Single tube reactor intelligent cleaning method based on multiple surfactant complexing

Abstract

The application provides a single-pipe reactor intelligent cleaning method based on multiple surfactant compounding, relates to the technical field of formula, and comprises the following steps: obtaining reactor operation parameters to determine a fouling state, identifying a fouling type based on a deep learning model and outputting an optimal surfactant combination formula, wherein the formula comprises at least two different types of surfactants and concentration ratios thereof. During the cleaning process, a variable frequency driving device is used to control the injection speed of the cleaning liquid in stages, so that the fouling layer is softened, peeled off and flushed. The application can accurately identify the fouling type, provide a customized cleaning scheme, improve the cleaning efficiency and prolong the service life of the equipment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of formula, in particular to a single tube reactor intelligent cleaning method based on multiple surfactant complexing. BACKGROUND

[0002] In industrial production, single tube reactors, as key equipment in chemical and petroleum industries, generally face scaling problems during long-term operation. Scaling on the inner wall of the reactor not only reduces heat transfer efficiency and increases energy consumption, but also leads to a decrease in reaction space, affects product quality, and even causes safety hazards.

[0003] Traditional single tube reactor cleaning methods mainly rely on fixed formula cleaning agents and standardized cleaning processes. These methods have obvious shortcomings: first, the scaling types and characteristics formed under different working conditions are different, and a single cleaning formula cannot adapt to various scaling conditions, resulting in uneven cleaning effects; second, the cleaning process lacks real-time monitoring and dynamic adjustment mechanisms, making it difficult to optimize cleaning parameters according to the actual situation of scaling layer softening and peeling, resulting in prolonged cleaning time or poor results; finally, existing technologies usually use single or simple mixed surfactants, which cannot fully exert the synergistic effect of different types of surfactants, and have limited ability to remove complex scaling. SUMMARY

[0004] The embodiments of the present application provide a single tube reactor intelligent cleaning method based on multiple surfactant complexing, which can solve the problems in the prior art.

[0005] In a first aspect, the embodiments of the present application provide a single tube reactor intelligent cleaning method based on multiple surfactant complexing, comprising:

[0006] Obtaining real-time operating parameters of a single tube reactor, and determining whether the scaling state of the single tube reactor reaches a preset cleaning threshold according to the real-time operating parameters;

[0007] When the scaling state reaches the preset cleaning threshold, identifying the scaling type of the single tube reactor based on a deep learning model, the deep learning model being obtained by training scaling image features, reactor operating parameters, and cleaning effect evaluation data in historical cleaning data, outputting an optimal surfactant combination formula corresponding to the scaling type, the surfactant combination formula including at least two of anionic surfactants, nonionic surfactants, and amphoteric surfactants, and concentration ratios of various surfactants;

[0008] Mixing the surfactants in the optimal surfactant combination formula according to the concentration ratios to obtain a cleaning solution;

[0009] The injection speed of the cleaning liquid is controlled by a variable frequency driving device, wherein, at the initial stage of cleaning, the injection speed of the cleaning liquid is a first preset speed for softening the scale layer; when it is detected that the softening degree of the scale layer reaches a first preset threshold, the injection speed of the cleaning liquid is increased to a second preset speed for peeling the softened scale layer; when it is detected that the peeling degree of the scale layer reaches a second preset threshold, the injection speed of the cleaning liquid is reduced to a third preset speed for flushing the residual scale.

[0010] The method further comprises training a deep learning model:

[0011] The deep learning model comprises a feature extraction layer, a feature fusion layer and a classification output layer, wherein: the feature extraction layer adopts an improved residual network structure for extracting deep features of the scale image; the feature fusion layer adopts an attention mechanism for adaptively allocating weights to image features and operating parameter features; and the classification output layer adopts a softmax classifier to output a probability distribution of the scale type.

[0012] A surfactant formula knowledge base is established based on historical cleaning data, the surfactant formula knowledge base comprises a corresponding relationship between different scale types and optimal surfactant combination formulas, wherein each formula comprises a selected combination of anionic surfactants, nonionic surfactants and amphoteric surfactants and an optimal concentration ratio thereof, and the optimal concentration ratio is obtained by optimizing historical cleaning effect data by a multi-objective optimization algorithm.

[0013] The historical cleaning effect data comprises at least one of anionic surfactant concentration data, nonionic surfactant concentration data, amphoteric surfactant concentration data, corresponding cleaning time data, cleaning cost data and scale removal rate data.

[0014] The optimal concentration ratio obtained by optimizing the historical cleaning effect data by the multi-objective optimization algorithm comprises:

[0015] A fitness function is suggested based on the historical cleaning effect data, and the historical cleaning effect data is coded into a binary chromosome sequence;

[0016] A plurality of chromosomes are randomly generated to form an initial population;

[0017] The fitness values of the chromosomes in the initial population are calculated, and the chromosomes are sorted in descending order of the fitness values;

[0018] The top 5% of the chromosomes with the highest fitness values are selected as high-quality chromosomes and reserved for the next generation;

[0019] The remaining chromosomes are crossed, the crossed chromosomes are mutated, and the mutated chromosomes are locally optimized, and the gene values in the chromosomes are adjusted within a preset adjustment range;

[0020] The optimized chromosomes are combined with the top 5% chromosomes in the fitness value sorting to form a new population;

[0021] The fitness values of the chromosomes in the new population are calculated, and if the difference between the maximum fitness values of the new and old populations is less than a preset difference threshold or the maximum number of iterations is reached, the iteration is terminated; otherwise, the new population is taken as the initial population, and the step of selecting high-quality chromosomes is returned to continue iteration;

[0022] The optimal chromosome obtained after iteration is decoded to obtain the optimal concentration ratio of anionic surfactant, nonionic surfactant and amphoteric surfactant.

