Multi-working-condition energy-saving optimization control method for butyl rubber device

Abstract

The invention discloses a butyl rubber device multi-working-condition energy-saving optimization control method, and belongs to the technical field of industrial process control. The method comprises the following steps: dividing a device into three working conditions of primary cooling standby, secondary cooling use and normal hexane flushing according to a production stage, and respectively constructing dynamic sub-models; a binary classification algorithm (BC) is adopted to quickly narrow a parameter range, a genetic algorithm (GA) and a particle swarm optimization (PSO) are combined to carry out hybrid optimization, and the inlet pressure of a compressor and the temperature of a heat exchanger are adjusted in real time through a DCS interface. According to the method, the problem of poor adaptability of a single algorithm under multiple working conditions is solved, staged accurate optimization is realized, the annual energy-saving benefit is improved by 15-20%, and the method has the characteristics of high efficiency and self-adaption.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial process control, and particularly relates to a multi-condition energy-saving optimization control method for a butyl rubber device, which is applicable to multi-stage dynamic energy-saving optimization in butyl rubber production, and particularly aims at the coordinated regulation of parameters of a refrigeration system under different conditions. Background Art

[0002] The production of butyl rubber is characterized by multiple stages and multiple conditions, such as primary cooling standby, secondary cooling operation, and n-hexane flushing. The energy consumption characteristics of each stage are significantly different. Existing optimization methods do not perform differential processing for different conditions, resulting in the following problems. First, the optimal values of the same control parameter are significantly different under different conditions. For example, the optimal value of the blower inlet pressure (PIC-103) is 68 kPa under the primary cooling condition, while it needs to be reduced to 64 kPa under the n-hexane flushing condition. The existing technology uses a unified parameter setting, resulting in a 10% increase in energy consumption under the latter condition. Second, traditional algorithms (such as particle swarm optimization algorithm, genetic algorithm) have poor adaptability in multi-condition scenarios and are prone to falling into local optima. For example, in a certain device under the secondary cooling condition, the particle swarm optimization algorithm did not converge after 200 iterations due to the too large parameter search space, taking 120 hours; while under the n-hexane condition, the genetic algorithm had insufficient optimization accuracy of ±1 °C due to unreasonable coding methods (such as binary coding being unable to accurately represent temperature decimal places), resulting in an energy-saving rate of only 2%. In addition, existing technologies lack a real-time condition recognition mechanism and rely on manual switching of optimization models. For example, a certain factory failed to detect in time that the feed rate of the reactor decreased from 10,500 kg / h to 8,200 kg / h (indicating a condition switch), and continued to use the original optimization parameters, resulting in an increased risk of compressor surge and a 15% increase in annual maintenance costs. Finally, traditional methods do not consider the energy transfer problem between conditions. For example, in a certain case, energy saving was achieved in a single condition by reducing the power of the ethylene compressor, but due to the lack of coordination of the propylene refrigeration system, the cold supply was insufficient, the reaction temperature got out of control, and finally additional steam consumption was required for compensation, resulting in an 8% increase in overall energy consumption. These problems seriously restrict the energy efficiency improvement of butyl rubber devices, and there is an urgent need for an optimization control scheme that models according to conditions, collaborates multiple algorithms, and supports real-time switching to cope with complex and changeable industrial scenarios.

[0003] This patent proposes an innovative solution to the above-mentioned technical bottlenecks. The present invention constructs a dynamic optimization system for different working conditions, establishes sub-models for three working conditions: primary cooling, secondary cooling, and n-hexane flushing respectively, and uses an LSTM neural network (inputting 10 parameters, sampling frequency 1 minute, classification accuracy 98.7%) to identify the production stage in real time and switch models within 5 seconds. A hybrid optimization algorithm is adopted. The binary classification algorithm quickly converges the parameter search range from 60 - 70 kPa to 63 - 64 kPa (10 iterations / 2 hours). The genetic algorithm reduces the compressor power by 130 kW through decimal coding and dynamic mutation mechanism (mutation rate 0.1 → 0.002). The particle swarm algorithm finely tunes it to 67.3 kPa ± 0.1 kPa within the range of ±2 kPa. The optimized parameters are written into the DCS system via the OPC UA protocol, and the compressor pressure is controlled in real time (fluctuation ±0.5 kPa), and the optimization program is restarted when it exceeds the limit continuously for 5 times, reducing the annual failure rate by 40%. The system realizes cold quantity buffer coordination through working condition prediction, and the overall energy saving rate reaches 17.6% Summary of the Invention

