling structure and a pressure relief unit are generated, including increasing the flow rate of the water cooling structure by 10%-30% and increasing the pressure relief unit by 0.05-0.1 MPa.

[0077] In this embodiment, the signal acquisition module, data fusion processing module, intelligent prediction and decision-making module, and execution module work together.

[0078] The signal acquisition module comprehensively and accurately collects pressure, temperature, speed and viscosity data of different areas inside the reactor by arranging pressure, temperature, speed and viscosity sensors at multiple locations such as the top, bottom, middle and radial walls of the reactor and the center of the reactor, providing comprehensive raw data support for subsequent control.

[0079] The data fusion processing module uses Kalman filtering to remove high-frequency noise, 3 The criteria remove outliers and normalize signals, effectively improving data quality. At the same time, by calculating characteristic parameters such as viscosity gradient, temperature deviation rate and pressure change rate, the spatiotemporal fusion of multidimensional data is achieved.

[0080] The intelligent prediction and decision-making module is based on a neural network prediction model with a multi-input-multi-output structure. Combined with the high-precision prediction capability after training (viscosity gradient prediction error ≤5%, coking risk level prediction accuracy ≥90%, local overheating probability prediction error ≤8%), it can accurately output the viscosity gradient prediction value, coking risk level and local overheating probability, and generate targeted collaborative control commands accordingly.

[0081] The execution module relies on the jacketed partitioned heating structure, water cooling structure, proportional pressure relief valve, and variable frequency motor to accurately respond to coordinated control commands. When the viscosity gradient prediction value is >10%, it generates adjustment commands for the stirring drive unit and the jacketed partitioned heating structure, namely, increasing the speed of the stirring drive unit by 5%-20%, reducing the temperature of the upper independent heating zone of the jacketed partitioned heating structure by 2-5℃, and increasing the temperature of the lower independent heating zone of the jacketed partitioned heating structure by 1-3℃. When the coking risk level is ≥3, it generates adjustment commands for the pressure relief unit and the stirring drive unit, namely, increasing the pressure relief unit by 0.02-0. When the pressure relief range is 0.05 MPa, the rotation speed of the stirring drive unit is increased by 10%-15%, and the local overheating probability is >60%, adjustment commands for the water cooling structure and the pressure relief unit are generated. That is, the flow rate of the water cooling structure is increased by 10%-30%, and the pressure relief range of the pressure relief unit is increased by 0.05-0.1 MPa. Ultimately, adaptive and coordinated control of the pressure and temperature of the reactor is achieved, ensuring the stability and safety of the reactor operation process. At the same time, the uniformity of the material reaction is improved, and the adverse conditions such as viscosity unevenness, coking, and local overheating are effectively avoided from affecting the reaction process.

[0082] The embodiments disclosed in this invention are preferred embodiments, but are not limited thereto. Those skilled in the art can easily understand the spirit of this invention based on the above embodiments and make different extensions and variations, but as long as they do not depart from the spirit of this invention, they are all within the protection scope of this invention.

Claims

1. An adaptive pressure and temperature coordinated control system for a reactor based on multi-parameter fusion, characterized in that, It includes a signal acquisition module, a data fusion and processing module, an intelligent prediction and decision-making module, and an execution module; The signal acquisition module is used to acquire in real time the pressure at the top and bottom of the reactor, the temperature at the top, the temperature in the middle, the temperature at the bottom and radial temperatures of the reactor wall, the temperature at the center of the reactor, the stirring rate, and the viscosity of the upper and lower material regions. The data fusion processing module is used to preprocess and fuse the collected signals to obtain feature parameters; The intelligent prediction and decision-making module outputs the viscosity gradient prediction value, coking risk level and local overheating probability based on the fused feature parameters through a neural network prediction model, and generates collaborative control commands accordingly. The execution module is used to respond to coordinated control commands and dynamically adjust the heating and cooling status, pressure, and stirring rate of the reactor. The signal acquisition module includes multiple pressure sensors, multiple temperature sensors, a speed sensor, and multiple viscosity sensors; The pressure sensor is used to collect the pressure at the top and bottom of the reactor, respectively. The temperature sensors are used to collect the temperature at the top of the reactor, the middle of the reactor, the bottom of the reactor, the radial temperature at the reactor wall, and the temperature at the center of the reactor. The speed sensor is used to collect the stirring rate of the reactor; The viscosity sensor is used to collect the viscosity of the material in the upper layer and the viscosity of the material in the lower layer of the reactor, respectively. When the data fusion processing module preprocesses the acquired signals, it includes using Kalman filtering to remove high-frequency noise from the signals acquired by the signal acquisition module, and then passing the signals through 3D filtering. The criteria are to remove outliers and normalize the signal to the [0,1] interval; The data fusion processing module extracts feature parameters including viscosity gradient, temperature deviation rate, and pressure change rate. The execution module includes a heating and cooling unit, a pressure relief unit, and a stirring drive unit; The heating and cooling unit includes a jacketed partitioned heating structure and a water cooling structure. The jacketed partitioned heating structure includes two independent heating zones, each equipped with an electric heating tube. The water cooling structure is equipped with an electric regulating valve. The pressure relief unit uses a proportional pressure relief valve; The stirring drive unit uses a variable frequency motor; The coordinated control commands include: When the viscosity gradient prediction value is greater than 10%, an adjustment command for the stirring drive unit and an adjustment command for the jacket partition heating structure are generated, including increasing the speed of the stirring drive unit by 5%-20%, lowering the temperature of the upper independent heating zone of the jacket partition heating structure by 2-5℃, and raising the temperature of the lower independent heating zone of the jacket partition heating structure by 1-3℃. When the coking risk level is ≥3, adjustment commands are generated for the pressure relief unit and the stirring drive unit, including increasing the pressure relief unit's pressure relief amplitude by 0.02-0.05MPa and increasing the stirring drive unit's rotation speed by 10%-15%. When the probability of local overheating is greater than 60%, an adjustment command for the water cooling structure and a pressure relief unit are generated, including increasing the flow rate of the water cooling structure by 10%-30% and increasing the pressure relief unit by 0.05-0.1 MPa.

2. The adaptive pressure and temperature coordinated control system for a reactor based on multi-parameter fusion according to claim 1, characterized in that, The neural network prediction model in the intelligent prediction and decision-making module is a multi-input multi-output structure, including an input layer, a hidden layer, and an output layer; The number of neurons in the input layer is consistent with the dimension of the fused feature parameters. The hidden layer contains 16 neurons and uses the ReLU activation function. The output layer contains 3 neurons, which correspond to the viscosity gradient prediction value, coking risk level and local overheating probability, respectively. The coking risk level is 1-5.

3. The adaptive pressure and temperature coordinated control system for a reactor based on multi-parameter fusion according to claim 2, characterized in that, The training process of a neural network prediction model is as follows: Historical operating data of the reactor was collected as training samples. This historical operating data included pressure, temperature, stirring rate, viscosity signal and corresponding actual viscosity gradient, coking risk and local overheating state under normal operating conditions, viscosity uneven operating conditions, coking operating conditions and local overheating conditions. The Adam optimization algorithm is used, with mean squared error as the loss function for model training.

4. The adaptive pressure and temperature coordinated control system for a reactor based on multi-parameter fusion according to claim 3, characterized in that, During model training, training is stopped when the neural network prediction model has a prediction error of ≤5% for viscosity gradient, a prediction accuracy of ≥90% for coking risk level, and a prediction error of ≤8% for local overheating probability.

Images