[0023] The injection speed of the cleaning liquid is controlled by a variable frequency drive device, which comprises:

[0024] The variable frequency drive device is started to control the rotation speed of the cleaning liquid pump, so that the cleaning liquid is injected into the reactor at a first preset speed, wherein the first preset speed is determined by the following steps:

[0025] Detect the initial hardness value of the scaling layer;

[0026] According to the initial hardness value, the corresponding injection speed parameter is matched from the preset speed-hardness correspondence database;

[0027] The injection speed parameter is set as the first preset speed;

[0028] During the injection of the cleaning liquid at the first preset speed, the softening degree of the scaling layer is monitored in real time, wherein the softening degree is calculated by detecting at least one parameter of the acoustic impedance, conductivity and permeability of the scaling layer.

[0029] When the softening degree of the scaling layer is detected to reach a first preset threshold, the injection speed of the cleaning liquid is increased to a second preset speed for peeling off the softened scaling layer, which comprises:

[0030] When the softening degree of the scaling layer is detected to reach a first preset threshold, the output frequency of the variable frequency drive device is adjusted to a target frequency value, so that the rotation speed of the cleaning liquid pump is increased, thereby increasing the injection speed of the cleaning liquid to a second preset speed, wherein the second preset speed is 1.5-3 times the first preset speed;

[0031] During the injection of the cleaning liquid at the second preset speed, the weight of the peeled scaling layer is monitored in real time, and when the weight reaches a preset value or the thickness of the scaling layer is detected to be less than a preset thickness, the injection speed of the cleaning liquid is reduced to enter the post-cleaning treatment stage.

[0032] generating a data record of the cleaning process, the data record comprising scale layer softening process data, scale layer peeling process data, and cleaning liquid injection speed variation data, and storing the data record in a cleaning process database.

[0033] In a second aspect of the embodiments of the present application, a single-tube reactor intelligent cleaning system based on a plurality of surfactant combinations is provided, comprising:

[0034] A first unit is configured to acquire real-time operation parameters of the single-tube reactor, and determine whether a scale state of the single-tube reactor reaches a preset cleaning threshold according to the real-time operation parameters.

[0035] A second unit is configured to identify a scale type of the single-tube reactor based on a deep learning model when the scale state reaches the preset cleaning threshold, the deep learning model being obtained by training scale image features, reactor operation parameters, and cleaning effect evaluation data in historical cleaning data, and output an optimal surfactant combination formula corresponding to the scale type, the surfactant combination formula comprising at least two of anionic surfactants, nonionic surfactants, and amphoteric surfactants, and concentration ratios of various surfactants.

[0036] A third unit is configured to mix the surfactants in the optimal surfactant combination formula according to the concentration ratios to obtain a cleaning liquid.

[0037] A fourth unit is configured to control an injection speed of the cleaning liquid by using a variable frequency driving device, wherein the injection speed of the cleaning liquid is a first preset speed in an initial cleaning stage, for softening a scale layer; when a softening degree of the scale layer reaches a first preset threshold, the injection speed of the cleaning liquid is increased to a second preset speed, for peeling the softened scale layer; and when a peeling degree of the scale layer reaches a second preset threshold, the injection speed of the cleaning liquid is reduced to a third preset speed, for flushing residual scale.

[0038] In a third aspect of the embodiments of the present application, an electronic device is provided, comprising:

[0039] a processor;

[0040] a memory for storing processor-executable instructions;

[0041] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0042] In a fourth aspect of the embodiments of the present application, a computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0043] The beneficial effects of the present application are as follows:

[0044] The present application realizes accurate identification and efficient cleaning of single-tube reactor fouling through the intelligent cleaning method of surfactant compounding based on deep learning, significantly improving the cleaning efficiency and effect.

[0045] The compounding scheme of multiple surfactants synergistic effect can exert their respective advantages for different properties of fouling substances, forming a powerful cleaning synergistic effect, effectively solving the problem that traditional single surfactant cannot cope with complex fouling.

[0046] The accurate control of the cleaning liquid injection speed is realized through the frequency conversion driving device, the three-stage cleaning strategy of "softening-separation-flushing" is adopted, the equipment wear is reduced, the service life of the reactor is prolonged, the cleaning agent consumption and cleaning time are reduced, and the dual improvement of environmental protection and economic benefit is realized. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 The figure is a flowchart of the intelligent cleaning method of single-tube reactor based on multiple surfactant compounding of the present application embodiment;

[0048] Figure 2 The figure is a logic flowchart of the multi-objective optimization algorithm. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical scheme and advantages of the present application embodiment clearer, the technical scheme in the present application embodiment will be described clearly and completely in combination with the drawings in the present application embodiment. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the scope of protection of the present application.

[0050] The technical scheme of the present application will be described in detail in specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in some embodiments.

[0051] Reference Figure 1 and Figure 2 The intelligent cleaning method of single-tube reactor based on multiple surfactant compounding of the present application embodiment comprises:

[0052] Obtain the real-time running parameters of the single-tube reactor, and determine whether the fouling state of the single-tube reactor reaches a preset cleaning threshold according to the real-time running parameters;

[0053] When the fouling state reaches the preset cleaning threshold, a deep learning model is used to identify the fouling type of the single-tube reactor, the deep learning model is trained by the fouling image features, reactor operating parameters, and cleaning effect evaluation data in historical cleaning data, and outputs an optimal surfactant combination formula corresponding to the fouling type, the surfactant combination formula includes at least two of anionic surfactant, nonionic surfactant, and amphoteric surfactant, and the concentration ratio of various surfactants;

[0054] The surfactants in the optimal surfactant combination formula are mixed according to the concentration ratio to obtain a cleaning solution.

[0055] A frequency conversion driving device is used to control the injection speed of the cleaning solution, wherein, in the initial stage of cleaning, the injection speed of the cleaning solution is a first preset speed, which is used to soften the fouling layer; when it is detected that the softening degree of the fouling layer reaches a first preset threshold, the injection speed of the cleaning solution is increased to a second preset speed, which is used to peel off the softened fouling layer; when it is detected that the peeling degree of the fouling layer reaches a second preset threshold, the injection speed of the cleaning solution is reduced to a third preset speed, which is used to flush the residual fouling.

