Waste tire continuous thermal cracking feeding control method and system based on machine learning

Abstract

The invention discloses a waste tire continuous thermal cracking feeding control method and system based on machine learning, and relates to the technical field of plastic waste recovery. The method comprises the following steps: acquiring a historical rear end socket temperature sequence and a historical internal pressure sequence of a cracker in a historical time window, and a rotation angular velocity and feeding rate sequence in a future preset time window; on the basis of the long and short-term memory network, predicting a predicted end socket temperature sequence and a predicted internal pressure sequence in a future window; with improvement of temperature, pressure and feeding rate stability as multiple targets, an adaptive feeding rate sequence is searched and output in a feeding rate adjustment space through digital twinborn simulation and an intelligent optimization algorithm in combination with the prediction sequence; and finally, carrying out feeding control on the cracker according to the sequence in a preset time window. According to the method, prospective optimization control over the cracking working condition is achieved, the hysteresis quality of traditional control is effectively overcome, and the stability, safety and automation level of the continuous thermal cracking process are improved.

Description

A Machine Learning-Based Control Method and System for Continuous Pyrolysis Feeding of Waste Tires Technical Field

[0001] This invention relates to the field of plastic waste recycling technology, specifically to a machine learning-based continuous pyrolysis feeding control method and system for waste tires. Background Technology

[0002] Continuous pyrolysis technology for waste tires is a key approach to realizing the resource utilization of waste rubber. This process decomposes the high-molecular polymers in tires into high-value-added products such as fuel oil, carbon black, and combustible gas by heating in an oxygen-free or oxygen-deficient environment. In this continuous industrial process, the reaction conditions inside the pyrolysis unit, especially the temperature in the rear end cap area and the pressure inside the chamber, directly affect the conversion efficiency of the pyrolysis reaction, the product quality distribution, and the long-term safety and reliability of the equipment.

[0003] Currently, the control of continuous pyrolysis feed in industrial practice largely relies on the operator's experience and judgment, or employs simple feedback control strategies based on fixed threshold triggers. For example, when the temperature or pressure exceeds a certain setpoint, the feed rate is passively adjusted. This approach has significant limitations: First, due to the large thermal inertia and hysteresis of the pyrolysis reaction, simple feedback control is slow to respond and struggles to suppress periodic or sudden fluctuations in parameters, leading to poor process stability. Second, it lacks comprehensive consideration of the coupled effects of multiple variables; feed, temperature, pressure, and pyrolyzer rotation speed are interconnected, and adjusting a single parameter may trigger a chain reaction of oscillations in other parameters. Finally, methods relying on human experience are not only inefficient but also difficult to continuously optimize for different batches and raw material characteristics, hindering the improvement of automation and intelligence in the production process and failing to meet the stringent requirements of modern industry for efficient, safe, and stable production. Summary of the Invention

[0004] This invention addresses the technical problems of existing technologies that rely on human experience and cannot achieve continuous adaptive adjustment, and whose feedback control response is slow due to the thermal inertia and hysteresis of the pyrolysis reaction. It provides a method and system for continuous pyrolysis feeding control of waste tires based on machine learning.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0006] In a first aspect, the present invention provides a method for continuous pyrolysis feed control of waste tires based on machine learning, comprising: monitoring and acquiring historical back end cap temperature sequences and historical internal pressure sequences of the pyrolyzer within a historical time window, and acquiring rotational angular velocity and feed rate sequences set for the pyrolyzer within a future preset time window; based on the historical back end cap temperature sequences, historical internal pressure sequences, rotational angular velocity, and feed rate sequences, predicting and acquiring predicted back end cap temperature sequences and predicted internal pressure sequences of the pyrolyzer within the preset time window; using preset back end cap temperature thresholds and preset internal pressure thresholds of the pyrolyzer as constraints, and with improving the stability of back end cap temperature, internal pressure, and feed rate as multiple optimization objectives, optimizing the pyrolyzer feed rate within the preset time window according to the predicted back end cap temperature sequences and predicted internal pressure sequences, and outputting an adapted feed rate sequence; and controlling the continuous pyrolysis feed of waste tires into the pyrolyzer according to the adapted feed rate sequence within the preset time window.

[0007] Secondly, the present invention provides a waste tire continuous pyrolysis feeding control system based on machine learning, comprising: a monitoring and acquisition module, used to monitor and acquire the historical back end cap temperature sequence and historical internal pressure sequence of the pyrolyzer within a historical time window, and to acquire the rotational angular velocity and feed rate sequence of the pyrolyzer set within a future preset time window; a prediction and acquisition module, used to predict and acquire the predicted back end cap temperature sequence and predicted internal pressure sequence of the pyrolyzer within the preset time window based on the historical back end cap temperature sequence, historical internal pressure sequence, rotational angular velocity, and feed rate sequence; an optimization and output module, used to optimize the pyrolyzer feed rate within the preset time window based on the predicted back end cap temperature sequence and predicted internal pressure sequence, with the preset back end cap temperature threshold and preset internal pressure threshold of the pyrolyzer as constraints, and with improving the stability of the back end cap temperature, internal pressure, and feed rate as multiple optimization objectives, and output an adapted feed rate sequence; and a feeding control module, used to control the waste tire continuous pyrolysis feeding of the pyrolyzer according to the adapted feed rate sequence within the preset time window.

[0008] The beneficial effects of this invention are as follows: Compared with existing technologies, this invention firstly overcomes the response delay problem caused by thermal inertia and hysteresis in traditional feedback control by integrating historical process parameter sequences with future planned operating parameters and using machine learning models to make forward-looking predictions of key operating states of the pyrolysis unit, thus achieving early perception of operating condition fluctuations. Secondly, by using the threshold conditions for safe operation as hard constraints and setting the stability of multiple parameters such as temperature, pressure, and feed rate as optimization objectives, a more comprehensive and refined multi-objective optimization framework is constructed, which can comprehensively coordinate the complex coupling relationships between various variables and avoid chain oscillations caused by the adjustment of a single parameter. Thirdly, by simulating and evaluating a large number of feeding schemes in a virtual pyrolysis reaction simulation space, the optimal feeding strategy with the best overall performance can be quickly screened and optimized, realizing the transformation from passive response to active optimization, reducing the dependence on operator experience, improving the automation level, operational stability, and overall safety margin of the continuous pyrolysis production process, and providing a reliable technical means for the efficient and intelligent management of waste tire resource utilization. Attached Figure Description

[0009] Figure 1 is a flowchart of the waste tire continuous pyrolysis feeding control method based on machine learning provided by the present invention; Figure 2 is a structural diagram of the waste tire continuous pyrolysis feeding control system based on machine learning provided by the present invention.

[0010] In the attached diagram, the components represented by each number are as follows: monitoring and acquisition module 11, prediction and acquisition module 12, optimization output module 13, and feed control module 14. Detailed Implementation

[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0013] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0014] Example 1, as shown in Figure 1, provides a machine learning-based continuous pyrolysis feed control method for waste tires, including: S10: monitoring and acquiring the historical back-end temperature sequence and historical internal pressure sequence of the pyrolyzer within a historical time window, and acquiring the rotational angular velocity and feed rate sequence set for the pyrolyzer within a future preset time window; the pyrolyzer is the core reaction device for the thermochemical conversion of waste tires. In continuous pyrolysis of waste tires, the continuous pyrolysis reaction inside is a dynamic process with significant thermal inertia and nonlinear characteristics. Therefore, in order to achieve precise feedforward and optimized control of the feed rate, it is first necessary to acquire the key historical process parameter sequence that can characterize the current and recent reaction state, using the historical back-end temperature sequence and historical internal pressure sequence to characterize the recent thermodynamic state and reaction intensity of the pyrolyzer. Simultaneously, it is necessary to acquire the pre-set future operating parameters, using the rotational angular velocity and feed rate sequence set within the future preset time window as forward-looking operations.

[0015] Specifically, monitoring and acquiring the historical back end temperature sequence and historical internal pressure sequence of the pyrolyzer within a historical time window includes: continuously monitoring the back end temperature of the pyrolyzer through a K-type thermocouple and continuously monitoring the internal pressure of the pyrolyzer chamber through a pressure transmitter during the operation of the pyrolyzer, and acquiring the historical back end temperature sequence and historical internal pressure sequence of the pyrolyzer within a historical time window. The pyrolyzer is a continuous pyrolyzer used for continuous pyrolysis of waste tires.