[0004] The core of the present invention lies in the coordination of modeling for different working conditions and the hybrid optimization algorithm. The specific solution includes the following innovative points:

[0005] 1. Multi-working condition division and construction of dynamic sub-models. Working condition definition:

[0006] Primary cooling standby working condition, the feed rate of the reactor ≥ 10,500 kg / h, mainly optimizing the power consumption of the ethylene compressor and the blower. The constraint conditions are the complete liquefaction of ethylene and the compressor speed ≤ 8500 rpm.

[0007] Secondary cooling working condition in operation, the feed rate is 8,000 - 10,500 kg / h, focusing on the collaborative optimization of the propylene compressor and the heat exchanger. The objective function is to maximize the total refrigeration capacity.

[0008] n-hexane flushing working condition, the feed rate ≤ 8,000 kg / h, focusing on reducing the energy consumption of the circulating water system, and it is required to meet the heat exchanger outlet temperature ≥ 25 °C.

[0009] Sub-model construction, deleting irrelevant modules (such as the granulation section) for each working condition, and retaining the core equipment (compressor, heat exchanger) and control points. For example, only the refrigeration module and the raw material feeding module are retained in the n-hexane flushing working condition model.

[0010] 2. Design of the hybrid optimization engine. The binary classification algorithm (BC) quickly locates, and iteratively reduces the parameter range through the dichotomy method. For example, the PIC-103 pressure is quickly converged from 60 - 70 kPa to 66 - 68 kPa. The specific steps include:

[0011] Set the upper and lower bounds of the initial parameters: a1, b1 = (60, 70). In the first round, generate candidate values: 65 kPa, 70 kPa, calculate the objective function values, and select the better interval (65 - 70 kPa) to enter the next round until the range converges to ±1 kPa. Adopt decimal coding and an adaptive mutation mechanism to enhance the robustness of the algorithm. The parameter values are directly mapped to decimal numbers. For example, PIC-103 = 66.5 kPa is encoded as 6650. The initial mutation rate is 0.1, which decreases by 0.002 for each generation to avoid excessive perturbation in the later stage. After BC and GA narrow the search space, PSO performs local optimization to improve the result accuracy. Set the inertia weight to 0.8, the learning factors C1 = C2 = 0.5, and the maximum number of iterations to 50.

[0012] 3. Real-time control and adaptive switching, DCS interface integration. The optimized parameters are written into the DCS system in real time through the OPC UA protocol to adjust the setpoint (SP) of the compressor inlet pressure. The working condition recognition module analyzes the DCS data stream based on the LSTM neural network to judge the current working condition in real time and switch the optimization model. The specific implementation includes:

[0013] 10 key parameters such as the reactor feed rate, circulating water temperature, and compressor power; two-layer LSTM (number of hidden units = 64), and the output layer is a Softmax classifier; covering three months of production data, with a classification accuracy rate of 98.7%. Specific implementation method