[0056] In an optional embodiment, the method further comprises training a deep learning model:

[0057] The deep learning model includes a feature extraction layer, a feature fusion layer, and a classification output layer, wherein: the feature extraction layer uses an improved residual network structure to extract deep features of the fouling image; the feature fusion layer uses an attention mechanism to adaptively allocate weights to the image features and operating parameter features; the classification output layer uses a softmax classifier to output the probability distribution of the fouling type;

[0058] A surfactant formula knowledge base is established based on historical cleaning data, the surfactant formula knowledge base contains the correspondence between different fouling types and optimal surfactant combination formulas, wherein each formula includes the selected combination of anionic surfactant, nonionic surfactant, and amphoteric surfactant, and the optimal concentration ratio thereof, and the optimal concentration ratio is obtained by optimizing historical cleaning effect data through a multi-objective optimization algorithm.

[0059] The deep learning model includes a feature extraction layer, a feature fusion layer, and a classification output layer, and the training process is as follows:

[0060] 5000 images of different types of fouling were collected, including calcium carbonate fouling, calcium sulfate fouling, silicon fouling, iron fouling and organic fouling, about 1000 images for each type. The images were collected by an industrial camera with a resolution of 1920x1080 pixels, and were uniformly cropped to 224x224 pixels in the preprocessing stage.

[0061] The feature extraction layer uses an improved residual network structure, which is modified based on the ResNet-50 architecture. The main improvements include: introducing a 1x1 convolution layer in the standard residual block to reduce the parameter amount; replacing part of the 3x3 convolution with separable convolution to improve computational efficiency; adding dilated convolution in the deep network to expand the receptive field, with a dilation rate of 2 and 4. This layer contains 5 residual block groups, each containing 3-6 residual units, with a total parameter amount of about 18 million and an output feature dimension of 2048.

[0062] The feature fusion layer uses an attention mechanism to process image features and operating parameter features. Operating parameters include system temperature, pressure, flow rate, pH value and conductivity, etc. The attention mechanism is realized through the following steps: first, map the operating parameters to the same feature space as the image features through a fully connected layer; then calculate the correlation matrix of the two sets of features; generate attention weights based on the correlation matrix; finally, fuse the weighted features. The key parameters in the attention module include weight matrices W_q, W_k and W_v, with dimensions (2048, 256), (10, 256) and (2048, 2048) respectively.

[0063] The classification output layer uses a softmax classifier to map the fused features to the class number dimension (i.e. 5 classes) through a fully connected layer, and then applies the softmax function to output the probability distribution of the fouling type.

[0064] The model training uses the mini-batch gradient descent method with a batch size of 32, and the initial learning rate is set to 0.001. The cosine annealing strategy is used to dynamically adjust the learning rate during training. The training lasts for 50 cycles, using the cross-entropy loss function and the Adam optimizer. To prevent overfitting, weight decay (L2 regularization) with a coefficient of 0.0001 and Dropout technology with a random dropout rate of 0.3 are implemented. The accuracy on the training dataset reaches 97.5%, and on the validation set reaches 95.2%.

[0065] Based on 3 years of production history data, 2500 cleaning records were collected to establish a surfactant formulation knowledge base. Each record contains the fouling type, the surfactant combination used, the concentration ratio and the cleaning effect score (range 1-10).

[0066] The knowledge base structure includes the following key fields: scale type ID, scale severity (light, medium, heavy), anionic surfactant type and concentration, non-ionic surfactant type and concentration, amphoteric surfactant type and concentration, cleaning time, cleaning temperature, cleaning effect score, energy consumption index, and environmental protection index.

[0067] For each scale type, the optimal surfactant combination formula is determined by a multi-objective optimization algorithm. The optimization objectives include: maximum cleaning effect, minimum cleaning time, minimum energy consumption, and maximum environmental protection. Genetic algorithm is used for optimization, with the following specific parameter settings: population size 100, iteration number 200, crossover probability 0.8, and mutation probability 0.05.

[0068] Taking calcium carbonate scale as an example, the optimal formula combination is: anionic surfactant selected as sodium dodecyl benzene sulfonate (concentration 0.15%), non-ionic surfactant selected as polyoxyethylene ether (concentration 0.08%), and amphoteric surfactant selected as cocamide propyl betaine (concentration 0.03%). Under medium hardness water quality conditions (calcium hardness 150-300 mg / L), cleaning temperature 65°C, and cleaning time 30 minutes, the cleaning efficiency can reach more than 94%.

[0069] For calcium sulfate scale, the optimal formula is: anionic surfactant selected as alpha-olefin sulfonate (concentration 0.18%), non-ionic surfactant selected as alkyl polyglycoside (concentration 0.12%), and amphoteric surfactant selected as cocamide propyl hydroxyl sulfobetaine (concentration 0.05%). Under the condition of adjusting pH value to 8.5-9.0, cleaning temperature 70°C, and cleaning time 45 minutes, the cleaning efficiency can reach more than 91%.

[0070] For silicon scale, the optimal formula is: anionic surfactant selected as alkyl sulfate (concentration 0.20%) and polycarboxylate (concentration 0.15%) complex, non-ionic surfactant selected as polyoxyethylene fatty alcohol ether (concentration 0.10%), and no amphoteric surfactant. Fluoride chelating agent (concentration 0.05%) needs to be added, pH value adjusted to 10.0-10.5, cleaning temperature 80°C, and cleaning time 60 minutes to achieve a cleaning efficiency of more than 87%.

[0071] The knowledge base regular update mechanism includes: triggering the update process after completing 50 new cleaning operations; adjusting existing formula parameters through Bayesian optimization technology; and conducting a comprehensive update once a quarter, including new surfactants.

[0072] In the actual application of the heat exchanger system in a certain petrochemical enterprise, the system first collects the images of the fouling surface and the operating parameters, and identifies the mixed scale mainly composed of calcium carbonate (calcium carbonate content is about 75%, and organic scale content is about 25%) through the trained deep learning model. The fouling severity score is 7.8 (full score 10), which belongs to moderate to heavy fouling.

[0073] The system queries the formula closest to the scale condition from the surfactant formula knowledge base, and fine-tunes according to the scale thickness (measured value 2.8 mm) and operating parameters (temperature 75℃, pressure 0.6 MPa) to give the final optimized formula: anionic surfactant sodium dodecyl benzene sulfonate (0.17%) and sodium dodecyl sulfate (0.05%) are compounded, non-ionic surfactant polyoxyethylene ether (0.09%), amphoteric surfactant cocamidopropyl betaine (0.04%), and citric acid chelating agent (0.12%) are added.