[0016] First, during the continuous operation of the pyrolyzer, two core process parameters need to be continuously acquired at high frequency. The rear end cap temperature is measured and recorded in real time using a K-type thermocouple directly installed at the rear end cap of the pyrolyzer. The K-type thermocouple is a contact temperature sensor based on the nickel-chromium / nickel-silicon thermoelectric effect, suitable for measurements in the medium to high temperature range. Specifically, the rear end cap temperature represents the temperature of the critical mechanical seal area at the pyrolyzer's discharge end. As a primary reference indicator for the pyrolysis temperature, it needs to be stable within an optimal range, such as 430℃ ± 10℃. This rear end cap temperature is directly related to equipment operational safety, heat loss, and the stability of the pyrolysis reaction. For example, the traditional control rule is to automatically increase the feed rate to increase heat absorption and thus lower the temperature when the rear end cap temperature is higher than the set upper limit, such as 450℃, and automatically decrease the feed rate to reduce the cooling effect and thus raise the temperature when the rear end cap temperature is lower than the set lower limit, such as 410℃.

[0017] The internal pressure of the pyrolysis chamber is measured and recorded in real time via a pressure transmitter connected to the pyrolysis chamber. A pressure transmitter is an instrument that senses fluid pressure and converts it into a standard electrical signal for remote transmission and display. Specifically, the internal pressure of the chamber represents the rate of gaseous product formation and system tightness during the pyrolysis reaction, and is a key indicator of the reaction intensity and safety risk. Its value needs to be strictly controlled within a safe window, such as -50 Pa to +150 Pa. For example, a traditional control rule might be to reduce the feed rate to slow the reaction intensity when the system pressure is too high.

[0018] Specifically, this continuous data acquisition process covers a specified historical time window, resulting in two chronologically ordered data sequences: the historical back end cap temperature sequence and the historical internal pressure sequence. The historical time window refers to a continuous period of time immediately preceding the current moment, used for retrospective analysis. This window is set according to the dynamic characteristics of the process and the requirements of the prediction model; for example, its length can be 30 minutes or 1 hour. Meanwhile, the pyrolysis unit specifically refers to a continuous pyrolysis reactor used to process waste tire materials.

[0019] In addition, it is necessary to obtain the pre-set operating parameters for future preset time windows, including the rotational angular velocity of the pyrolyzer spindle and the planned feed rate sequence for adding waste tire material into the pyrolyzer. The rotational angular velocity is usually a set value, while the feed rate sequence is time-series data that specifies the planned feed rate value for each future moment or time period. Making the changes in the feed rate sequence as smooth as possible can avoid shocking the equipment.

[0020] Specifically, obtaining the rotational angular velocity and feed rate sequence of the pyrolyzer within a future preset time window can provide known input conditions for future operations for subsequent prediction and optimization. The future preset time window refers to the continuous time period following the current moment that the control strategy will cover and execute. It is set according to the length of the control cycle and the forward-looking requirements of the optimization decision; for example, the length could be 15 minutes or 30 minutes in the future.

[0021] S20: Based on the historical back end cap temperature sequence, historical internal pressure sequence, rotational angular velocity, and feed rate sequence, predict the predicted back end cap temperature sequence and predicted internal pressure sequence of the pyrolyzer within the preset time window; after obtaining the historical back end cap temperature sequence, historical internal pressure sequence, rotational angular velocity, and feed rate sequence, further, the above data needs to be used as input and fed into a pre-trained long short-term memory network prediction model to simulate the complex dynamic reaction process and heat and mass transfer mechanism inside the pyrolyzer, and obtain the predicted back end cap temperature and predicted internal pressure values ​​corresponding to each sampling moment in the future preset time window, thereby generating a complete predicted back end cap temperature sequence and predicted internal pressure sequence, providing an accurate operating condition prediction basis for subsequent feed rate optimization control based on forward-looking information.

[0022] Specifically, based on the historical back end cap temperature sequence, historical internal pressure sequence, rotational angular velocity, and feed rate sequence, the predicted back end cap temperature sequence and predicted internal pressure sequence of the pyrolyzer within the preset time window are predicted. This includes: constructing a back end cap temperature prediction plugin and an internal pressure prediction plugin based on a long short-term memory network; using the back end cap temperature prediction plugin, predicting the predicted back end cap temperature sequence of the pyrolyzer within the preset time window based on the historical back end cap temperature sequence, rotational angular velocity, and feed rate sequence; and using the internal pressure prediction plugin, predicting the predicted internal pressure sequence of the pyrolyzer within the preset time window based on the historical internal pressure sequence, rotational angular velocity, and feed rate sequence.

[0023] First, it is necessary to construct a rear end temperature prediction module and an internal pressure prediction module based on a Long Short-Term Memory (LSTM) network. The LSTM network is a special type of recurrent neural network structure that excels at processing time-series data and capturing its long-term dependencies, making it suitable for modeling and predicting thermal decomposition processes that possess temporal continuity and inertia.

[0024] Specifically, the rear end temperature prediction module is a dedicated component trained on a large amount of historical operating data with a long short-term memory network as its core. It is used to establish a complex nonlinear mapping relationship between the historical rear end temperature sequence, rotational angular velocity, and feed rate sequence and the future rear end temperature sequence. It can accurately simulate the dynamic impact of feed operation and heat transfer effect on the temperature evolution of the pyrolyzer outlet. The internal pressure prediction module is used to establish a mapping relationship between the historical internal pressure sequence, rotational angular velocity, and feed rate sequence and the future internal pressure sequence. It can effectively characterize the dynamic balance process between reaction intensity, gaseous product generation, and system exhaust.

[0025] Specifically, a back end cap temperature prediction plugin is constructed based on a Long Short-Term Memory (LSTM) network. This includes: collecting sample back end cap temperature sequence sets, sample rotational angular velocity sets, and sample feed rate sequence sets based on historical operational monitoring records of similar pyrolyzers; using historical back end cap temperature sequences within historical time windows corresponding to different scenarios for different sample back end cap temperature sequences, sample rotational angular velocities, and sample feed rate sequences as sample predicted back end cap temperature sequences to obtain a sample predicted back end cap temperature sequence set. The historical time window has the same time span as the preset time window. Using the sample back end cap temperature sequence set, sample rotational angular velocity set, and sample feed rate sequence set as input, and using the sample predicted back end cap temperature sequence set as supervision, the LSM network is trained until convergence to generate the back end cap temperature prediction plugin.

[0026] First, based on historical operation monitoring records accumulated during the long-term operation of similar pyrolyzers with similar process conditions to the target pyrolyzer, sample data from three key dimensions were systematically collected: first, a sample set of back-end temperature sequences, containing actual monitored back-end temperature change curves over multiple historical time periods; second, a sample set of rotational angular velocities, recording the setpoint rotational speed of the pyrolyzer spindle within the corresponding time period; and third, a sample set of feed rate sequences, recording the planned and actual feed rate changes of waste tire material within the corresponding time period.

[0027] Simultaneously, each combination consisting of a specific sample back end cap temperature sequence, sample rotational angular velocity, and sample feed rate sequence is defined as an independent process operation scenario. For each operation scenario, the actual monitored back end cap temperature change sequence within a historical time window immediately following that scenario needs to be extracted, and this historical back end cap temperature sequence is set as the target output that the model should predict under that scenario, i.e., the sample predicted back end cap temperature sequence. It should be noted that the duration of the selected historical time window must be strictly consistent with the duration of the future preset time window required for prediction in subsequent practical applications, for example, both set to 30 minutes, to ensure that the model learns the dynamic evolution law of process parameters at the same time scale during training. By performing the above processing procedure on all available historical operation scenarios one by one, a complete set of sample predicted back end cap temperature sequences can be obtained, providing sufficient supervision signals for model training.

