Adaptive optimization control method and system for energy consumption of an internal mixer and storage medium

Abstract

The application discloses a method and system for adaptive optimization control of energy consumption of an internal mixer and a storage medium, and belongs to the technical field of intelligent control of rubber or plastic processing equipment. The method collects power, temperature, speed and pressure data with the ram down time as the starting point, constructs a process state index integrating temperature rise and time progress, and calculates an energy marginal utility index accordingly. Through online learning, the mixing rule evolution state is identified, and a high-consistency data fitting energy efficiency reference function is used. Based on the deviation of the real-time state point and the reference function, a double-layer controller is used to dynamically generate target power and match the optimal rotor speed and ram pressure combination to drive the system to run along the efficient trajectory. The discharge timing is intelligently determined in combination with the process completion degree, energy efficiency stability and trajectory closeness. The application does not require an offline model, significantly reduces the energy consumption per unit product while ensuring mixing quality, and realizes adaptive closed-loop optimization of energy consumption.

Description

Technical Field

[0001] This application relates to the field of intelligent control technology for rubber or plastic processing equipment, specifically to an adaptive optimization control method, system, and storage medium for energy consumption of internal mixers. Background Technology

[0002] Internal mixers, as core equipment in rubber and plastics mixing processes, account for a very high proportion of energy consumption in the entire production process. While current mainstream control strategies offer adjustable ranges for key process parameters such as speed and pressure, their setting and adjustment heavily rely on operator experience, essentially representing a semi-open-loop feedforward control based on a "formula-parameter" mapping. The fundamental flaw of this approach is its inability to respond promptly and quantitatively to fluctuations in energy utilization efficiency caused by real-time changes in material conditions during mixing. Faced with batch variations in rubber compounds, environmental fluctuations, and equipment status changes, operators or control systems typically can only perform coarse, conservative global parameter compensation based on lagging indicators such as final temperature, time, or power integral (e.g., increasing or decreasing speed throughout the process), failing to dynamically identify and optimize the marginal utility of energy at each moment during mixing. This results in energy input often not being at the optimal efficiency point, leading to the common dilemma of maintaining quality while consuming excessive energy.

[0003] Existing research has attempted to introduce model-based optimization or real-time power feedback, but two major bottlenecks remain: First, static models built offline struggle to match online dynamic processes, exhibiting weak adaptability; second, the lack of a core state indicator that can accurately and in real-time integrate "energy consumption cost" and "process benefits" leads to ambiguous optimization objectives. More importantly, existing methods generally lack the ability to autonomously determine whether the mixing process has entered a stable and efficient operating phase online, making it prone to misjudgments in the initial unstable phase when material properties change drastically, potentially triggering control oscillations.

[0004] Therefore, there is an urgent need to propose a control method that can identify the regularity of mixing online, adaptively construct an energy efficiency reference benchmark, and implement closed-loop optimization based on real-time energy efficiency deviation, so as to achieve refined control of energy consumption and maximize the marginal utility of energy. Summary of the Invention

[0005] This application provides an adaptive optimization control method, system, and computer-readable storage medium for energy consumption of internal mixers, aiming to solve the technical problems of low energy utilization efficiency and poor control robustness caused by the lack of dynamic energy efficiency perception and adaptive optimization mechanism in the current internal mixing process.

[0006] The first aspect disclosed in this application provides an adaptive optimization control method for the energy consumption of an internal mixer, the method comprising:

[0007] The sampling start time is the moment when the top plug is pressed down. At each sampling moment, the instantaneous power of the motor, the temperature of the material in the mixing chamber, the rotor speed and the pressure of the top plug are collected synchronously during the mixing process.

[0008] For each sampling time, based on the material temperature in the mixing chamber at that sampling time, a process state index is calculated, and based on the rate of change of the instantaneous power of the motor and the process state index between adjacent sampling times at that sampling time, an energy marginal utility index is calculated. The process state index is a normalized process progress index, which represents the degree to which the mixing state approaches the preset target at that sampling time. The energy marginal utility index is used to quantify the process state progress caused by a unit energy input, so as to characterize the energy utilization efficiency of the mixing process at that sampling time.

[0009] Based on the instantaneous power of the motor and the energy marginal utility index, an energy state phase diagram is constructed, and the two are mapped to real-time state points in the energy state phase diagram;

[0010] The learning phase begins from the sampling start time. At each sampling time, the following operations are performed until the learning phase ends: Based on the real-time state points obtained up to the sampling time, calculate their consistency index. Based on the number of consistency indices that are not less than a preset consistency threshold, determine that the learning phase has entered a regular evolution state. Then, mark the sampling time as the last sampling time of the learning phase and end the learning phase. Extract the real-time state points corresponding to the sampling times in the learning phase where the consistency index is not less than the preset consistency threshold. Construct an energy efficiency reference function with the instantaneous power of the motor as the independent variable and the energy marginal utility index as the dependent variable. The regular evolution state refers to the fact that the material in the mixing chamber has completed the main dispersion and its mixing state has entered a stable operating stage.

[0011] In response to the last sampling moment of the learning phase, a two-layer optimization controller is started. The outer controller obtains the real-time state point corresponding to each sampling moment after the end of the learning phase. Based on the instantaneous power of the motor in the state point, the characteristic trajectory corresponding to the energy efficiency reference function is obtained. Then, the reference point with the shortest Euclidean distance to the state point on the trajectory is searched, and the power value corresponding to the reference point is recorded as the trajectory reference power. Based on the trajectory reference power, the target power is generated. The inner loop controller responds to the target power and solves the optimal combination command of rotor speed and top bolt pressure with the goal of maximizing the current energy marginal utility index.

[0012] The optimal combination of instructions is input to the actuator of the internal mixer to perform real-time control of the mixing process, optimize the marginal utility of energy, and achieve adaptive optimization of energy consumption in this mixing process.