[0074] The cleaning operation parameters are: circulating flow rate 1.2 m / s, cleaning temperature 68℃, pH value adjusted to 9.2, and cleaning time set to 40 minutes. After cleaning, the detection shows that the scale removal rate reaches 96.3%, which is 14.2 percentage points higher than the original cleaning method of the enterprise (removal rate 82.1%), the cleaning time is shortened by 25%, and the cleaning agent consumption is reduced by 18%.

[0075] Through actual application verification, this method not only improves the cleaning efficiency, but also reduces energy consumption and cleaning agent consumption, which has significant economic and environmental benefits.

[0076] In an alternative embodiment, the historical cleaning effect data includes at least one of anionic surfactant concentration data, non-ionic surfactant concentration data, amphoteric surfactant concentration data, and corresponding cleaning time data, cleaning cost data, and scale removal rate data.

[0077] The historical cleaning effect data includes anionic surfactant concentration data, cleaning time data, and scale removal rate data. First, a historical cleaning effect database is established to record the cleaning time and scale removal rate under different anionic surfactant concentrations. For example, when the concentration of anionic surfactant sodium dodecyl benzene sulfonate is 0.5%, the cleaning time is 30 minutes and the scale removal rate is 85.3%; when the concentration is 0.8%, the cleaning time is 25 minutes and the scale removal rate is 92.7%; when the concentration is 1.2%, the cleaning time is 20 minutes and the scale removal rate is 97.6%.

[0078] By analyzing the historical data, the relationship between the concentration of anionic surfactant and the cleaning effect is determined. The data fitting method is used to analyze the relationship between the concentration and the scale removal rate, and the optimal concentration range is obtained as 0.8%-1.2%. Within this range, the scale removal rate can reach more than 90%, and the cleaning time is controlled within 30 minutes, so the cleaning efficiency is relatively high.

[0079] In an alternative embodiment, the optimal concentration ratio is obtained by optimizing the historical cleaning effect data using a multi-objective optimization algorithm, including:

[0080] Based on the historical cleaning effect data, a fitness function is suggested, and the historical cleaning effect data is encoded into a binary chromosome sequence;

[0081] Randomly generate multiple chromosomes to form an initial population;

[0082] Calculate the fitness values of each chromosome in the initial population, and sort them from large to small according to the fitness values;

[0083] Select the top 5% of the chromosomes with the highest fitness values as high-quality chromosomes and reserve them for the next generation;

[0084] Perform crossover operation on the remaining chromosomes, mutation operation on the chromosomes after crossover, and local optimization on the chromosomes after mutation, and adjust the gene values in the chromosomes within a preset adjustment range;

[0085] Merge the optimized chromosomes with the top 5% of the chromosomes with the highest fitness values to form a new population;

[0086] Calculate the fitness values of each chromosome in the new population. If the difference between the maximum fitness values of the new and old populations is less than a preset difference threshold or the maximum number of iterations is reached, terminate the iteration; otherwise, use the new population as the initial population and return to the step of selecting high-quality chromosomes for further iteration;

[0087] Decode the optimal chromosome obtained after the iteration is terminated to obtain the optimal concentration ratio of anionic surfactant, nonionic surfactant, and amphoteric surfactant.

[0088] The cleaning effect score can be evaluated by multiple indicators such as stain removal rate, material damage degree, and cleaning time, and the score range is 0-100 points.

[0089] For example, the historical data can include the following records:

[0090] - Ratio 1: anionic surfactant 0.5%, nonionic surfactant 0.3%, amphoteric surfactant 0.2%, cleaning effect score 85 points;

[0091] - Ratio 2: anionic surfactant 0.4%, nonionic surfactant 0.4%, amphoteric surfactant 0.2%, cleaning effect score 82 points;

[0092] - Ratio 3: anionic surfactant 0.6%, nonionic surfactant 0.2%, amphoteric surfactant 0.2%, cleaning effect score 88 points;

[0093] Based on historical data, a fitness function is established. The fitness function needs to consider both the cleaning effect and the total amount of surfactant. The better the cleaning effect and the less the total amount of surfactant, the higher the fitness value. The fitness function can be represented as a weighted combination of the cleaning effect score and the total amount of surfactant.

[0094] The fitness value is calculated as follows: the cleaning effect score multiplied by 0.7 plus the surfactant saving rate multiplied by 0.3. The surfactant saving rate is calculated as follows: 1 minus the ratio of the total amount of surfactant under the current ratio to the maximum allowed amount.

[0095] The surfactant concentration ratio is encoded as a binary chromosome sequence. The concentration value of each surfactant is represented by an 8-bit binary number, with a value range of 0-2.55% and an accuracy of 0.01%. A total of 24 bits of binary numbers are required for the three types of surfactants.

[0096] For example, an anionic surfactant concentration of 0.5% can be encoded as 00110010, a nonionic surfactant concentration of 0.3% can be encoded as 00011110, and an amphoteric surfactant concentration of 0.2% can be encoded as 00010100. The complete chromosome is: 001100100001111000010100.

[0097] An initial population of 100 chromosomes is randomly generated. Each bit of each chromosome is randomly assigned a value of 0 or 1, while ensuring that the decoded concentration values are within a reasonable range. The reasonable range is: anionic surfactant 0.1%-1.0%, nonionic surfactant 0.1%-0.8%, amphoteric surfactant 0.1%-0.5%, and the total concentration of the three types of surfactants does not exceed 2.0%.

[0098] The fitness values of the chromosomes in the initial population are calculated. First, the chromosomes are decoded into surfactant concentration values, and then the fitness values are calculated according to the fitness function.

[0099] For example, the chromosome 001100100001111000010100 represents, after decoding, anionic surfactant 0.5%, nonionic surfactant 0.3%, amphoteric surfactant 0.2%, and total concentration 1.0%. Assuming the maximum allowable amount is 2.0%, the surfactant saving rate is 0.5. If the cleaning effect score of this ratio is 85, the fitness value is 85 x 0.7 + 50 x 0.3 = 74.5.