[0028] Further, after data preparation, the model training phase begins. The previously obtained set of sample back-end temperature sequences, sample rotational angular velocities, and sample feed rate sequences are used as training input features, and the corresponding sample predicted back-end temperature sequences are used as supervision signals, i.e., training targets. The long short-term memory network model is trained until convergence, resulting in a fully trained back-end temperature prediction plugin. The convergence condition is set based on whether the model's prediction error on the independent validation set no longer significantly decreases or reaches a preset upper limit for the number of iterations. For example, the root mean square error of the validation set decreases by less than 1‰ within 10 consecutive training cycles, or the total number of training iterations reaches 1000.

[0029] For example, since there is a highly nonlinear temporal dependency between the temperature sequence of the pyrolyzer back end cap, the internal pressure sequence and operating parameters such as feed rate and rotational angular velocity, and long short-term memory networks have significant advantages in sequence data modeling and long-term temporal dependency capture, long short-term memory networks are selected to construct the back end cap temperature prediction plug-in and the internal pressure prediction plug-in, respectively.

[0030] Specifically, the network structures of both the rear end cap temperature prediction plugin and the internal pressure prediction plugin mainly consist of an input layer, a temporal feature abstraction layer, and a sequence prediction output layer. The input layer receives a normalized, time-step-organized input sequence, which includes historical rear end cap temperature and internal pressure sequences, a preset rotational angular velocity sequence, and a preset feed rate sequence. The temporal feature abstraction layer employs a multi-layer long short-term memory (LSTM) network structure. The number of network units in each layer is configured according to the dimension and complexity of the input sequence. The network uses a gating mechanism to filter and transmit information, capturing long-term dependencies within the sequence. Each LSTM layer can be followed by a Dropout layer with a dropout rate set between 0.1 and 0.3 to effectively suppress model overfitting and improve its generalization prediction ability for new operating conditions. The sequence prediction output layer uses a fully connected network with a linear activation function to map the final temporal features output by the LSTM network to a continuous predicted value step-by-step, thereby generating a complete predicted rear end cap temperature sequence or predicted internal pressure sequence.

[0031] During training, key hyperparameters included a learning rate of 0.001, 200 training epochs, and a batch size of 32. The learning rate was set to balance the stability and convergence speed of gradient descent. The number of training epochs ensured the model fully learned the dynamic patterns of the pyrolysis process. The batch size balanced the efficiency of time-series data processing with memory resource consumption. Specifically, a supervised learning training method was adopted. Sample back-end temperature sequences and sample internal pressure sequences were collected from historical monitoring records of similar pyrolyzers, combined with corresponding sample rotational angular velocity sequences and sample feed rate sequences as input feature sample sets. Simultaneously, based on historical time windows, corresponding future actual back-end temperature change sequences and internal pressure change sequences were extracted to form sample predicted back-end temperature sequences and sample predicted internal pressure sequences, serving as supervision label sets. The input feature sample set and corresponding label sample set for each plugin were divided into training, validation, and test sets in a 7:2:1 ratio.

[0032] Furthermore, the sample feature sequences in the training set are used as input, and the corresponding future actual temperature or pressure sequences are used as supervision signals. The weight parameters of the Long Short-Term Memory (LSTM) network are iteratively optimized using a time backpropagation algorithm in conjunction with the Adam optimizer. The mean square error loss function is used to measure the overall deviation between the predicted and actual sequences at each time step. The training process is monitored using a validation set. Training is terminated when the decrease in the root mean square error of the validation set is less than 1‰ over 10 consecutive training cycles, resulting in the converged back-end temperature prediction plugin and internal pressure prediction plugin.

[0033] After construction, the pre-trained back-end temperature prediction plugin and internal pressure prediction plugin are used for specific predictions. Specifically, for the prediction of the back-end temperature, the historical back-end temperature sequence, the preset rotational angular velocity, and the preset feed rate sequence are input into the back-end temperature prediction plugin. This plugin calculates and outputs the predicted back-end temperature values ​​of the pyrolyzer at a series of time points within a preset future time window, i.e., the predicted back-end temperature sequence.

[0034] Similarly, for internal pressure prediction, historical internal pressure sequences, the same preset rotational angular velocity, and the preset feed rate sequence are input into the internal pressure prediction plugin. This plugin calculates and outputs predicted internal pressure values ​​for the pyrolyzer at a series of time points within the same preset time window in the future, i.e., the predicted internal pressure sequence.

[0035] The predicted back end cap temperature sequence and the predicted internal pressure sequence together constitute a quantitative prediction of the pyrolyzer's short-term operating status, which can provide a key data foundation for subsequent optimization decisions.

[0036] S30: Using the preset rear head temperature threshold and preset internal pressure threshold of the pyrolyzer as constraints, and with the goal of improving the stability of the rear head temperature, internal pressure, and feed rate as multiple optimization objectives, the feed rate of the pyrolyzer within the preset time window is optimized based on the predicted rear head temperature sequence and predicted internal pressure sequence, and an adapted feed rate sequence is output. Specifically, using the preset rear head temperature threshold and preset internal pressure threshold of the pyrolyzer as constraints, and with the goal of improving the stability of the rear head temperature, internal pressure, and feed rate as multiple optimization objectives, the feed rate of the pyrolyzer within the preset time window is optimized based on the predicted rear head temperature sequence and predicted internal pressure sequence, and an adapted feed rate sequence is output. This includes: obtaining the feed rate adjustment space of the pyrolyzer, and randomly generating several initial feed rate sequences based on the feed rate adjustment space; optimizing the feed rate within the preset time window based on the predicted rear head temperature sequence and predicted internal pressure sequence, and outputting an adapted feed rate sequence. The pyrolysis process is simulated using a pressure sequence and several initial feed rate sequences, outputting several simulated back end cap temperature sequences and several simulated internal pressure sequences. Constrained by satisfying the preset back end cap temperature threshold and the preset internal pressure threshold, the initial feed rate sequences are filtered based on the simulated back end cap temperature sequences and the simulated internal pressure sequences to obtain several qualified feed rate sequences, as well as several simulated back end cap temperature sequences and several simulated internal pressure sequences of the qualified feed rate sequences. With improving the stability of the back end cap temperature, internal pressure, and feed rate as the multiple optimization objectives, the fitness of multiple feed schemes is evaluated and determined based on the multiple simulated back end cap temperature sequences and the multiple simulated internal pressure sequences. Based on the fitness of the multiple feed schemes and the multiple qualified feed rate sequences, the pyrolyzer feed rate is optimized, and an adapted feed rate sequence is output.

[0037] Specifically, multi-objective optimization decision-making is implemented to find the optimal feed rate sequence with the best overall performance from numerous possible feed schemes. First, the search scope needs to be defined, i.e., the allowable feed rate adjustment space of the pyrolyzer under current operating conditions needs to be obtained. This feed rate adjustment space is determined jointly by the equipment's physical limits and process safety requirements. Based on this feed rate adjustment space, a large number of differentiated initial feed rate sequences are generated through random sampling, serving as the starting population for the optimization search.

[0038] Secondly, it is necessary to evaluate the future operating conditions that each initial feed rate sequence may trigger. Specifically, combining the obtained predicted back end cap temperature sequence and predicted internal pressure sequence, each initial feed rate sequence is used as an input variable to simulate the pyrolysis process within a preset time window. Each pyrolysis process simulation will output the corresponding simulated back end cap temperature sequence and simulated internal pressure sequence, thereby predicting the future operating trajectory of the pyrolyzer under this feed scheme.

[0039] Specifically, the pyrolysis process is simulated based on the predicted back end cap temperature sequence, the predicted internal pressure sequence, and several initial feed rate sequences, respectively, and several simulated back end cap temperature sequences and several simulated internal pressure sequences are output. This includes: randomly selecting a first initial feed rate sequence from the several initial feed rate sequences; performing operational simulation of the pyrolyzer based on a digital twin to construct a pyrolysis reaction simulation space; and within the pyrolysis reaction simulation space, performing pyrolysis reaction simulation within the preset time window based on the first initial feed rate sequence, the predicted back end cap temperature sequence, and the predicted internal pressure sequence, outputting a first simulated back end cap temperature sequence and a first simulated internal pressure sequence, and adding them to the several simulated back end cap temperature sequences and several simulated internal pressure sequences.