[0013] The second aspect disclosed in this application provides an adaptive optimization control system for the energy consumption of an internal mixer, the system comprising:

[0014] The multi-source synchronous sampling module is used to collect the instantaneous power of the motor, the temperature of the material in the mixing chamber, the rotor speed and the pressure of the top plug at each sampling moment, with the moment when the top plug is pressed down as the sampling start time.

[0015] The energy marginal calculation module is used to calculate the process state index based on the material temperature in the mixing chamber at each sampling time, and to calculate the energy marginal utility index based on the rate of change between the instantaneous power of the motor and the process state index at adjacent sampling times. The process state index is a normalized process progress index, which represents the degree to which the mixing state approaches the preset target at the sampling time. The energy marginal utility index is used to quantify the process state progress caused by a unit energy input, so as to characterize the energy utilization efficiency of the mixing process at the sampling time.

[0016] A phase diagram construction module is used to construct an energy state phase diagram based on the instantaneous power of the motor and the energy marginal utility index, and map the two to real-time state points in the energy state phase diagram;

[0017] The energy state feature learning module is used to enter the learning phase from the sampling start time. At each sampling time, the following operations are performed until the end of the learning phase: Based on the real-time state points obtained up to the sampling time, its consistency index is calculated, and based on the number of consistency indices that are not less than a preset consistency threshold, it is determined that the system has entered a regular evolution state. Then, the sampling time is marked as the last sampling time of the learning phase, and the learning phase ends. The real-time state points corresponding to the sampling times in the learning phase where the consistency index is not less than the preset consistency threshold are extracted, and an energy efficiency reference function is constructed with the instantaneous power of the motor as the independent variable and the energy marginal utility index as the dependent variable. The regular evolution state refers to the fact that the material in the mixing chamber has completed the main dispersion and its mixing state has entered a stable operating stage.

[0018] A dual-layer energy efficiency optimization control module is used to respond to the last sampling moment of the learning phase by starting the dual-layer optimization controller. The outer controller obtains the real-time state point corresponding to each sampling moment after the end of the learning phase. Based on the instantaneous power of the motor in the state point, it obtains the feature trajectory corresponding to the energy efficiency reference function. Then, it searches for the reference point with the shortest Euclidean distance to the state point on the trajectory and records the power value corresponding to the reference point as the trajectory reference power. Based on the trajectory reference power, a target power is generated. The inner loop controller responds to the target power to maximize the current energy marginal utility index and solves the optimal combination command of rotor speed and top bolt pressure.

[0019] The optimal instruction execution module is used to input the optimal combination of instructions to the actuator of the internal mixer to perform real-time control of the mixing process, optimize the marginal utility of energy, and achieve adaptive optimization of energy consumption in this mixing process.

[0020] A third aspect of this application discloses a computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the steps in the adaptive optimization control method for energy consumption of the internal mixer.

[0021] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0022] Reduce ineffective energy input during the mixing process: By online determination of whether the mixing process has entered a regular evolution state, avoid blindly maintaining high power operation during the stage of violent material disturbance, and effectively reduce redundant energy consumption during the transition period.

[0023] The dynamic guidance system operates in the high-efficiency range: based on the characteristic trajectory defined by the energy efficiency reference function constructed during the learning phase, the shortest Euclidean distance from the current state point to the trajectory is calculated at each sampling time, and the nearest reference point on the trajectory is determined. Then, the power value of the reference point is used as the guidance benchmark to generate the target power command, so that the system state continuously converges to the characteristic trajectory, thereby ensuring that the motor energy input always operates near the optimal working path with the maximum output per unit energy consumption.

[0024] Achieve batch-adaptive energy-saving control: Instead of relying on fixed formulation parameters or offline models, it autonomously establishes energy efficiency benchmarks for each batch of mixing process, overcoming energy consumption fluctuations caused by differences in rubber compounds and improving the universality and robustness of energy-saving strategies.

[0025] In summary, the technical solution provided in this application enables on-demand supply, precise control, and adaptive optimization of energy consumption in the internal mixer, significantly improving energy utilization efficiency and reducing the power consumption per unit product while ensuring mixing quality.

[0026] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart illustrating an adaptive optimization control method for energy consumption of a mixing mill in one embodiment.

[0029] Figure 2 This is an architecture diagram of an adaptive optimization control system for the energy consumption of an internal mixer in one embodiment.

[0030] Figure labeling: 11 Multi-source synchronous sampling module, 12 Energy margin calculation module, 13 Phase diagram construction module, 14 Energy state feature learning module, 15 Two-layer energy efficiency optimization control module, 16 Optimal instruction execution module. Detailed Implementation

[0031] This application provides an adaptive optimization control method, system, and storage medium for the energy consumption of an internal mixer. The internal mixing process suffers from energy waste due to the lack of real-time perception and dynamic optimization capabilities for energy utilization efficiency.

[0032] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0033] It should be noted that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0034] Example 1, as Figure 1 As shown, this application provides an adaptive optimization control method for the energy consumption of an internal mixer, including:

[0035] The sampling start time is the moment when the top plug is pressed down. At each sampling moment, the instantaneous power of the motor, the material temperature in the mixing chamber, the rotor speed and the pressure of the top plug are collected synchronously during the mixing process.

[0036] Specifically, the moment the top plug is depressed marks the official start of the mixing process—at this point, the rubber compound is completely enclosed in the mixing chamber, and the rotor begins to apply shearing and kneading actions to the material, marking the system's entry into a dynamic stage with a clear energy-process coupling relationship. Before this (such as during the feeding and pre-compression stages), the material is not fully filled or is subjected to uneven stress, and the collected data lacks steady-state evolution characteristics. Incorporating this into learning or control would introduce significant noise. Therefore, selecting the moment the top plug is depressed as the sampling starting point ensures that all subsequent state data reflects the true mixing dynamics. The instantaneous power of the motor is output in real time by the inverter or power sensor of the main drive motor of the internal mixer; the rotor speed and top plug pressure are standard operating parameters of the internal mixer, typically provided directly by the equipment's built-in encoder and pressure transmitter; the material temperature inside the mixing chamber is obtained through thermocouples or infrared temperature measuring devices installed on the walls of the mixing chamber or the rotor shaft. All of the above signals are standard monitoring quantities in modern internal mixer control systems, requiring no additional dedicated sensors. Synchronous sampling and timestamp alignment using existing PLCs or industrial controllers are sufficient.