[0100] After calculating the fitness values of all chromosomes, sort them from large to small according to the fitness values.

[0101] Select the top 5% of the chromosomes in terms of fitness value as high-quality chromosomes and directly retain them to the next generation. In a population of 100 chromosomes, the top 5 chromosomes are retained.

[0102] Perform crossover operation on the remaining 95% of the chromosomes. Use double-point crossover, randomly select two crossover points, and exchange the gene segments of the two parent chromosomes between the two points. The crossover probability is set to 0.8, i.e., there is an 80% probability of crossover operation.

[0103] For example, parent chromosome 1 is 001100100001111000010100, and parent chromosome 2 is 001010110010100100001010. If the randomly selected crossover points are the 8th and 16th positions, then the child chromosomes after crossover are 001100100010100100010100 and 001010110001111000001010.

[0104] Perform mutation operation on the chromosomes after crossover. For each bit in the chromosome, mutate it with a probability of 0.05, i.e., change 0 to 1 or change 1 to 0.

[0105] For example, child chromosome 1 is 001100100010100100010100. If the 5th bit is mutated, the mutated chromosome is 001110100010100100010100.

[0106] Perform local optimization on the mutated chromosome. Adjust the gene values in the chromosome within a preset adjustment range to make them closer to the local optimal solution. The preset adjustment range is ±10% of the current value.

[0107] For example, if the mutated chromosome decodes to anionic surfactant concentration of 0.55%, the local optimization can adjust the value within the range of 0.495%-0.605%, and select the concentration value that maximizes the fitness value.

[0108] The optimized chromosomes are combined with the top 5% chromosomes in terms of fitness value to form a new population. The fitness values of the chromosomes in the new population are calculated.

[0109] An iteration termination condition is set: if the difference between the maximum fitness values of the new and old populations is less than a preset difference threshold of 0.001 or the maximum number of iterations is reached, the iteration is terminated; otherwise, the new population is taken as the initial population, and the step of selecting high-quality chromosomes is returned to continue iteration.

[0110] After the iteration is terminated, the chromosome with the highest fitness value is obtained, and the optimal concentration ratio of the anionic surfactant, the nonionic surfactant, and the amphoteric surfactant is obtained by decoding the chromosome.

[0111] Actual application case: The cleaning process of an electronic component is optimized, and the optimal concentration ratio obtained by the above method is: anionic surfactant 0.42%, nonionic surfactant 0.35%, amphoteric surfactant 0.18%, and total concentration 0.95%. The cleaning effect score under this ratio is 92 points, which is increased by 7 points compared with the original process, and the surfactant consumption is reduced by 15%, which improves the cleaning effect and reduces the cost.

[0112] The method can also be applied to the optimization of other types of cleaning agent formulations, such as alkaline cleaning agents and acidic cleaning agents, by adjusting the chromosome encoding method and the fitness function. Through this method, the optimal formula that meets the specific cleaning requirements can be quickly found, the cleaning efficiency is improved, and the environmental burden is reduced.

[0113] In an alternative embodiment, the injection speed of the cleaning liquid is controlled by a variable frequency drive device, which includes:

[0114] The variable frequency drive device is started to control the rotation speed of the cleaning liquid pump, so that the cleaning liquid is injected into the reactor at a first preset speed, wherein the first preset speed is determined by the following steps:

[0115] The initial hardness value of the scaling layer is detected.

[0116] According to the initial hardness value, the corresponding injection speed parameter is matched from the preset speed-hardness correspondence database.

[0117] The injection speed parameter is set as the first preset speed.

[0118] During the injection of the cleaning liquid at the first preset speed, the softening degree of the scaling layer is monitored in real time, wherein the softening degree is calculated by detecting at least one parameter of the acoustic impedance, conductivity, and permeability of the scaling layer.

[0119] In this embodiment, the injection speed of cleaning liquid is controlled by the frequency conversion driving device, so as to realize efficient cleaning of the fouling layer in the reactor. The specific implementation process is as follows:

[0120] The output frequency of the cleaning liquid pump is controlled by the frequency conversion driving device. The frequency conversion driving device controls the rotating speed of the cleaning liquid pump by adjusting the output frequency range (usually 0-50 Hz), so as to accurately adjust the injection speed of the cleaning liquid.

[0121] Before starting the cleaning process, the initial hardness value of the fouling layer needs to be detected. The hardness detection of the fouling layer can adopt a portable hardness tester to obtain the hardness value by contact measurement, and the hardness unit can adopt Shore hardness (HS). For different types of fouling, such as calcium carbonate scale, silicate scale, calcium sulfate scale, etc., the initial hardness value usually distributes between 35HS and 85HS.

[0122] The system pre-establishes a speed-hardness corresponding relationship database, which is obtained based on a large amount of experimental data analysis and stored in the data management module of the control system. The typical corresponding relationship stored in the database is as follows:

[0123] - Hardness value 35-45HS: corresponding injection speed 2.5-3.0 liters / minute;

[0124] - Hardness value 46-55HS: corresponding injection speed 2.0-2.5 liters / minute;

[0125] - Hardness value 56-65HS: corresponding injection speed 1.5-2.0 liters / minute;

[0126] - Hardness value 66-75HS: corresponding injection speed 1.0-1.5 liters / minute;

[0127] - Hardness value 76-85HS: corresponding injection speed 0.5-1.0 liters / minute;

[0128] For example, if the initial hardness value of the fouling layer in a reactor is detected as 58HS, the control system will match the corresponding injection speed range of 1.5-2.0 liters / minute from the database. In actual application, the system will take the median value of 1.75 liters / minute in this range as the first preset speed.

[0129] After determining the first preset speed, the frequency conversion driving device starts to work according to the speed parameter. The output frequency of the frequency converter is automatically adjusted according to the preset speed. For example, for the injection speed of 1.75 liters / minute, the output frequency of the frequency converter is about 26.3 Hz, and the rotating speed of the cleaning liquid pump is adjusted to about 1578 rpm accordingly.