[0040] Specifically, pyrolysis simulation is a key step in enabling forward-looking virtual testing of candidate feed rate schemes. First, from several randomly generated initial feed rate sequences, one is arbitrarily selected as the scheme to be evaluated and marked as the first initial feed rate sequence.

[0041] The simulation relies on a high-fidelity virtual simulation environment. Therefore, it is necessary to construct a digital mapping model based on digital twin technology that highly corresponds to the physical pyrolyzer in terms of geometry, thermodynamic properties, and reaction kinetics, forming a pyrolyzer simulation space capable of realistically simulating the actual pyrolyzer reaction process. This pyrolyzer simulation space integrates multi-physics coupling equations for heat transfer, fluid flow, material transport, and pyrolyzer chemical reactions, and can dynamically calculate the internal state of the pyrolyzer based on the input operating conditions.

[0042] Specifically, constructing the simulation space for the pyrolysis reaction is a digital modeling process. First, a precise three-dimensional digital model needs to be established at the geometric level to fully reproduce the overall structural dimensions of the physical pyrolysis reactor and the spatial layout of key components such as internal stirring blades, heating coils, and inlet / outlet ports. This forms the physical framework of the virtual space. Second, accurate material property parameters need to be assigned to the geometric model at the physical property level, including the thermal conductivity and specific heat capacity of the reactor walls, as well as key thermophysical parameters such as the density and viscosity of the internal materials. These parameters form the fundamental basis for subsequent numerical calculations in heat transfer and fluid dynamics.

[0043] The most crucial element lies in the embedding and integration of mechanistic models. The essence of the pyrolysis reaction simulation space lies in its integration of a series of mathematical models characterizing the inherent laws of the pyrolysis process. These mainly include heat transfer models for calculating heat conduction, convection, and radiation inside and outside the reactor, especially for the complex heat exchange processes between the heating source and the material, and between the material and the reactor wall; fluid flow models that use computational fluid dynamics to simulate the mixing, flow, and separation behavior of gas-liquid-solid multiphase materials during pyrolysis; material transport models that describe the transport, propulsion, residence time distribution, and discharge of pyrolysis residues within the reactor; and a pyrolysis chemical reaction kinetic model based on theories such as the Arrhenius equation. This pyrolysis chemical reaction kinetic model describes the complex reaction network and reaction rates of each step in the decomposition of tire rubber polymers under specific temperature and pressure conditions, resulting in the formation of oil, gas, and carbon.

[0044] The above models are closely linked through multiphysics coupling equations. For example, the thermal effect of a chemical reaction directly interferes with the temperature field distribution, and changes in the temperature field in turn affect the chemical reaction rate and fluid viscosity characteristics. The fluid flow state simultaneously determines the uniformity of heat transfer and material distribution. The final pyrolysis reaction simulation space is a complex software system capable of solving such coupled partial differential equations or algebraic equations. When different operating conditions, including the first initial feed rate sequence, are input into the pyrolysis reaction simulation space, and the predicted back-end temperature sequence and predicted internal pressure sequence are used as the initial state or boundary perturbation, the pyrolysis reaction simulation space can dynamically deduce the spatiotemporal evolution of state parameters such as temperature, pressure, concentration, and flow rate inside the pyrolyzer in future time periods, and thus output reliable first simulated back-end temperature sequence and first simulated internal pressure sequence.

[0045] Specifically, within this pyrolysis reaction simulation space, a simulation run is initiated for a preset time window. The selected first initial feed rate sequence is used as the core operating variable. At the same time, the predicted back end temperature sequence and the predicted internal pressure sequence obtained above are used as initial boundary conditions or disturbance inputs and injected into the pyrolysis reaction simulation space. Based on the built-in mechanism, the pyrolysis reaction simulation space dynamically calculates the evolution of the internal state of the pyrolyzer over time under this specific feed strategy.

[0046] After the simulation is completed, the temperature change data and internal pressure change data of the pyrolyzer rear end cap region within the preset time window can be extracted to form the simulation results corresponding to the first initial feed rate sequence, namely the first simulated rear end cap temperature sequence and the first simulated internal pressure sequence. The first simulated rear end cap temperature sequence and the first simulated internal pressure sequence are added to several simulated rear end cap temperature sequences and several simulated internal pressure sequences used to summarize all simulation results, respectively, as the basis for subsequent screening and evaluation of the candidate scheme.

[0047] Specifically, the above process will be repeated for each initial feed rate sequence, thereby generating corresponding simulated operating data for all candidate schemes.

[0048] Furthermore, the initial feeding schemes are preliminarily screened based on mandatory constraints for safe operation. Specifically, using preset back-end temperature thresholds and preset internal pressure thresholds as criteria, several simulated back-end temperature sequences and several simulated internal pressure sequences output from the pyrolysis process simulation are checked item by item to determine whether their values ​​at each time point within a preset time window meet the corresponding threshold requirements. Initial feeding rate sequences whose simulated temperature or pressure values ​​at any time point exceed the corresponding safety thresholds are eliminated, retaining only schemes that meet the constraints at all time points within the entire time window, thus forming a set consisting of multiple qualified feeding rate sequences. Simultaneously, the simulated back-end temperature sequence and simulated internal pressure sequence corresponding to each qualified feeding rate sequence are recorded for subsequent refined evaluation.

[0049] Among them, the preset rear end temperature threshold and the preset internal pressure threshold are the limit allowable values ​​to ensure the mechanical structure safety and reaction stability of the pyrolyzer, respectively, representing the safety boundary of process operation. They are set comprehensively according to equipment design specifications, material tolerance limits and safety production procedures. For example, the rear end temperature threshold can be set to 450℃ and the internal pressure threshold can be set to 150Pa.

[0050] Furthermore, within the set of qualified solutions that have passed the safety screening, a more refined selection process is conducted. With the goal of improving the stability of the back end cap temperature, internal pressure, and feed rate, the fitness of multiple feeding schemes is evaluated and determined based on multiple simulated back end cap temperature sequences and multiple simulated internal pressure sequences. A higher fitness value indicates that the scheme, while meeting safety constraints, can achieve smaller fluctuations in process parameters and more stable operation.

[0051] Specifically, with the goal of improving the stability of the back end cap temperature, internal pressure, and feed rate, multiple feed scheme fitness levels are evaluated and determined based on multiple simulated back end cap temperature sequences and multiple simulated internal pressure sequences. This includes: randomly selecting a first qualified feed rate sequence from the multiple qualified feed rate sequences, and obtaining a first simulated back end cap temperature sequence and a first simulated internal pressure sequence for the first qualified feed rate sequence; obtaining the standard back end cap temperature and standard internal pressure during pyrolysis operation; calculating the deviation of the first simulated back end cap temperature sequence based on the standard back end cap temperature to obtain a first temperature deviation sequence; calculating the deviation of the first simulated internal pressure sequence based on the standard internal pressure to obtain a first pressure deviation sequence; and evaluating and determining the fitness level of a first feed scheme based on the first temperature deviation sequence, the first pressure deviation sequence, and the first qualified feed rate sequence, with the goal of improving the stability of the back end cap temperature, internal pressure, and feed rate, and adding it to the fitness levels of the multiple feed schemes.

[0052] First, from the multiple qualified feed rate sequences obtained after constraint screening, one is randomly selected as the current evaluation object and marked as the first qualified feed rate sequence. At the same time, the first simulated back end temperature sequence and the first simulated internal pressure sequence obtained in the previous simulation are obtained corresponding to the first qualified feed rate sequence.

[0053] A clear reference baseline is required for stability assessment. Therefore, it is necessary to obtain the process parameter values ​​that the pyrolyzer is expected to maintain under ideal stable operating conditions, namely the standard back end cap temperature and standard internal pressure. The standard back end cap temperature and standard internal pressure are process setpoints determined based on optimal product quality, highest energy efficiency, or long-term operating experience. For example, the standard back end cap temperature can be set to 430℃±10℃, and the standard internal pressure can be set to -50Pa to +150Pa.