[0037] For each sampling time, based on the material temperature in the mixing chamber at that sampling time, a process state index is calculated, and based on the rate of change of the instantaneous power of the motor and the process state index between adjacent sampling times at that sampling time, an energy marginal utility index is calculated. The process state index is a normalized process progress index, which represents the degree to which the mixing state approaches the preset target at that sampling time. The energy marginal utility index is used to quantify the process state progress caused by a unit energy input, so as to characterize the energy utilization efficiency of the mixing process at that sampling time.

[0038] Furthermore, based on the material temperature inside the mixing chamber at the sampling time, a process state index is calculated, including:

[0039] A preset sampling period is used, with the moment the top bolt is pressed down as the sampling start time t0. A series of discrete sampling times t are generated according to the preset sampling period. i i = 0, 1, 2, ...;

[0040] At each sampling time t i Synchronously collect the instantaneous power P(t) of the motor during the mixing process i ), material temperature T(t) in the mixing chamber i ), rotor speed V(t) i ) and the pressure of the top bolt F(t) i );

[0041] Based on the sampling time t i The collected material temperature T(t) in the mixing chamber i ), calculate sampling time t i The process state index S(t)i The formula is as follows:

[0042] ;

[0043] Where T0 is the initial temperature of the material, T(t) i (t) represents the sampling time. i The material temperature in the mixing chamber, T target Based on the preset temperature reference endpoint for this mixing process, t max The maximum mixing time preset for this mixing process is t, where t is the time from t0 to t... i The mixing time, α(t) i ) and β(t i ) represents the dynamically adjusted weighting coefficients, and satisfies α(t) i )+β(t i =1.

[0044] Specifically, the process state index S(t) i Temperature is a dimensionless scalar defined in the range [0, 1] that unifies two key physical quantities (temperature and time) with different dimensions, both reflecting the progress of mixing, into a single dimensionless progress scale. A value closer to 0 indicates that mixing has just begun, while a value closer to 1 indicates that mixing is nearing completion. Temperature term It directly characterizes the degree of accumulation of heat energy generated by shearing and friction, and is the most direct physical entity for the mixing and dispersion effect; the time term This provides an absolute timeline reference, independent of material properties, ensuring that T... target The preset target temperature characterizes the thermodynamic conditions required for the completion of this mixing and dispersion process. max The preset maximum mixing time characterizes the time guarantee required for uniform distribution. This is achieved through dynamic weighting α(t). i ) and β(t i Adaptive adjustment of S(t) i The design prioritizes temperature completion during the temperature rise phase and time completion during the plateau phase, thus providing a unified description of the multi-stage mixing process. This design not only accommodates process differences between various formulations but also accurately reflects the impact of disturbances such as cooling intervention and feed deviations on the progress, providing a reliable basis for energy marginal utility analysis and adaptive debinding. Target temperature T target Based on the type of raw rubber, the type and amount of fillers, and the thermal stability of the anti-scorching system in the rubber compound formulation, the median of the typical temperature range for complete dispersion is determined using historical batch data. For example, for tread compounds filled with high-structure carbon black, such as SMR20+60phrN220, the temperature rise plateau for sufficient dispersion typically occurs in the 150–160°C range; therefore, a T0 temperature range can be set. target=155℃. This value reflects the minimum thermodynamic driving force required to achieve effective dispersion, not the discharge command. Maximum mixing time t max The value of t is determined based on the internal mixer rotor configuration, rotational speed, fill factor, and Mooney viscosity of the rubber compound, combined with the empirical upper limit of the minimum shear time required for distribution mixing. For example, under conditions of a 37° rotor and 60 rpm, most general-purpose rubber compounds can achieve macroscopic uniform distribution within 300–400 seconds, so t is often taken as... max =400s. This value is intended to provide a normalized scale for the time-dominant phase and to prevent indefinite mixing due to abnormal temperature rise (such as excessive cooling), and is not mandatory at t. max The glue is discharged continuously. Therefore, the above formula not only has clear physical meaning and engineering feasibility, but is also the key quantitative basis for realizing the perception-evaluation-optimization closed-loop control of this invention. This invention also supports setting α and β as preset constants, setting the value range of α to [0.6, 0.8] and the value range of β to [0.4, 0.2]. Conservative formula: α=0.7, β=0.3, for ordinary conveyor belts, general seals, and low-end tire sidewall rubber; aggressive formula: α=0.8, β=0.2, for green tire tread rubber and low rolling resistance high-performance rubber compounds; high-quality requirements: α=0.6, β=0.4, for aircraft tires, high-speed rail shock absorbers, and medical rubber products.

[0045] Furthermore, α(t) i ) and β(t i The weighting coefficients are dynamically adjusted, including:

[0046] , ;

[0047] Where κ is the sensitivity coefficient, with a value ranging from [0.5, 1.5]. This is the normalized temperature difference.