[0130] During the injection of the cleaning liquid, the system monitors the softening degree of the fouling layer in real time. The softening degree detection adopts comprehensive evaluation of multiple parameters:

[0131] Acoustic impedance detection uses an ultrasonic probe to transmit 5MHz ultrasonic signals through the scale layer and calculates the acoustic impedance value based on the reflected signals. In the initial state, the acoustic impedance value of hard scale is usually between 12-18MRayl, and as the softening degree increases, the acoustic impedance value will decrease to 3-6MRayl.

[0132] Conductivity detection uses a four-electrode conductivity sensor to measure the change in conductivity of the scale layer's contact surface with the cleaning solution. In the initial state, the conductivity of the contact surface between hard scale and cleaning solution is usually 0.05-0.1mS / cm, and as the softening degree increases, the conductivity value will increase to 0.8-1.2mS / cm.

[0133] Permeability detection uses a micro-pressure difference sensor to measure the pressure difference of the cleaning solution passing through the scale layer, and indirectly calculates the permeability. In the initial state, the permeability of hard scale is usually 0.005-0.01mD (millidarcy), and as the softening degree increases, the permeability will increase to 0.5-1.0mD.

[0134] The system comprehensively evaluates the softening degree of the scale layer based on real-time data of these three parameters. Taking the scale layer with an initial hardness of 58HS as an example, as the cleaning process progresses, its acoustic impedance decreases from 15.2MRayl to 5.8MRayl, its conductivity increases from 0.08mS / cm to 0.95mS / cm, and its permeability increases from 0.008mD to 0.72mD, indicating that the scale layer is gradually softening.

[0135] When the softening degree reaches the preset threshold (such as the acoustic impedance value decreases to less than 40% of the original value, the conductivity increases to more than 8 times the original value, and the permeability increases to more than 70 times the original value), the system determines that the scale layer has been sufficiently softened, and can adjust the injection speed or enter the next cleaning stage.

[0136] In actual application cases, a petrochemical enterprise's reactor inner wall formed a calcium carbonate scale layer with an average hardness of 62HS, with a thickness of about 8.5mm. The cleaning process was controlled using this method, with an initial injection speed of 1.8 liters per minute, a cleaning solution of 5% organic acid solution, and a temperature control of 45±2℃. After the start of cleaning, the system continuously monitored the softening degree parameters: the acoustic impedance gradually decreased from 16.8MRayl, the conductivity gradually increased from 0.07mS / cm, and the permeability gradually increased from 0.006mD. After 42 minutes of cleaning, the acoustic impedance decreased to 6.2MRayl, the conductivity increased to 0.85mS / cm, and the permeability increased to 0.61mD, indicating that the scale layer had been sufficiently softened, and the cleaning efficiency reached the expected target.

[0137] The injection speed of the cleaning liquid is precisely controlled by the frequency conversion driving device, and combined with real-time softening degree monitoring, so that the cleaning effect is ensured while the cleaning liquid consumption is maximally saved, the cleaning cost is reduced, the cleaning efficiency is improved, and the service life of the equipment is prolonged.

[0138] In an alternative embodiment, when the softening degree of the fouling layer is detected to reach a first preset threshold, the injection speed of the cleaning liquid is increased to a second preset speed for peeling off the softened fouling layer, comprising:

[0139] When the softening degree of the fouling layer is detected to reach the first preset threshold, the output frequency of the frequency conversion driving device is adjusted to a target frequency value, so that the rotating speed of the cleaning liquid pump is increased, thereby increasing the injection speed of the cleaning liquid to the second preset speed, wherein the second preset speed is 1.5-3 times of the first preset speed.

[0140] During the injection of the cleaning liquid at the second preset speed, the weight of the peeled off fouling layer is monitored in real time, and when the weight reaches a preset value or the thickness of the fouling layer is detected to be less than a preset thickness, the injection speed of the cleaning liquid is reduced, and the cleaning enters a post-treatment stage.

[0141] A data record of the cleaning process is generated, which includes fouling layer softening process data, fouling layer peeling process data, and cleaning liquid injection speed change data, and the data record is stored in a cleaning process database.

[0142] The system continuously monitors the softening of the fouling layer by a fouling layer softening degree detection device. The device includes at least three probes embedded in the fouling layer, and the change of the resistance value between the probes is used to represent the softening degree of the fouling layer. In the initial stage of cleaning, the system injects the cleaning liquid at a first preset speed, which is usually set to 2.5-3.5 liters / minute. This speed is sufficient to allow the cleaning liquid to penetrate into the fouling layer, but will not cause impact damage to the unsoftened fouling layer. The temperature of the cleaning liquid is maintained in the range of 45-55℃, and the pH value is adjusted to 9.5-10.5 to accelerate the softening process.

[0143] When the fouling layer softening degree detection device detects that the softening degree reaches a first preset threshold, the system enters a peeling stage. The first preset threshold is usually set to a decrease of 60%-70% of the initial resistance value. For example, if the initial resistance value is 5000 ohms, when the resistance value between the probes is reduced to 1500-2000 ohms, it is determined that the softening degree reaches the preset threshold.

[0144] In the stripping stage, the system controller sends instructions to the frequency converter to adjust the output frequency from the initial value to the target value. In practical applications, if the initial output frequency of the frequency converter is 20 Hz, the target value can be set to 40-50 Hz, which makes the rotational speed of the cleaning liquid pump increase from the initial 1200 rpm to 2400-3000 rpm.

[0145] The increase in rotational speed leads to an increase in the injection speed of the cleaning liquid to a second preset speed, usually 1.5-3 times the first preset speed, i.e. 3.75-10.5 L / min. In a specific embodiment, the first preset speed is 3 L / min, and the second preset speed is set to 7.5 L / min, which is 2.5 times the first preset speed. This speed is sufficient to generate sufficient impact force to strip the softened scale layer while avoiding the potential damage to the pipeline caused by excessively high flow rate.

[0146] During the injection of the cleaning liquid at the second preset speed, the system monitors the stripping process in real time through the weight sensor or thickness detection device. The weight sensor is arranged at the outlet of the discharge pipeline to measure the weight of the discharged scale; the thickness detection device monitors the thickness of the remaining scale layer in real time through the ultrasonic thickness measurement principle.