[0054] Secondly, using the standard back end cap temperature as a benchmark, each predicted back end cap temperature in the first simulated back end cap temperature sequence is compared with it, and the temperature deviation value at each time point is calculated. All the calculated deviation values ​​are arranged in chronological order to form the first temperature deviation sequence. This first temperature deviation sequence directly reflects the future fluctuation of the back end cap temperature from the ideal set point under this feeding scheme. Similarly, using the standard internal pressure as a benchmark, the first simulated internal pressure sequence is processed in the same way to calculate the first pressure deviation sequence, which is used to characterize the future fluctuation of the internal pressure.

[0055] Finally, the fitness is evaluated by combining the first temperature deviation sequence, the first pressure deviation sequence, and the first qualified feed rate sequence. The core idea of ​​this fitness calculation is that an excellent feed scheme should minimize the fluctuations of the temperature deviation sequence and the pressure deviation sequence, while the change in the feed rate sequence itself should also be as gradual as possible, i.e., the feed should be more stable.

[0056] Specifically, the fitness of the first feeding scheme is evaluated and determined based on the first temperature deviation sequence, the first pressure deviation sequence, and the first qualified feed rate sequence, including: calculating the first temperature deviation mean and the first temperature deviation coefficient of variation based on the first temperature deviation sequence; calculating the first pressure deviation mean and the first pressure deviation coefficient of variation based on the first pressure deviation sequence; calculating the first rate coefficient of variation based on the first qualified feed rate sequence; and calculating the fitness of the first feeding scheme by weighting the first temperature deviation mean, the first temperature deviation coefficient of variation, the first pressure deviation mean, the first pressure deviation coefficient of variation, and the first rate coefficient of variation, wherein the fitness of the first feeding scheme is negatively correlated with the first temperature deviation mean, the first temperature deviation coefficient of variation, the first pressure deviation mean, the first pressure deviation coefficient of variation, and the first rate coefficient of variation.

[0057] Specifically, the calculation of the adaptability of the first feeding scheme is a process of integrating multiple stability quantification indicators. Its purpose is to use a single value to comprehensively characterize the overall performance of the feeding scheme in maintaining process stability.

[0058] First, a statistical analysis is performed on the first temperature deviation sequence. The arithmetic mean of all deviation values ​​in the first temperature deviation sequence is calculated to obtain the first temperature deviation mean, which reflects the average degree to which the simulated back end cap temperature deviates from the standard value. Simultaneously, the coefficient of variation of the first temperature deviation sequence is calculated, which is the ratio of the standard deviation to the mean, to obtain the first temperature deviation coefficient of variation. This first temperature deviation coefficient of variation eliminates the influence of dimensions and is used to reflect the relative dispersion of temperature fluctuations, i.e., instability.

[0059] Secondly, the first pressure deviation sequence is processed in the same way. The mean value of the first pressure deviation is calculated to measure the average magnitude of the overall internal pressure deviation from the standard value. At the same time, the coefficient of variation of the first pressure deviation is calculated to measure the relative severity of pressure fluctuations.

[0060] Furthermore, the first qualified feed rate sequence itself is analyzed. The coefficient of variation of this first qualified feed rate sequence is calculated to obtain the first rate coefficient of variation. This first rate coefficient of variation is used to reflect the smoothness of the change in the feed rate within a preset time window. The smaller the coefficient of variation, the more stable the feed rate and the smaller the impact on the equipment.

[0061] Finally, the mean value of the first temperature deviation, the coefficient of variation of the first temperature deviation, the mean value of the first pressure deviation, the coefficient of variation of the first pressure deviation, and the coefficient of variation of the first rate are weighted and integrated into a first feed scheme fitness value. Specifically, in the weighted calculation, the weight coefficients of each stability index are set based on their comprehensive impact on the safety, efficiency, and equipment life of the pyrolysis process. For example, the weight coefficient of the mean value of the first temperature deviation can be set to 0.15, the weight coefficient of the coefficient of variation of the first temperature deviation can be set to 0.15, the weight coefficient of the mean value of the first pressure deviation can be set to 0.15, the weight coefficient of the coefficient of variation of the first pressure deviation can be set to 0.15, and the weight coefficient of the coefficient of variation of the first rate can be set to 0.40. In this example, the stability of the feed rate is given a high weight because it is directly related to the uniformity of material input and has a primary impact on mitigating mechanical shock to the equipment and maintaining the balance of reactants; the mean and fluctuation stability of temperature and pressure are given relatively balanced and important weights to jointly ensure the thermodynamic stability and safety boundary of the reaction process.

[0062] Specifically, the smaller the mean value of the first temperature deviation, the smaller the coefficient of variation of the first temperature deviation, the smaller the mean value of the first pressure deviation, the smaller the coefficient of variation of the first pressure deviation, and the smaller the coefficient of variation of the first rate, the greater the calculated fitness value of the first feeding scheme. This ensures that the fitness value can directly and consistently reflect the relative merits of the candidate feeding schemes. A scheme with a higher fitness value represents a better overall performance in simultaneously ensuring stable back end cap temperature, stable internal pressure, and stable feeding rate, thus providing a clear and reliable target guidance and evaluation benchmark for subsequent intelligent optimization algorithms.

[0063] Finally, based on the calculated fitness values ​​of multiple feeding schemes and the corresponding sequences of multiple qualified feeding rates, a specific optimization algorithm is used for iterative optimization.

[0064] Specifically, based on the fitness of multiple feed schemes and multiple qualified feed rate sequences, the pyrolyzer feed rate is optimized, and a suitable feed rate sequence is output. This includes: setting the qualified feed rate sequence as the initial solution; arranging multiple initial solutions according to the feed scheme fitness from largest to smallest to generate an initial solution sequence; setting the first Q solutions of the initial solution sequence as excellent solutions and the last J solutions as inferior solutions; performing random iso-clustering on the J inferior solutions based on the Q excellent solutions to obtain Q solution sets, where the sum of Q and J is the number of initial solutions, J is M times Q, and M is greater than or equal to 10; and based on the Q solution sets, within each solution set, with the excellent solutions as the direction, optimizing the solutions... The inferior solutions within the set are optimized and adjusted to obtain Q initial updated solution sets. If the updated inferior solution does not satisfy the feed rate adjustment space, a qualified feed rate sequence that did not appear during the optimization process is randomly selected from the feed rate adjustment space and replaced. The solution with the highest fitness of the feed scheme in the solution set is selected as the optimal solution, and the Q initial updated solution sets are updated to obtain Q updated solution sets. Based on the Q updated solution sets, the pyrolyzer feed rate optimization continues until convergence, and Q current updated solution sets are output. The optimal solution with the highest fitness of the feed scheme in the Q current updated solution sets is selected as the suitable feed rate sequence.

[0065] First, all qualified feed rate sequences are defined as initial solutions to the optimization problem. Based on the calculated fitness of the feed schemes, all initial solutions are sorted in descending order, with the solution with the highest fitness placed first, forming an ordered sequence of initial solutions.

[0066] Secondly, the sorted initial solution sequence is divided. The top-ranking, high-performance Q solutions in the initial solution sequence are separately defined as excellent solutions; the bottom-ranking, low-performance J solutions are defined as inferior solutions. The sum of the numbers of Q and J equals the total number of initial solutions, and the number of J is set to be M times the number of Q, where M is an integer greater than or equal to 10. This setting indicates that the number of excellent solutions is far less than the number of inferior solutions, allowing a pattern where a few high-quality individuals guide the majority of individuals in the search during the evolutionary process. This effectively balances the algorithm's convergence speed and global exploration capability, preventing the search process from prematurely falling into local optima.

[0067] Next, random equal-value clustering is performed. Specifically, the Q selected optimal solutions are used as fixed cluster centers, and all J inferior solutions are randomly assigned to the categories represented by each optimal solution, forming Q independent solution sets. Each solution set strictly contains one core optimal solution and a number of inferior solutions equal to J divided by Q, thus constructing multiple parallel local search units guided by high-performance solutions. Subsequently, an optimization and adjustment process is initiated within each solution set. Specifically, using the optimal solutions in the current solution set as the evolutionary reference benchmark and optimization direction, targeted correction and performance improvement are implemented for each inferior solution belonging to that solution set. The correction strategy can be manifested as a vector movement of a certain step size in the solution space along the direction pointing to the optimal solution, or by partially inheriting the structural features of the optimal solution through operations such as cross-recombination. Through such adjustments, the performance of inferior solutions is expected to be improved.