[0048] Specifically, in the initial stage of mixing, the temperature rise rate is rapid and the temperature is highly sensitive to shear work; therefore, temperature dependence should be given a high weight. ≈1; The mixing process enters a period of regular evolution, with temperature rise slowing down or even plateauing. Temperature becomes less sensitive to progress, so the weight of temperature should be reduced, and the process should instead rely on time, α→0. α(t i The core item fT(t) i )= ,fT(t i )∈[0,1], and S(t) i The formula for the time term is consistent with that for α(t). i ) and S(t i The temperature terms of S(t) share the same signal source, causing fluctuations caused by temperature measurement noise to vary cooperatively in both, thus canceling each other out to some extent. iThe overall perturbation of the final value improves the robustness of the system; considering the temperature rise plateau period, fT(t) i ) is approximately constant and is denoted as f At this time, α(t) i )=1-κ·f f Since it is a constant, ,S(t i ) is time t i The linear function ensures that S(t) is a function of the plateau period. i It can grow smoothly and predictably, without nonlinear oscillations caused by the coupling of weights and processes. This provides a stable basis for judging the consistency of state point motion during the "learning phase". If S(t) i The irregular changes in fT(t) make it difficult to extract a stable energy efficiency reference function. Dynamic weights are designed to encode the law of diminishing marginal returns on energy. This law is directly manifested as: as fT(t) changes... i The increase in temperature rise indicates the progression of the temperature increase process, and the increase in fT(t) per unit energy input. i The increment, i.e., the rate of temperature rise, is decreasing. Therefore, fT(t) i ) itself acts as a regulator of α(t) i The basis for this is the most direct and accurate mapping. κ is the sensitivity coefficient, a temperature-dependent reliability decay coefficient, used to ensure S(t) iThe κ value is strictly monotonically non-decreasing as the mixing process progresses (i.e., heating or delaying will not cause a regression in progress assessment), with a value range of [0.5, 1.5], and a preferred value of 1.0. The κ value range is based on the physical limit analysis of the mixing process: the lower limit is 0.5 as an engineering basis. The most conservative physical meaning is that when the temperature rise is close to the target, the system still retains a weight of α = 1 - 0.5 × 1 = 0.5, indicating that the temperature term is always given a weight of no less than 50%. This setting is suitable for rubber compounds that are sensitive to temperature and have a smooth temperature rise curve, ensuring that temperature is always the dominant evaluation indicator. This setting ensures that even if the temperature rise temporarily stalls, the system's assessment of the process status retains sufficient temperature weighting, effectively preventing the risk of insufficient mixing or premature termination due to decreased temperature signal sensitivity. It is a conservative and robust parameter strategy, its core rationale being to mitigate the risk of misjudgment of progress caused by a temperature rise plateau by consistently maintaining a high temperature weighting. The engineering basis for the 1.5 limit, in its most aggressive physical sense, is that when the temperature rise reaches approximately 1 / 1.5 ≈ 67%, the weight α decays to 0, indicating that the system completely stops relying on temperature at two-thirds of the process. For mixtures with a significant temperature rise plateau, such as rubber compounds with a large amount of carbon black, where the later shear heat generation and cooling reach equilibrium and the temperature stagnates, this setting allows the system to decisively switch to an evaluation mode primarily based on time or other parameters before temperature failure occurs, preventing system malfunction. For example, using ethylene propylene diene monomer (EPDM) rubber with a Mooney viscosity of approximately 50 at 125℃ (ML(1+4) 125℃), this type of rubber exhibits stable heat generation and a smooth temperature rise curve, making it a typical synthetic rubber sensitive to temperature history. Adding 40 phr (per 100 parts of rubber) of fast extrusion carbon black (FEF) and an appropriate amount of plasticizer, with T0 = room temperature, T... target =130℃, t max =300 seconds. On the same internal mixer, the adaptive optimization control method of this invention was run with κ=0.7 and κ=1.3 respectively. When κ=0.7, the system consistently gave a high weight to the temperature term throughout the mixing process due to the slow decrease of the α weight. The state point evolved smoothly, and the Mooney viscosity eventually stabilized within the target value (e.g., 55±2), indicating that the mixing was uniform and sufficient. When κ=1.3, the system rapidly reduced its dependence on temperature in the later stage of temperature rise (after about 80% completion). Since the actual temperature rise had not yet reached the target, the system prematurely judged that the progress was close to completion based on the time term, leading to the premature termination of optimization and the switch to conventional control. The Mooney viscosity of the final discharged material was significantly higher (e.g., reaching above 65), showing over-mixing characteristics. This proves that in a formulation with a smooth temperature rise, excessively rapid weight decay can lead to misjudgment. This formulation uses standard Malaysian natural rubber (SMR20) and is filled with 60 phr of high abrasion-resistant carbon black (HAF). Due to its high carbon black content, this formula tends to exhibit a significant temperature plateau during the later stages of mixing as shear heat generation and cooling reach equilibrium. T0 = room temperature, T... target =145℃, t max=400 seconds. On the same internal mixer, using κ=0.7 and κ=1.3 respectively, when κ=1.3: the system detected a plateau after the temperature rise reached about 85%. Since the α weight has rapidly decayed to a low level, the system can quickly and smoothly switch to a state evaluation dominated by the time term, avoiding "idling" during the temperature plateau period. The optimized controller enters the stable tracking state earlier, shortening the total mixing cycle by about 8%, reducing unit energy consumption, and achieving the target carbon black dispersion in the final masterbatch. When κ=0.7: during the temperature rise plateau period, the system still assigns a high weight to the temperature term, continuously expecting the temperature to rise, resulting in a prolonged learning phase. The system repeatedly performs ineffective optimization attempts during the plateau period, causing the state point evolution to oscillate, ultimately extending the total mixing cycle by more than 15% and increasing energy consumption. The balance of the preferred value 1.0 is the theoretical standard decay model, which is robust in most common formulations. It represents the intuitive logic of reducing trust by the amount of progress completed, making it easy to understand and adjust parameters.

[0049] Furthermore, based on the rate of change of the instantaneous power of the motor and the process state index at the sampling time between adjacent sampling times, the energy marginal utility index is calculated, including:

[0050] ;

[0051] Among them, Eu(t) i (t) represents the sampling time. i The current marginal utility index of energy, P(t) i (t) represents the sampling time. i Instantaneous power of the motor under the following conditions, t i-1 Sampling time t i The previous sampling time, t i -t i-1 Let S(t) be the sampling period Δt. i-1 The current process status index corresponding to the previous sampling time.