[0147] When the weight of the discharged scale reaches the preset value or the thickness of the remaining scale layer is less than the preset thickness, the system enters the post-cleaning processing stage. The preset weight value is usually set according to 90%-95% of the initial estimated weight of the scale layer. For example, if the initial estimated weight of the scale layer is 5 kg, the preset weight value can be set to 4.5-4.75 kg. The preset thickness is usually set to 0.5-1 mm, and the residual scale at this thickness does not significantly affect the performance of the equipment.

[0148] When entering the post-cleaning processing stage, the system gradually reduces the output frequency of the frequency converter to 25-30 Hz, so that the rotational speed of the cleaning liquid pump is reduced to 1500-1800 rpm, and the injection speed is correspondingly reduced to 3.5-4.5 L / min. This speed is sufficient to flush the residual scale while reducing water consumption.

[0149] During the entire cleaning process, the system continuously records multiple key data. The scale softening process data includes the softening start time, softening duration, scale resistance value change curve during softening, etc. The scale stripping process data includes the stripping start time, stripping duration, stripping weight per minute, total stripping scale weight, etc. The cleaning liquid injection speed change data includes the injection speed value at different stages, the variable speed time point, the cumulative injection amount, etc.

[0150] These data are transmitted in real time to the central control system through the communication module and stored in the cleaning process database in a predetermined format. The database uses a relational structure, treating each cleaning as an independent record containing metadata such as device ID, cleaning date, cleaning duration, cleaning efficiency, and the detailed process data mentioned above.

[0151] In a specific implementation case, when cleaning a heat exchanger, the initial detection of the scale layer thickness is 8.5 mm, and the estimated weight is 6.2 kg. The system first injects cleaning fluid at a speed of 3 liters per minute, with a temperature of 50℃ and a pH value of 10.2. After 28 minutes, the scale layer resistance value decreases from the initial 5500 ohms to 1870 ohms, and the softening degree reaches the preset threshold. The system adjusts the frequency of the variable frequency drive device from 22Hz to 48Hz, and the injection speed of the cleaning fluid is increased to 7.2 liters per minute.

[0152] During the high-speed injection phase, the weight sensor records the peeling weight data every 5 minutes: 2.1 kg is peeled off in the first 5 minutes, 1.8 kg is peeled off in the second 5 minutes, 1.2 kg is peeled off in the third 5 minutes, and 0.7 kg is peeled off in the fourth 5 minutes. The cumulative peeling weight reaches 5.8 kg, which exceeds the preset value of 5.6 kg (90% of the initial estimated weight). At the same time, the thickness detection device shows that the remaining scale layer thickness is 0.8 mm, which is lower than the preset thickness of 1 mm.

[0153] The system then enters the post-cleaning processing phase, reducing the cleaning fluid injection speed to 4 liters per minute for 15 minutes to flush residual scale. The entire cleaning process lasts 68 minutes, with a cleaning efficiency of over 95% and no damage to the pipeline or equipment.

[0154] The complete cleaning process data is stored in the database for subsequent cleaning process optimization and equipment maintenance plan development. By analyzing historical data, the system can gradually optimize various parameter settings, further improving cleaning efficiency and reducing resource consumption.

[0155] The embodiment of the present application is based on a single-tube reactor intelligent cleaning system with multiple surfactant compounds, which includes:

[0156] The first unit is used to obtain real-time running parameters of the single-tube reactor, and to determine whether the scaling state of the single-tube reactor reaches a preset cleaning threshold according to the real-time running parameters;

[0157] The second unit is configured to identify the fouling type of the single-tube reactor based on a deep learning model when the fouling state reaches the preset cleaning threshold, the deep learning model is obtained by training the fouling image features, reactor operation parameters and cleaning effect evaluation data in historical cleaning data, and outputs an optimal surfactant combination formula corresponding to the fouling type, the surfactant combination formula includes at least two of anionic surfactant, non-ionic surfactant and amphoteric surfactant, and concentration ratios of various surfactants;

[0158] The third unit is configured to mix the surfactants in the optimal surfactant combination formula according to the concentration ratios to obtain a cleaning solution.

[0159] The fourth unit is configured to control the injection speed of the cleaning solution by using a variable frequency driving device, wherein the injection speed of the cleaning solution is a first preset speed in the initial cleaning stage, and is used for softening the fouling layer; when it is detected that the softening degree of the fouling layer reaches a first preset threshold, the injection speed of the cleaning solution is increased to a second preset speed, and is used for peeling the softened fouling layer; when it is detected that the peeling degree of the fouling layer reaches a second preset threshold, the injection speed of the cleaning solution is reduced to a third preset speed, and is used for flushing the residual fouling.

[0160] The third aspect of the embodiment of the present application,

[0161] An electronic device is provided, comprising:

[0162] A processor;

[0163] A memory for storing processor-executable instructions;

[0164] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0165] The fourth aspect of the embodiment of the present application,

[0166] A computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0167] The present application can be a method, device, system and / or computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions loaded thereon for performing various aspects of the present application.