[0068] It is important to note that the boundary constraints defined by the feed rate adjustment space must be strictly followed during this process. For any newly generated solution after adjustment, it must be verified that the feed rate values ​​at all times are within the allowable range of the feed rate adjustment space. If any component of a new solution exceeds this boundary, the solution is deemed invalid. For invalid solutions, a qualified feed rate sequence that has not appeared in the current round or any previous iteration and satisfies the constraints must be randomly generated within the complete feed rate adjustment space to replace it. This mechanism aims to maintain the diversity of the population within the feasible region and avoid premature convergence. After completing the above adjustment and validity processing for inferior solutions in all solution sets, Q updated solution sets are obtained. Each updated solution set contains the original excellent solution and the inferior solutions after optimization and replacement, laying the foundation for subsequent iterative evolution.

[0069] Furthermore, the Q updated solution sets are used as the starting population for a new round of iterations, and the complete steps, including optimization adjustment, constraint verification and replacement of invalid solutions, and updating of optimal solutions, are repeatedly executed. Each iteration aims to promote the evolution of solutions within each solution set towards better performance, while maintaining population diversity by randomly generating new solutions to replace invalid solutions. This iterative process continues until a preset convergence condition is met. The convergence condition is the criterion for determining whether the algorithm should terminate the iteration, and is usually set based on the balance between the optimization objective and computational resources. For example, the convergence condition may include: after 10 consecutive iterations, the fitness value of the optimal solution in all solution sets increases by less than a preset minimum threshold, such as one ten-thousandth.

[0070] When the convergence condition is met, the iteration stops, and the final set of Q solutions obtained at this time is called the Q current updated solution sets. In the Q current updated solution sets, all solutions in each set are traversed, and the fitness values ​​of the feeding schemes are compared. The solution with the highest fitness value is selected. This solution represents the optimal feeding scheme found throughout the entire optimization search process, which, under the premise of satisfying all safety and operational constraints, best improves the stability of the rear end cap temperature, internal pressure, and feed rate.

[0071] Finally, this optimal solution was formally determined as the appropriate feed rate sequence and output for actual feed control of the pyrolyzer.

[0072] S40: Within the preset time window, the pyrolyzer is controlled to continuously pyrolyze waste tires according to the adapted feed rate sequence.

[0073] Specifically, throughout the entire duration of the preset time window, the waste tire feed control of the pyrolyzer follows the instructions specified by the adapted feed rate sequence. Using this sequence as a set trajectory, the instantaneous flow rate of waste tire material entering the pyrolyzer is dynamically and precisely controlled by adjusting the drive motor speed or valve opening of the feeding device, such as the screw feeder. This ensures that the material tracks and matches the target rate value at the corresponding moment in the adapted feed rate sequence in real time. Ultimately, a complete intelligent control loop is achieved, from offline predictive optimization to online closed-loop execution, directly transforming machine learning-based multi-objective optimization decisions into continuous and precise physical execution actions. This execution process ensures a high degree of consistency between process operation and optimization objectives, enabling the feed strategy to proactively suppress operating condition fluctuations rather than passively compensate for them. In actual production, this effectively achieves multiple objectives of improving temperature stability, pressure stability, and feed smoothness, ultimately ensuring the efficient, safe, and stable operation of the continuous pyrolysis process.

[0074] In summary, the embodiments of this application have at least the following technical effects: Compared with the prior art, this application firstly, by integrating machine learning prediction and digital twin simulation, constructs a forward-looking perception and high-fidelity extrapolation capability for pyrolyzer operating conditions, effectively overcoming the response delay problem caused by thermal inertia and hysteresis in traditional feedback control, and realizing proactive prediction and intervention of fluctuations in key parameters such as temperature and pressure. Secondly, by using process safety thresholds as hard constraints and setting the stability of the rear end cap temperature, internal pressure, and feed rate as optimization objectives, a comprehensive and refined multi-objective optimization framework is established. Thirdly, through elite-guided population partitioning, localized directed evolution, and a strict feasibility maintenance mechanism, it can quickly and reliably search for the optimal feed rate sequence that satisfies multiple constraints and has the best overall performance in the complex solution space, improving the automation level and quality of optimization decision-making. Finally, the optimization decision is seamlessly integrated into the actual execution stage.

[0075] In summary, the technical solution of this application significantly reduces the reliance on operator experience, fundamentally improves the operational stability, safety, and overall economic benefits of the waste tire continuous pyrolysis production process, and provides a practical and feasible technical path for realizing the intelligent and efficient operation of this industrial process.

[0076] Example 2, as shown in Figure 2, based on the same inventive concept as the machine learning-based continuous pyrolysis feed control method for waste tires provided in Example 1, this embodiment of the invention also provides a machine learning-based continuous pyrolysis feed control system for waste tires, including: a monitoring and acquisition module 11, used to monitor and acquire the historical back end temperature sequence and historical internal pressure sequence of the pyrolyzer within a historical time window, and to acquire the rotational angular velocity and feed rate sequence of the pyrolyzer set within a future preset time window; and a prediction and acquisition module 12, used to predict the feed rate sequence based on the historical back end temperature sequence, historical internal pressure sequence, rotational angular velocity, and feed rate sequence. The pyrolyzer is configured to take the predicted back end cap temperature sequence and the predicted internal pressure sequence within the preset time window; the optimization output module 13 is used to optimize the pyrolyzer feed rate within the preset time window based on the predicted back end cap temperature threshold and the preset internal pressure threshold of the pyrolyzer, with the goal of improving the stability of the back end cap temperature, internal pressure and feed rate as multiple optimization objectives, and output an adapted feed rate sequence; the feed control module 14 is used to control the continuous pyrolysis feed of waste tires into the pyrolyzer according to the adapted feed rate sequence within the preset time window.

[0077] The monitoring and acquisition module 11 is specifically used to: monitor and acquire the historical back end temperature sequence and historical internal pressure sequence of the pyrolyzer within a historical time window, including: continuously monitoring the back end temperature of the pyrolyzer through a K-type thermocouple and continuously monitoring the internal pressure of the pyrolyzer chamber through a pressure transmitter during the operation of the pyrolyzer, and acquiring the historical back end temperature sequence and historical internal pressure sequence of the pyrolyzer within a historical time window, wherein the pyrolyzer is a continuous pyrolyzer used for continuous pyrolysis of waste tires.

[0078] Specifically, the prediction acquisition module 12 is used to: predict and acquire the predicted rear head temperature sequence and predicted internal pressure sequence of the pyrolyzer within the preset time window based on the historical rear head temperature sequence, historical internal pressure sequence, rotational angular velocity, and feed rate sequence, including: constructing a rear head temperature prediction plugin and an internal pressure prediction plugin based on a long short-term memory network; using the rear head temperature prediction plugin to predict and acquire the predicted rear head temperature sequence of the pyrolyzer within the preset time window based on the historical rear head temperature sequence, rotational angular velocity, and feed rate sequence; and using the internal pressure prediction plugin to predict and acquire the predicted internal pressure sequence of the pyrolyzer within the preset time window based on the historical internal pressure sequence, rotational angular velocity, and feed rate sequence.

[0079] Specifically, a back end cap temperature prediction plugin is constructed based on a Long Short-Term Memory (LSTM) network. This includes: collecting sample back end cap temperature sequence sets, sample rotational angular velocity sets, and sample feed rate sequence sets based on historical operational monitoring records of similar pyrolyzers; using historical back end cap temperature sequences within historical time windows corresponding to different scenarios for different sample back end cap temperature sequences, sample rotational angular velocities, and sample feed rate sequences as sample predicted back end cap temperature sequences to obtain a sample predicted back end cap temperature sequence set. The historical time window has the same time span as the preset time window. Using the sample back end cap temperature sequence set, sample rotational angular velocity set, and sample feed rate sequence set as input, and using the sample predicted back end cap temperature sequence set as supervision, the LSM network is trained until convergence to generate the back end cap temperature prediction plugin.