[0052] Specifically, ; where △S(t) i The change in the process condition index between adjacent sampling times is expressed as the ratio of the change to the sampling period. That is, the rate of change of the discretization, approximately the reciprocal S(t) i Eu(t) i ) represents the current sampling time t i Near the point of origin, it represents the instantaneous rate of progress in the overall process state that can be achieved by a unit energy input; it measures the marginal output of marginal energy. In the mixing process, during the rapid heating phase, a large amount of energy input is quickly converted into temperature rise and dispersion, resulting in a large ΔS and a high Eu(t) value. iA high Eu value indicates that each unit of electricity is effectively driving the process. When entering a regular evolutionary state, energy input is mainly used to maintain the shear and thermal balance of the materials, failing to significantly change their overall state; ΔS≈0, and the Eu value approaches zero, indicating that the additional energy input contributes negligibly to advancing the overall process objective. This invention successfully quantifies and explicitly detects the implicit state of "energy input but process stagnation." Traditional control methods, lacking such indicators, often continue blindly during plateau periods, resulting in energy waste. In this invention, Eu(t)... i As a real-time guiding signal for "energy utilization efficiency," its ultimate value lies in driving the control system to actively avoid inefficient zones and pursue efficient zones, thereby achieving global energy consumption optimization. The system's optimization goal is not to maintain a constant high Eu value, but to maximize Eu(t). i The integral of ) is the total technological progress or the driving force of Eu(t). i ) evolves along a preset efficient trajectory. When Eu(t) i When the value drops to near zero: the outer controller determines whether the current state point deviates from the characteristic trajectory. If it is already on the trajectory but the Eu value is low, it indicates that the trajectory segment itself is not energy efficient, and a new power setpoint may need to be explored. The inner controller will maximize Eu(t) i With the goal of [missing information], try adjusting the combination of rotational speed and pressure. Even during the plateau period, if the material can regain effective shear through parameter fine-tuning, even if ΔS changes from 0 to a very small positive value, Eu(t) i The value will change from 0 to a positive decimal, allowing the system to identify a better operation point. Global optimization is achieved by continuously tracking Eu(t). i The system can automatically identify each stage from efficient heating to plateau maintenance and endpoint determination, and adopt the most economical energy utilization strategy at each stage. For example, during the plateau period, the optimal strategy might be to appropriately reduce the power so that Eu(t) i The value is maintained at a small but positive level, rather than achieving Eu≈0 at high power. This is related to the process completion judgment. Eu(t) i The value remains close to zero, combined with the process state index S(t). i The total amount of 1 has approached or reached 1, which together constitutes the intelligent mixing endpoint criterion: the overall process has basically met the target (S(t)). i When the temperature (Eu(t)) ≈ 1, and further energy input is insufficient to effectively advance the process (Eu(t) ≈ 0), then degreasing the glue at this point can avoid unnecessary over-refining energy consumption while ensuring quality. This is more scientific and economical than simply relying on temperature or time thresholds.

[0053] When Eu(t) i When the energy consumption is less than 0, it indicates that the current energy consumption is causing the process to regress, such as excessive cooling or shear damage to the structure, which is an important early warning signal.

[0054] Based on the current instantaneous power of the motor and the current energy marginal utility index, an energy state phase diagram is constructed, and the two are mapped to real-time state points in the energy state phase diagram.

[0055] Furthermore, an energy state phase diagram is constructed, including:

[0056] Define a two-dimensional coordinate system, with the horizontal axis representing the instantaneous power of the motor and the vertical axis representing the energy marginal utility index;

[0057] At each sampling moment, the collected instantaneous power of the motor and the corresponding energy marginal utility index are mapped to a real-time state point in the coordinate system;

[0058] As the mixing process progresses, the real-time state points are continuously accumulated to form a dynamically evolving energy state phase diagram, which is used to characterize the evolution trajectory of the energy efficiency state.

[0059] Specifically, the system constructs an energy state phase diagram to intuitively and quantitatively characterize the dynamic mapping relationship between energy input and process efficiency during the internal mixing process. The energy state phase diagram uses the instantaneous power of the motor as the horizontal axis and the energy marginal utility index as the vertical axis, forming a two-dimensional state plane. At each sampling time t... i The P(t) data will be obtained synchronously. i ) and Eu(t i This is mapped to a point in the coordinate system, denoted as the real-time state point Q. i =(P(t i Eu(t) i As the mixing process progresses, the system continuously updates the diagram, forming an evolutionary trajectory composed of continuous or discrete state points. The energy state phase diagram reveals the nonlinear law of energy utilization efficiency. The energy efficiency of the mixing process does not change monotonically with power, but usually exhibits a single-peak characteristic of "first rising and then falling": insufficient shear at low power, excessive frictional heat generation at high power, and only reaches the maximum value of energy marginal utility in a certain intermediate power range. The energy state phase diagram can visualize this inherent law, allowing the control system to see the efficient operating region. It provides a data foundation for adaptive learning. During the learning phase, the system automatically identifies whether the mixing has entered a stable and efficient evolutionary state by analyzing the distribution characteristics of state points in the energy state phase diagram (such as central tendency and consistency). Only when the state points show a convergent and smooth trajectory in the phase diagram are they used to construct the energy efficiency reference function, thereby avoiding the inclusion of transitional or abnormal data in the baseline model. The energy efficiency reference function supporting the deviation calculation of the outer optimization controller is essentially an optimal energy efficiency curve (i.e., Eu=Lref(P)) in the energy state phase diagram. During the control phase, the outer controller compares the current real-time state point with the reference curve at the same power (i.e., the vertical deviation). It determines whether the current operation deviates from the optimal energy efficiency point and generates a target power adjustment command accordingly.

[0060] Starting from the sampling start time, the learning phase begins, and the following operations are repeated at each sampling time until the learning phase ends:

[0061] (a) Calculate the consistency index based on the real-time state points in the energy state phase diagram obtained up to the sampling time;

[0062] (b) Based on the number of consistency indicators that are not less than the preset consistency threshold, determine whether the mixing process has entered a regular evolution state;

[0063] (c) If it is determined that the process has entered a regular evolution state, then the sampling time is marked as the last sampling time of the learning phase, the learning phase ends, the real-time state point corresponding to the sampling time in the learning phase where the consistency index is not less than the preset consistency threshold is extracted, and an energy efficiency reference function is constructed with the instantaneous power of the motor as the independent variable and the energy marginal utility index as the dependent variable. The regular evolution state refers to the fact that the material in the mixing chamber has completed the main dispersion and its mixing state has entered a stable operating stage.