[0014] The core of the present invention lies in constructing a multi-condition dynamic optimization system to achieve refined energy-saving control of the butyl rubber plant through the collaboration of real-time data perception and intelligent algorithms. During system deployment, first configure a data acquisition module on the DCS side to read 12 key parameters such as the reactor feed rate, compressor speed, and inlet and outlet temperatures of the heat exchanger at a frequency of twice per second. Among them, a Coriolis mass flowmeter (range 0 - 15,000 kg / h, accuracy ±0.35%) is used for feed rate detection, and an A-class platinum resistance sensor (-50°C - 150°C, ±0.15°C) is selected for temperature measurement. These real-time data are input into the working condition recognition module equipped with an Intel Xeon processor for working condition classification through a pre-trained two-layer LSTM neural network. When the feed rate remains stable above 10,500 kg / h for five consecutive minutes and the speed of ethylene compressor B400 breaks through 8,200 rpm, the system automatically determines that it enters a primary cooling standby working condition. At this time, the corresponding optimized sub-model is loaded into the dynamic model library, the core equipment parameters of the ethylene refrigeration section are retained, and irrelevant variables such as the granulation unit are eliminated.

[0015] The optimization algorithm adopts a three - level progressive architecture. The binary classification algorithm first quickly locates key parameters such as the inlet pressure of the compressor. Taking the pressure optimization of PIC - 104 as an example, the algorithm performs bisection iteration within the initial range of 60 - 70 kPa. Two candidate values are generated in each round and the objective function value (compressor power consumption + heat transfer efficiency weight coefficient) is calculated. After ten rounds of calculation, the search range converges to the interval of 63.2 - 64.8 kPa, and the time consumption is shortened by 82% compared with the traditional exhaustive method. Subsequently, the genetic algorithm takes over. The pressure value is accurately represented to 0.1 kPa using decimal coding (e.g., 67.3 kPa is encoded as 6730), and it cooperates with an adaptive mutation mechanism to dynamically adjust the search strategy. The initial mutation rate is set to 0.15. Whenever the improvement amplitude of the optimal solution is less than 0.08% for five consecutive generations, the mutation mechanism is triggered, and the mutation rate is temporarily increased to 0.3 to jump out of the local optimum. In a typical optimization process, this algorithm reduces the power of the ethylene compressor from 5,850 kW to 5,703 kW after 50 generations of iteration, and the energy consumption of the blower is synchronously reduced by 7.8%. Finally, the particle swarm algorithm performs local optimization within the fine range of ±0.5 kPa. After 80 iterations of 50 particles, the optimized solutions of PIC - 104 = 67.28 kPa and TIC - 102 = - 30.76 °C are output, and the overall energy consumption of the system is reduced by another 0.3 percentage points.

[0016] The optimized parameters are transmitted to the DCS control system in real - time through the industrial - level OPC UA protocol. The compressor speed regulation adopts the fuzzy PID algorithm, with the proportional band set to 2.5%, and the integral time is dynamically adjusted according to the load (120 - 180 seconds). The actual operation data shows that the inlet pressure fluctuation is stabilized within ±0.3 kPa. The heat exchanger valve control module is equipped with an intelligent positioner with a resolution of 0.05%, and the temperature control accuracy is improved to ±0.2 °C in cooperation with the feed - forward compensation algorithm. In terms of the safety protection mechanism, the system sets three - level early warnings: when the deviation of key parameters continuously exceeds the set threshold (pressure ±1 kPa, temperature ±1.5 °C) for five minutes, it automatically reverts to the nearest stable parameter set; the surge early warning module analyzes the compressor characteristic curve in real - time. If the operating point approaches the surge boundary (pressure difference > 85% of the design value), the speed limit mode (fluctuation range ±150 rpm) is immediately started and the standby control strategy is switched; the historical database stores the optimization records of the past 30 days, and the similar working condition matching function is enabled when the communication is interrupted to ensure the continuity of control.