[0080] Specifically, the optimization output module 13 is used to: constrain the preset rear end cap temperature threshold and preset internal pressure threshold of the pyrolyzer, and with improving the stability of the rear end cap temperature, internal pressure, and feed rate as multiple optimization objectives, optimize the pyrolyzer feed rate within the preset time window based on the predicted rear end cap temperature sequence and predicted internal pressure sequence, and output an adapted feed rate sequence. This includes: obtaining the feed rate adjustment space of the pyrolyzer, and randomly generating several initial feed rate sequences based on the feed rate adjustment space; performing pyrolysis processing simulations based on the predicted rear end cap temperature sequence, predicted internal pressure sequence, and several initial feed rate sequences, and outputting several simulated rear end cap temperature sequences and several simulated internal pressure sequences. Pressure sequence; constrained by satisfying the preset back end cap temperature threshold and the preset internal pressure threshold, based on the plurality of simulated back end cap temperature sequences and the plurality of simulated internal pressure sequences, the plurality of initial feed rate sequences are screened to obtain a plurality of qualified feed rate sequences, and a plurality of simulated back end cap temperature sequences and a plurality of simulated internal pressure sequences are obtained from the plurality of qualified feed rate sequences; with the goal of improving the stability of back end cap temperature, internal pressure and feed rate as multiple optimization objectives, the fitness of multiple feed schemes is evaluated and determined based on the plurality of simulated back end cap temperature sequences and the plurality of simulated internal pressure sequences; based on the fitness of the plurality of feed schemes and the plurality of qualified feed rate sequences, the pyrolyzer feed rate is optimized, and an adapted feed rate sequence is output.

[0081] Specifically, the pyrolysis process is simulated based on the predicted back end cap temperature sequence, the predicted internal pressure sequence, and several initial feed rate sequences, respectively, and several simulated back end cap temperature sequences and several simulated internal pressure sequences are output. This includes: randomly selecting a first initial feed rate sequence from the several initial feed rate sequences; performing operational simulation of the pyrolyzer based on a digital twin to construct a pyrolysis reaction simulation space; and within the pyrolysis reaction simulation space, performing pyrolysis reaction simulation within the preset time window based on the first initial feed rate sequence, the predicted back end cap temperature sequence, and the predicted internal pressure sequence, outputting a first simulated back end cap temperature sequence and a first simulated internal pressure sequence, and adding them to the several simulated back end cap temperature sequences and several simulated internal pressure sequences.

[0082] Specifically, with the goal of improving the stability of the back end cap temperature, internal pressure, and feed rate, multiple feed scheme fitness levels are evaluated and determined based on multiple simulated back end cap temperature sequences and multiple simulated internal pressure sequences. This includes: randomly selecting a first qualified feed rate sequence from the multiple qualified feed rate sequences, and obtaining a first simulated back end cap temperature sequence and a first simulated internal pressure sequence for the first qualified feed rate sequence; obtaining the standard back end cap temperature and standard internal pressure during pyrolysis operation; calculating the deviation of the first simulated back end cap temperature sequence based on the standard back end cap temperature to obtain a first temperature deviation sequence; calculating the deviation of the first simulated internal pressure sequence based on the standard internal pressure to obtain a first pressure deviation sequence; and evaluating and determining the fitness level of a first feed scheme based on the first temperature deviation sequence, the first pressure deviation sequence, and the first qualified feed rate sequence, with the goal of improving the stability of the back end cap temperature, internal pressure, and feed rate, and adding it to the fitness levels of the multiple feed schemes.

[0083] Specifically, the fitness of the first feeding scheme is evaluated and determined based on the first temperature deviation sequence, the first pressure deviation sequence, and the first qualified feed rate sequence, including: calculating the first temperature deviation mean and the first temperature deviation coefficient of variation based on the first temperature deviation sequence; calculating the first pressure deviation mean and the first pressure deviation coefficient of variation based on the first pressure deviation sequence; calculating the first rate coefficient of variation based on the first qualified feed rate sequence; and calculating the fitness of the first feeding scheme by weighting the first temperature deviation mean, the first temperature deviation coefficient of variation, the first pressure deviation mean, the first pressure deviation coefficient of variation, and the first rate coefficient of variation, wherein the fitness of the first feeding scheme is negatively correlated with the first temperature deviation mean, the first temperature deviation coefficient of variation, the first pressure deviation mean, the first pressure deviation coefficient of variation, and the first rate coefficient of variation.

[0084] Specifically, based on the fitness of multiple feed schemes and multiple qualified feed rate sequences, the pyrolyzer feed rate is optimized, and a suitable feed rate sequence is output. This includes: setting the qualified feed rate sequence as the initial solution; arranging multiple initial solutions according to the feed scheme fitness from largest to smallest to generate an initial solution sequence; setting the first Q solutions of the initial solution sequence as excellent solutions and the last J solutions as inferior solutions; performing random iso-clustering on the J inferior solutions based on the Q excellent solutions to obtain Q solution sets, where the sum of Q and J is the number of initial solutions, J is M times Q, and M is greater than or equal to 10; and based on the Q solution sets, within each solution set, with the excellent solutions as the direction, optimizing the solutions... The inferior solutions within the set are optimized and adjusted to obtain Q initial updated solution sets. If the updated inferior solution does not satisfy the feed rate adjustment space, a qualified feed rate sequence that did not appear during the optimization process is randomly selected from the feed rate adjustment space and replaced. The solution with the highest fitness of the feed scheme in the solution set is selected as the optimal solution, and the Q initial updated solution sets are updated to obtain Q updated solution sets. Based on the Q updated solution sets, the pyrolyzer feed rate optimization continues until convergence, and Q current updated solution sets are output. The optimal solution with the highest fitness of the feed scheme in the Q current updated solution sets is selected as the suitable feed rate sequence.

[0085] Specifically, the feed control module 14 is used to control the continuous pyrolysis feed of waste tires into the pyrolyzer according to the adapted feed rate sequence within the preset time window.

[0086] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0087] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0088] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A machine learning-based continuous pyrolysis feeding control method for waste tires, characterized in that, The method includes: monitoring and acquiring the historical back end cap temperature sequence and historical internal pressure sequence of the pyrolyzer within a historical time window, and acquiring the rotational angular velocity and feed rate sequence of the pyrolyzer set within a future preset time window; based on the historical back end cap temperature sequence, historical internal pressure sequence, rotational angular velocity, and feed rate sequence, predicting and acquiring the predicted back end cap temperature sequence and predicted internal pressure sequence of the pyrolyzer within the preset time window; using the preset back end cap temperature threshold and preset internal pressure threshold of the pyrolyzer as constraints, and with improving the stability of the back end cap temperature, internal pressure, and feed rate as multiple optimization objectives, optimizing the pyrolyzer feed rate within the preset time window according to the predicted back end cap temperature sequence and predicted internal pressure sequence, and outputting an adapted feed rate sequence; within the preset time window, controlling the continuous pyrolysis feed of waste tires into the pyrolyzer according to the adapted feed rate sequence.

2. The method for controlling the continuous pyrolysis feed of waste tires based on machine learning according to claim 1, characterized in that, The monitoring and acquisition of the historical back end temperature sequence and historical internal pressure sequence of the pyrolyzer within a historical time window includes: continuously monitoring the back end temperature of the pyrolyzer through a K-type thermocouple and continuously monitoring the internal pressure of the pyrolyzer chamber through a pressure transmitter during the operation of the pyrolyzer, and acquiring the historical back end temperature sequence and historical internal pressure sequence of the pyrolyzer within a historical time window. The pyrolyzer is a continuous pyrolyzer used for continuous pyrolysis of waste tires.

3. The method for controlling the continuous pyrolysis feed of waste tires based on machine learning according to claim 1, characterized in that, Based on the historical back end cap temperature sequence, historical internal pressure sequence, rotational angular velocity, and feed rate sequence, the predicted back end cap temperature sequence and predicted internal pressure sequence of the pyrolyzer within the preset time window are obtained, including: constructing a back end cap temperature prediction plugin and an internal pressure prediction plugin based on a long short-term memory network; using the back end cap temperature prediction plugin, predicting the predicted back end cap temperature sequence of the pyrolyzer within the preset time window based on the historical back end cap temperature sequence, rotational angular velocity, and feed rate sequence; and using the internal pressure prediction plugin, predicting the predicted internal pressure sequence of the pyrolyzer within the preset time window based on the historical internal pressure sequence, rotational angular velocity, and feed rate sequence.