[0064] Specifically, the learning phase refers to the initial dynamic adjustment period from the moment the top plug is pressed down until the mixing process enters a stable and efficient operating state. During this phase, the material is in a non-steady-state process with drastic changes in filling, compaction, and shear heat generation. The relationship between energy input and process response is highly time-varying and cannot be directly used to construct a reliable energy efficiency benchmark. To address this, the system employs an online real-time learning mechanism: continuously collecting real-time state points at each sampling moment and calculating its motion consistency index (reflecting the smoothness and convergence of the state trajectory) through a sliding window. The process stage where the mixing process enters a dynamic steady state refers to the point where material dispersion is basically complete, the temperature change rate approaches zero, motor power fluctuations converge, and a repeatable mapping relationship is presented between the energy marginal utility index and operating parameters. To avoid misjudging transient disturbances as steady state, this invention introduces a consistency index to comprehensively evaluate the temporal stability of temperature, power, and Eu. Only when this consistency index continuously meets the convergence threshold within a preset time window is the system determined to have truly entered a regular evolutionary state. Once a consistent evolutionary state is detected, the learning phase ends immediately, and only high-quality state points that meet the consistency criteria within this phase are used to fit the energy efficiency reference function. Because these points are all from the system's truly efficient operating range, the constructed reference benchmark closely matches the actual energy efficiency characteristics of the current batch of rubber compound, avoiding mismatch issues caused by differences in formulation, environment, or equipment in offline models. The typical learning phase duration is 10–60 seconds, depending on the rubber compound type and initial state, much shorter than a complete mixing cycle (typically 120–400 seconds), and does not affect production cycle time. This design ensures that it does not rely on prior knowledge, does not introduce human intervention, and does not use unstable data, thereby obtaining an adaptive, highly reliable energy efficiency reference benchmark, providing a reliable basis for subsequent closed-loop optimization.

[0065] Furthermore, a feature trajectory is fitted based on all real-time state points collected during the learning phase, including:

[0066] The learning phase begins at the start of the sampling period, and the following loop is executed at each sampling time until the learning phase ends:

[0067] (a) Up to the sampling time, record the most recent M consecutive state points to form a sliding state sequence {Q1, Q2, ..., Q...} M}, M≥3, the time period covered by the sliding state sequence constitutes a local time window;

[0068] (b) Based on the sliding state sequence {Q1, Q2, ..., Q...} M}, calculate the displacement vector between adjacent state points within it, and normalize each displacement vector to obtain multiple unit vectors;

[0069] (c) Calculate the magnitude of the average vector of the unit vector as the consistency index C, where 0≤C≤1, and store the calculated consistency index C in the pre-constructed consistency index sequence in chronological order.

[0070] (d) In the consistency index sequence, take the K most recent consecutive consistency indices to form a decision window, and count the number of indices that satisfy C≥ preset consistency threshold. The value of K should be such that the total time covered by the decision window is greater than twice the local time window.

[0071] (e) If the ratio of the number to K is less than the preset information ratio, it is determined that the mixing process has not yet entered the regular evolution state, waits for the next sampling time, and returns to (a);

[0072] (f) If the ratio of the number to K is greater than or equal to a preset confidence ratio, and the mixing process is determined to have entered a regular evolution state, then the sampling time is marked as the last sampling time t of the learning phase. R And thus conclude the learning phase;

[0073] After the learning phase is completed, the energy efficiency reference function is obtained by fitting a quadratic polynomial regression method.

[0074] Specifically, starting from the initial sampling time (i.e., the moment the top plug is pressed down), the system enters the learning phase and executes the following online discrimination and data filtering process at each sampling time until it is determined that the mixing process has entered a regular evolution state and the learning ends:

[0075] First, up to the current sampling time, the system records the most recent M consecutive real-time state points (M≥3), forming a sliding state sequence {Q1, Q2, ..., Q...}. M The time period covered by this sequence is called the local time window, used to capture the recent dynamic behavior of the system. Subsequently, based on this sequence, the displacement vector between adjacent state points is calculated. j=1,2...,M,P j+1 -P j For displacement vector V j The components on the horizontal axis of the energy state phase diagram represent the change in power, Eu. j+1 -Eu j For displacement vector V j The components on the vertical axis of the energy state phase diagram represent the changes in marginal utility of energy. Each displacement vector is then normalized to obtain... There are unit direction vectors. These unit vectors reflect the local uniformity of the system's direction of motion in the energy state phase diagram. Normalization involves dividing the vector by its own magnitude (length) to transform it into a unit direction vector with a length of 1 and unchanged direction. For example, the normalized result for a displacement vector is:

[0076] ,

[0077] This operation will convert displacement vectors V of different lengths. j The vectors are mapped to unit vectors of length 1. The direction of each unit vector is exactly the same as the original displacement vector, but its magnitude information is stripped. This makes subsequent consistency metrics depend only on the direction, and not on the speed of state point movement, i.e., the drastic degree of state change between adjacent sampling points. Further, the average vector of all unit vectors is calculated, and its magnitude is taken as the consistency index C. Since the average magnitude of the unit vectors is in the range [0, 1], the closer C is to 1, the smoother the state point trajectory, the more consistent the direction, and the more stable the system operation; conversely, it indicates that the process is in a state of drastic disturbance or divergence. This C value is stored in a pre-constructed consistency index sequence in chronological order. Further, the system... The average vector is obtained by averaging the unit vectors. The magnitude of this average vector is used as the consistency index C.