[0017] In the actual application of a 120,000-ton annual butyl rubber production unit of a petrochemical enterprise, this solution has demonstrated remarkable benefits. During the first cooling stage, by optimizing the operating parameters of the ethylene compressor, the annual operating hours were reduced by 700 hours, and the unit energy consumption was reduced by 8.2%; in the n-hexane flushing section, the storage tank pressure was dynamically adjusted to the range of 0.95 - 1.08 MPa, the circulating water consumption decreased from 85 cubic meters per hour to 63 cubic meters, and the heat transfer terminal temperature difference was reduced by 2.3 °C; the efficiency of handling abnormal conditions was greatly improved. In the third quarter of 2023, the number of system restarts caused by parameter overrun decreased by 79% year-on-year, and the fault recovery time was compressed from 47 minutes to 9 minutes. The unplanned equipment shutdown rate decreased from 6.3 times per year on average to 1.8 times, and the energy-saving and consumption-reducing effect passed the TUV certification.

[0018] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions claimed by the present invention.

Claims

1. A multi-condition energy-saving optimization control method for a butyl rubber device, characterized in that, It includes the following steps: (1) Divide the device operation into a primary cooling standby condition, a secondary cooling operation condition, and a n-hexane flushing condition according to the production stage; (2) Establish corresponding dynamic model sub-modules for each condition, and set optimization objectives and constraint conditions; (3) Use a binary classification algorithm (BC) to quickly optimize the key parameters. The binary classification algorithm iteratively narrows the parameter range through the dichotomy method; (4) Combine the genetic algorithm (GA) and the particle swarm optimization algorithm (PSO) to globally correct the optimization results; (5) Write the optimized parameters into the DCS system to adjust the compressor inlet pressure and heat exchanger temperature in real time.

2. The method according to claim 1, wherein The implementation steps of the binary classification algorithm in step (3) include: initializing the upper and lower boundaries of the parameters, generating candidate parameter combinations by dividing according to the dichotomy method; calculating the objective function values of each combination, and selecting the parameters corresponding to the minimum value as the benchmark for the next round of iteration; repeating the iteration until the parameter range converges to the set accuracy.

3. The method according to claim 1, characterized in that, The optimization objective of the primary cooling standby condition is to reduce the power of the ethylene compressor (B400) and the blower (B410). The constraint conditions include: the ethylene is completely liquefied after passing through the AE410 heat exchanger; the compressor speed does not exceed the characteristic curve range.

4. The method according to claim 1, wherein The improvements to the genetic algorithm in step (4) include: using decimal coding instead of binary coding to improve the parameter accuracy; introducing an adaptive mutation probability and dynamically adjusting the mutation rate according to the number of iterations.

5. The method according to claim 1, characterized in that The optimized parameters of the n-hexane flushing condition include: the ethylene compressor inlet pressure (PIC-104), the outlet temperature of the propylene heat exchanger (TI_A40005), and the storage tank pressure (PIC_A43102).

6. A multi-condition energy-saving optimization control device for a butyl rubber plant, characterized in that, A working condition identification module for real-time judgment of the current operating condition of the device; a dynamic model library storing dynamic model sub-modules corresponding to different conditions; a hybrid optimization engine integrating a binary classification algorithm, a genetic algorithm, and a particle swarm optimization algorithm; A DCS interface module for real-time transmission of the optimized parameters to the on-site control system.

7. The device according to claim 6, characterized in that, Quickly narrow the parameter search space through the binary classification algorithm; use the genetic algorithm for global exploration to avoid falling into local optima; use the particle swarm optimization algorithm for fine tuning.

8. The device according to claim 6, characterized in that, The sub-modules in the dynamic model library are constructed in the following way: deleting the modules irrelevant to the target section; adding a compressor inlet pressure control instrument and ignoring the automatic assignment table in UniSim Design.

9. The device according to claim 6, characterized in that The device also includes a stability monitoring module that real-time collects the PV value and SP value of the key instruments; if the deviation between the PV value and the SP value continuously exceeds the threshold, trigger the optimization program to run again.

10. The device according to claim 6, characterized in that The annual energy-saving benefit of the device is calculated by the following formula: Energy-saving benefit = Σ (energy-saving value of each condition × annual operating duration) × electricity price, where the electricity price is calculated at 0.73 yuan / kWh.