4. The waste tire continuous pyrolysis feeding control method based on machine learning according to claim 3, characterized in that, A post-end temperature prediction plugin based on a Long Short-Term Memory (LSTM) network is constructed, comprising: collecting sample post-end temperature sequence sets, sample rotational angular velocity sets, and sample feed rate sequence sets based on historical operation monitoring records of similar pyrolyzers; using historical post-end temperature sequences within historical time windows corresponding to different scenarios of sample post-end temperature sequences, sample rotational angular velocity sets, and sample feed rate sequences as sample predicted post-end temperature sequences to obtain a sample predicted post-end temperature sequence set, wherein the historical time window has the same time span as the preset time window; using the sample post-end temperature sequence set, sample rotational angular velocity set, and sample feed rate sequence set as input, and using the sample predicted post-end temperature sequence set as supervision, training the LSM network until convergence to generate the post-end temperature prediction plugin.

5. The method for controlling the continuous pyrolysis feed of waste tires based on machine learning according to claim 1, characterized in that, Constrained by preset rear end cap temperature threshold and preset internal pressure threshold of the pyrolyzer, and with the goal of improving the stability of rear end cap temperature, internal pressure, and feed rate, the feed rate of the pyrolyzer within the preset time window is optimized based on the predicted rear end cap temperature sequence and predicted internal pressure sequence, and an adapted feed rate sequence is output. This includes: obtaining the feed rate adjustment space of the pyrolyzer, and randomly generating several initial feed rate sequences based on the feed rate adjustment space; performing pyrolysis processing simulations based on the predicted rear end cap temperature sequence, predicted internal pressure sequence, and several initial feed rate sequences, and outputting several simulated rear end cap temperature sequences and several simulated internal pressure sequences; to meet the preset... Constrained by a rear end cap temperature threshold and a preset internal pressure threshold, several initial feed rate sequences are filtered based on several simulated rear end cap temperature sequences and several simulated internal pressure sequences to obtain several qualified feed rate sequences, as well as several simulated rear end cap temperature sequences and several simulated internal pressure sequences of the several qualified feed rate sequences. With the goal of improving the stability of rear end cap temperature, internal pressure, and feed rate, the fitness of several feed schemes is evaluated and determined based on the several simulated rear end cap temperature sequences and several simulated internal pressure sequences. Based on the fitness of the several feed schemes and the several qualified feed rate sequences, the pyrolyzer feed rate is optimized, and an adapted feed rate sequence is output.

6. The waste tire continuous pyrolysis feeding control method based on machine learning according to claim 5, characterized in that, Based on the predicted back end cap temperature sequence, the predicted internal pressure sequence, and several initial feed rate sequences, a pyrolysis process simulation is performed, outputting several simulated back end cap temperature sequences and several simulated internal pressure sequences. This includes: randomly selecting a first initial feed rate sequence from the several initial feed rate sequences; performing operational simulation of the pyrolyzer based on a digital twin to construct a pyrolysis reaction simulation space; within the pyrolysis reaction simulation space, performing a pyrolysis reaction simulation within the preset time window based on the first initial feed rate sequence, the predicted back end cap temperature sequence, and the predicted internal pressure sequence, outputting a first simulated back end cap temperature sequence and a first simulated internal pressure sequence, and adding them to the several simulated back end cap temperature sequences and several simulated internal pressure sequences.

7. The method for controlling the continuous pyrolysis feed of waste tires based on machine learning according to claim 5, characterized in that, With the goal of improving the stability of the back end cap temperature, internal pressure, and feed rate, multiple feed scheme fitness is evaluated and determined based on multiple simulated back end cap temperature sequences and multiple simulated internal pressure sequences. This includes: randomly selecting a first qualified feed rate sequence from the multiple qualified feed rate sequences, and obtaining a first simulated back end cap temperature sequence and a first simulated internal pressure sequence for the first qualified feed rate sequence; obtaining the standard back end cap temperature and standard internal pressure during pyrolysis operation; calculating the deviation of the first simulated back end cap temperature sequence based on the standard back end cap temperature to obtain a first temperature deviation sequence; calculating the deviation of the first simulated internal pressure sequence based on the standard internal pressure to obtain a first pressure deviation sequence; and evaluating and determining the fitness of a first feed scheme based on the first temperature deviation sequence, the first pressure deviation sequence, and the first qualified feed rate sequence, with the goal of improving the stability of the back end cap temperature, internal pressure, and feed rate, and adding this to the fitness of the multiple feed schemes.

8. The waste tire continuous pyrolysis feeding control method based on machine learning according to claim 7, characterized in that, The fitness of a first feeding scheme is evaluated and determined based on the first temperature deviation sequence, the first pressure deviation sequence, and the first qualified feed rate sequence, including: calculating the first temperature deviation mean and the first temperature deviation coefficient of variation based on the first temperature deviation sequence; calculating the first pressure deviation mean and the first pressure deviation coefficient of variation based on the first pressure deviation sequence; calculating the first rate coefficient of variation based on the first qualified feed rate sequence; and calculating the fitness of the first feeding scheme by weighting the first temperature deviation mean, the first temperature deviation coefficient of variation, the first pressure deviation mean, the first pressure deviation coefficient of variation, and the first rate coefficient of variation, wherein the fitness of the first feeding scheme is negatively correlated with the first temperature deviation mean, the first temperature deviation coefficient of variation, the first pressure deviation mean, the first pressure deviation coefficient of variation, and the first rate coefficient of variation.

9. The method for controlling the continuous pyrolysis feed of waste tires based on machine learning according to claim 6, characterized in that, Based on the fitness of multiple feed schemes and multiple qualified feed rate sequences, the pyrolyzer feed rate is optimized, and a suitable feed rate sequence is output. This includes: setting the qualified feed rate sequence as the initial solution; arranging multiple initial solutions according to the feed scheme fitness from largest to smallest to generate an initial solution sequence; setting the first Q solutions of the initial solution sequence as optimal solutions and the last J solutions as inferior solutions; performing random iso-clustering on the J inferior solutions based on the Q optimal solutions to obtain Q solution sets, where the sum of Q and J is the number of initial solutions, J is M times Q, and M is greater than or equal to 10; and based on the Q solution sets, within each solution set, focusing on optimal solutions, optimizing the solution set... The inferior solutions are optimized and adjusted to obtain Q initial updated solution sets. If the updated inferior solution does not meet the feed rate adjustment space, a qualified feed rate sequence that did not appear in the optimization process is randomly selected from the feed rate adjustment space and replaced. The solution with the highest fitness of the feed scheme in the solution set is selected as the optimal solution, and the Q initial updated solution sets are updated to obtain Q updated solution sets. Based on the Q updated solution sets, the pyrolyzer feed rate optimization continues until convergence, and Q current updated solution sets are output. The optimal solution with the highest fitness of the feed scheme in the Q current updated solution sets is selected as the adapted feed rate sequence.

10. A waste tire continuous pyrolysis feeding control system based on machine learning, characterized in that, The method for controlling the continuous pyrolysis feed of waste tires based on machine learning, as described in any one of claims 1-9, comprises: a monitoring and acquisition module for monitoring and acquiring historical back end cap temperature sequences and historical internal pressure sequences of the pyrolyzer within a historical time window, and acquiring rotational angular velocity and feed rate sequences set by the pyrolyzer within a future preset time window; a prediction and acquisition module for predicting and acquiring predicted back end cap temperature sequences and predicted internal pressure sequences of the pyrolyzer within the preset time window based on the historical back end cap temperature sequences, historical internal pressure sequences, rotational angular velocity, and feed rate sequences; an optimization and output module for optimizing the pyrolyzer feed rate within the preset time window based on the predicted back end cap temperature sequences and predicted internal pressure sequences, with preset back end cap temperature thresholds and preset internal pressure thresholds of the pyrolyzer as constraints, and with improving the stability of the back end cap temperature, internal pressure, and feed rate as multiple optimization objectives, and outputting an adapted feed rate sequence; and a feed control module for controlling the continuous pyrolysis feed of waste tires into the pyrolyzer according to the adapted feed rate sequence within the preset time window.

Images