[0078] ;

[0079] The index C has clear geometric and physical meaning: if the state points within the sliding window move smoothly along an approximately straight line trajectory in the energy state phase diagram (indicating stable system behavior and strong evolutionary regularity), then the directions of each unit vector are similar, and their average vector magnitude is close to 1; if the trajectory of the state points changes drastically, diverges, or jumps randomly (such as uncompacted materials in the early stage of mixing, rotor slippage, and other unsteady-state processes), then the directions of the unit vectors are dispersed and cancel each other out, and the average vector magnitude approaches 0. Therefore, C∈[0,1] essentially quantifies the directional consistency of the local trajectory and is an effective criterion for judging whether the mixing process has entered a regular evolutionary state. This method does not require a pre-set model and only relies on the temporal geometric characteristics of the state points, possessing strong robustness and real-time performance. Next, the most recent K consecutive consistency indices are selected in this sequence to form a judgment window, where the value of K satisfies: the total time covered by the judgment window is greater than twice the local time window. This design ensures that the criterion has sufficient temporal coverage and avoids misjudgment due to instantaneous stability. The time scale relationship is clarified, and typical values ​​are shown in Table 1. If C≥C in the judgment window th If the proportion of the indicators is not less than the preset information ratio η, then the mixing process is judged to have entered a regular evolutionary state. thA preset consistency threshold, preferably 0.85, is set, and η is preferably 0.7. At this point, the current sampling time is marked as the last sampling time of the learning phase, and learning is terminated. After the learning phase ends, the system extracts all samples satisfying C≥C from all sampling times within that phase. th The sampling times correspond to real-time state points (P, Eu), most of which come from stable and efficient operating ranges and have high reliability. Finally, using instantaneous motor power as the independent variable and energy marginal utility index as the dependent variable, a quadratic polynomial regression method is used to fit the selected data points to obtain the energy efficiency reference function Lref(P). This function represents the optimal power-energy efficiency mapping relationship of the current batch under the regular evolution state, and serves as the energy efficiency reference benchmark for subsequent control stages, used for real-time deviation calculation and target power generation.

[0080] Table 1

[0081]

[0082] Furthermore, after the learning phase ends, the energy efficiency reference function is obtained by fitting a quadratic polynomial regression method, including:

[0083] (a) Extract the real-time state points corresponding to the sampling times during the learning phase where the consistency index C is not less than the preset consistency threshold, and construct the fitting dataset. Each real-time state point is represented by the instantaneous motor power P(t) at that sampling time. i ) and the marginal utility index of energy Eu(t) i Composed of ), where i∈I, To meet the consistency index C(t) i )≥C th The set of sampling indices;

[0084] (b) The effective reference function is defined as a quadratic polynomial function, as shown in the following formula:

[0085] Eu(t i )=a·P(t i ) 2 +b·P(t i )+c;

[0086] Where a, b, and c are the coefficients to be solved;

[0087] (c) All state points (P(t) in the fitted dataset i Eu(t) i Using these as inputs, the coefficients a, b, and c are solved using the least squares method to minimize the sum of squared errors between the current energy marginal utility index of each state point in the fitted dataset and the predicted values ​​of the quadratic polynomial model. The formula for the sum of squared errors is:

[0088] ;

[0089] (d) Save the obtained coefficients a, b, and c as fixed parameters to form the energy efficiency reference function, L. ref (P)=Eu(P)=a·P 2 +b·P+c, at each sampling time after the learning phase ends, based on the corresponding collected instantaneous motor power P(t) i Substitute this value into the energy efficiency reference function to calculate the corresponding reference value at that sampling time, which will be used for subsequent energy efficiency deviation analysis and adaptive control.

[0090] Specifically, after the learning phase, the system selects all sampling moments within that phase where the consistency index C is not less than a preset consistency threshold, and extracts their corresponding real-time state points. Each state point consists of the instantaneous motor power P and the energy marginal utility index Eu at that moment, forming a high-quality fitting dataset. Based on this dataset, the system uses a quadratic polynomial regression method to construct the energy efficiency reference function L. ref (P). The coefficients a, b, and c are automatically solved using the least squares method to minimize the overall deviation between the predicted Eu values ​​and the actual observed values ​​across all selected data points. This process is automatically completed by the controller's built-in algorithm without manual intervention. The resulting energy efficiency reference function L... ref (P) is stored and used as the energy efficiency benchmark for subsequent control phases. At each sampling moment after the learning phase ends, the system substitutes the currently acquired instantaneous motor power P(ti) into L. ref (P) can be used to calculate the theoretically optimal energy marginal utility reference value at that power, which can be used to compare with the real-time Eu(t) i The comparison generates an energy efficiency deviation signal, which drives the target power adjustment.

[0091] In response to the last sampling moment of the learning phase, a two-layer optimization controller is activated. The outer controller acquires the real-time state point corresponding to each sampling moment after the end of the learning phase. Based on the instantaneous power of the motor at the state point, it obtains the characteristic trajectory corresponding to the energy efficiency reference function. Then, it searches for the reference point on the trajectory with the shortest Euclidean distance to the state point and records the power value corresponding to the reference point as the trajectory reference power. Based on the trajectory reference power, a target power is generated. The inner loop controller responds to the target power and, with the goal of maximizing the current energy marginal utility index, solves for the optimal combination command of rotor speed and top bolt pressure.

[0092] Furthermore, the optimal combination command for determining the rotor speed and the pressure of the upper jack is calculated, including:

[0093] Once it is determined that the mixing process has entered a regular evolutionary state, the response is at the last sampling time t.R The dual-layer optimization controller is activated, and the outer layer controller performs the following steps:

[0094] At each sampling time t after the end of the learning phase i i = R+1, R+2, ..., obtain the real-time state point (P(t) in the energy state phase diagram. i Eu(t) i ));

[0095] Based on the energy efficiency reference function, a characteristic trajectory is defined. Using the characteristic trajectory as a reference, a point (P(t)) is determined on the characteristic trajectory that corresponds to the real-time state point (P(t)). i Eu(t) i The reference point with the shortest Euclidean distance is denoted as (P). near Eu near );

[0096] Calculate the energy efficiency deviation ΔP = P(t) i )-P near , i = R+1, R+2, ...;

[0097] Based on the energy efficiency deviation, the target power is generated using the following formula:

[0098] P ref =P(t i )-λ·△P;

[0099] Among them, P ref Let λ be the instantaneous target power, λ∈(0,1], and λ be the preset gain coefficient.