[0168] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions recorded in the above embodiments can be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A single tube reactor smart cleaning method based on multiple surfactant complexing, characterized in that, The method comprises the following steps: acquiring real-time operation parameters of a single-tube reactor, and determining whether the fouling state of the single-tube reactor reaches a preset cleaning threshold according to the real-time operation parameters; when the fouling state reaches the preset cleaning threshold, identifying the fouling type of the single-tube reactor based on a deep learning model, the deep learning model being obtained by training the fouling image features, reactor operation parameters and cleaning effect evaluation data in historical cleaning data, and outputting an optimal surfactant combination formula corresponding to the fouling type, the surfactant combination formula comprising at least two of anionic surfactant, nonionic surfactant and amphoteric surfactant, and the concentration ratio of various surfactants; mixing the surfactants in the optimal surfactant combination formula according to the concentration ratio to obtain a cleaning solution; controlling the injection speed of the cleaning solution by using a frequency conversion driving device, wherein the injection speed of the cleaning solution is a first preset speed in the initial cleaning stage for softening the fouling layer, the injection speed of the cleaning solution is increased to a second preset speed when the softening degree of the fouling layer reaches a first preset threshold for peeling off the softened fouling layer, and the injection speed of the cleaning solution is reduced to a third preset speed when the peeling degree of the fouling layer reaches a second preset threshold for flushing the residual fouling; the method further comprises training the deep learning model: the deep learning model comprises a feature extraction layer, a feature fusion layer and a classification output layer, wherein the feature extraction layer adopts an improved residual network structure to extract deep features of the fouling image, the feature fusion layer adopts an attention mechanism to perform adaptive weight allocation on the image features and the operation parameter features, and the classification output layer adopts a softmax classifier to output the probability distribution of the fouling type; establishing a surfactant formula knowledge base based on historical cleaning data, the surfactant formula knowledge base comprising the corresponding relationship between different fouling types and optimal surfactant combination formulas, wherein each formula comprises the selected combination of anionic surfactant, nonionic surfactant and amphoteric surfactant and the optimal concentration ratio thereof, and the optimal concentration ratio is obtained by optimizing the historical cleaning effect data by a multi-objective optimization algorithm.

2. The method of claim 1, wherein, The historical cleaning effect data comprises at least one of anionic surfactant concentration data, nonionic surfactant concentration data, amphoteric surfactant concentration data, cleaning time data, cleaning cost data and fouling removal rate data.

3. The method of claim 2, wherein, The optimal concentration ratio obtained by optimizing the historical cleaning effect data by a multi-objective optimization algorithm comprises: suggesting an adaptive function based on the historical cleaning effect data, and encoding the historical cleaning effect data into a binary chromosome sequence; randomly generating multiple chromosomes to form an initial population; calculating the fitness values of the chromosomes in the initial population, and sorting them from large to small according to the fitness values; selecting the top 5% of the chromosomes with the highest fitness values as high-quality chromosomes to be reserved for the next generation; The remaining chromosomes are crossed, the crossed chromosomes are mutated, the mutated chromosomes are locally optimized, and the gene values in the chromosomes are adjusted within a preset adjustment range; The optimized chromosomes are combined with the top 5% chromosomes in the sorting of the fitness values to form a new population; The fitness values of the chromosomes in the new population are calculated, and if the difference between the maximum fitness values of the new and old populations is less than a preset difference threshold or the maximum number of iterations is reached, the iteration is terminated; otherwise, the new population is taken as the initial population, and the step of selecting high-quality chromosomes is returned to continue iteration; The optimal concentration ratio of the anionic surfactant, the nonionic surfactant and the amphoteric surfactant is obtained by decoding the optimal chromosome obtained after the iteration is terminated.

4. The method of claim 1, wherein, The injection speed of the cleaning liquid is controlled by using a variable frequency driving device, which comprises: Starting the variable frequency driving device to control the rotating speed of the cleaning liquid pump, so that the cleaning liquid is injected into the reactor at a first preset speed, wherein the first preset speed is determined by the following steps: Detecting the initial hardness value of the scaling layer; According to the initial hardness value, the corresponding injection speed parameter is matched from the preset speed-hardness corresponding relationship database; The injection speed parameter is set as the first preset speed; During the injection of the cleaning liquid at the first preset speed, the softening degree of the scaling layer is monitored in real time, wherein the softening degree is calculated by detecting at least one parameter of the acoustic impedance, conductivity and permeability of the scaling layer.

5. The method of claim 4, wherein, When it is detected that the softening degree of the scaling layer reaches the first preset threshold, the injection speed of the cleaning liquid is increased to a second preset speed for peeling the softened scaling layer, which comprises: When it is detected that the softening degree of the scaling layer reaches the first preset threshold, the output frequency of the variable frequency driving device is adjusted to a target frequency value, so that the rotating speed of the cleaning liquid pump is increased, thereby increasing the injection speed of the cleaning liquid to the second preset speed, wherein the second preset speed is 1.5-3 times of the first preset speed; During the injection of the cleaning liquid at the second preset speed, the weight of the peeled scaling layer is monitored in real time, and when the weight reaches a preset value or it is detected that the thickness of the scaling layer is less than a preset thickness, the injection speed of the cleaning liquid is reduced, and the cleaning enters a post-processing stage; A data record of the cleaning process is generated, which includes scaling layer softening process data, scaling layer peeling process data and cleaning liquid injection speed change data, and the data record is stored in a cleaning process database.

6. A single tube reactor intelligent cleaning system based on complexing of multiple surfactants for implementing the method as claimed in any one of claims 1 to 5, characterized in that, It comprises: The first unit is used for acquiring real-time running parameters of the single-tube reactor, and judging whether the scaling state of the single-tube reactor reaches a preset cleaning threshold according to the real-time running parameters; The second unit is used for identifying the scaling type of the single-tube reactor based on a deep learning model when the scaling state reaches the preset cleaning threshold, the deep learning model is obtained by training the scaling image features, reactor running parameters and cleaning effect evaluation data in the historical cleaning data, and outputs an optimal surfactant combination formula corresponding to the scaling type, the surfactant combination formula comprises at least two of anionic surfactant, nonionic surfactant and amphoteric surfactant, and the concentration ratio of various surfactants; a third unit configured to mix the surfactants in the optimal surfactant combination formula according to the concentration ratio to obtain a cleaning solution; a fourth unit configured to control the injection speed of the cleaning solution by using a variable frequency driving device, wherein the injection speed of the cleaning solution is a first preset speed at the initial stage of cleaning for softening the scale layer, the injection speed of the cleaning solution is increased to a second preset speed when the softening degree of the scale layer reaches a first preset threshold for peeling the softened scale layer, and the injection speed of the cleaning solution is decreased to a third preset speed when the peeling degree of the scale layer reaches a second preset threshold for flushing the residual scale.

7. An electronic device, comprising: comprise: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the instructions stored in the memory to perform the method of any one of claims 1 to 5.

8. A computer-readable storage medium having stored thereon computer program instructions, wherein, The computer program instructions, when executed by the processor, implement the method of any one of claims 1 to 5.

Images