[0100] After the learning phase is completed, the inner loop controller constructs a real-time, high-quality data window, including the following steps:

[0101] At each sampling time t after the end of the learning phase i The P(t) corresponding to the sampling time is only selected if the following conditions are met simultaneously. i Eu(t) i ), T(t) i ), V(t) i ) and F(t i ) are denoted as valid entries, where i = R+1, R+2, ...:

[0102] (1) The Euclidean distance calculated by the outer controller is less than the preset deviation threshold;

[0103] (2) The absolute value of the material temperature change rate in the mixing chamber calculated based on adjacent sampling times is less than the preset temperature change rate threshold; the valid entries are stored in real-time high-quality data window in chronological order, wherein the data window retains all valid entries within the most recent γ minutes, where γ is the preset time window length, and each valid entry contains at least each sampling time t after the feature trajectory has been constructed. i V(t) i ), F(t) i ), P(t i ) and Eu(t i ), i = R+1, R+2, ...;

[0104] In response to the received instantaneous target power P ref Retrieve one or more candidate entries from the data window that satisfy the condition that the difference between the instantaneous power of the motor and the target power is within a preset tolerance range.

[0105] If there are candidate entries, the entry with the highest energy marginal utility index is selected, and the rotor speed and upper bolt pressure recorded in that entry are output as the optimal combination command.

[0106] If no candidate entry is found, the rotor speed and the pressure command of the top bolt remain unchanged.

[0107] Specifically, once the mixing process is determined to have entered a regular evolutionary state, the system responds by activating a two-layer optimization controller at the last sampling moment of the learning phase. This controller consists of an outer energy efficiency decision layer and an inner execution mapping layer, which work together to achieve closed-loop optimization of target generation and instruction solving. The outer controller activates at each subsequent sampling moment t... i (i>R) executes in a loop, its core function being to act as a navigator, guiding the real-time operation based on a fixed characteristic trajectory. The specific process is as follows: First, at each sampling moment, the outer controller obtains the current real-time state point (P(t)) from the energy state phase diagram. i Eu(t) i Next, the controller uses the pre-stored characteristic trajectory function Eu=f(P) as a static reference. Through geometric calculation, it determines the projection point or neighboring point on the trajectory with the shortest Euclidean distance to the current real-time state point, and labels the coordinates of this point as (P). near Eu near ), where Eu(near) = f(P(near)). Subsequently, the controller calculates the energy efficiency deviation ΔP = P(t). i )-P near This deviation directly quantifies the deviation of the current actual operating power from the nearest reference power on the characteristic trajectory. Based on this deviation, the outer controller generates the instantaneous target power P. refIts generation logic is as follows: In order to reduce the deviation from the reference trajectory, the target power should be increased by increasing the reference power P based on the current power. near The direction is adjusted proportionally. The adjustment range is controlled by a preset gain coefficient λ to ensure a smooth adjustment process. Preferably, the gain coefficient λ ranges from 0.3 to 0.7. In a specific embodiment, λ is 0.5. This value achieves a good balance between response speed and control stability, enabling the system state to converge quickly and without overshoot to the characteristic trajectory, meeting the robustness requirements of industrial applications. The target power P ref It is always a positive value, reflecting the active energy input from the motor to the mixing chamber. In a preferred embodiment, the system further ensures that P... ref ≥P min , where P min The minimum power threshold required to maintain the basic shear action of the mixing process. The characteristic trajectory is explicitly defined by an energy efficiency reference function, and any point on it can be calculated by specifying a power value P and substituting it into the function. In one embodiment, to improve online calculation efficiency, after the learning phase, the system operates within the effective power range of the motor [P]. min P max The data is discretized using a preset step size, such as 0.5kW, to generate a table of trajectory points. Each entry contains a power value P. d and its corresponding reference energy efficiency value Lref(P) d At each subsequent sampling time, the outer controller iterates through the table and calculates the Euclidean distance between each trajectory point and the current real-time state point, determines the trajectory point with the smallest distance, and refines it through local interpolation to finally obtain the optimal reference power P. near (t iThis method is computationally efficient and robust, and is suitable for industrial real-time control systems. II. Inner layer controller: Execution instruction retrieval based on historical high-quality data. The inner layer controller is responsible for converting the instantaneous target power generated by the outer layer into specific rotor speed and upper bolt pressure combination instructions. In essence, it is a data-driven execution mapper, avoiding reliance on complex mechanism models. Its workflow is as follows: Construct a real-time high-quality data window. At each sampling moment after the learning phase ends, the current working condition data is recorded as a "valid entry" and stored in the data window only when the following conditions are met simultaneously: (1) The shortest Euclidean distance from the real-time state point to the energy efficiency reference trajectory is less than 0.1; (2) The absolute value of the material temperature change rate at adjacent sampling moments is less than 2℃ / s. The data window adopts a sliding time window mechanism to retain all valid entries within the most recent γ minutes. The value range of γ is usually 30 to 90 seconds, preferably γ = 60 seconds. This duration is sufficient to cover the steady-state operation segment of the typical mixing process, usually lasting more than 30 seconds, while avoiding the inclusion of unsteady-state data from premature batches or transitional phases, ensuring that the data window contains only high-efficiency, thermally balanced, and effective operation records. Furthermore, γ is significantly longer than the learning phase duration (10–60 seconds) and the local decision window (approximately 5–10 seconds), forming a clear phase transition: the learning phase is used to construct feature trajectories, and the subsequent regular evolution phase achieves online adaptive optimization through a sliding data window. Instruction retrieval and output: When the target power P issued by the outer layer is received... ref Then, the inner controller retrieves all data that meet the requirements from the data window. Candidate entries, The mean absolute deviation (MAD) of motor power for all valid entries within the data window is used to characterize the level of natural fluctuations under the current steady-state operation. The preset relative proportionality coefficient ranges from 0.01 to 0.03, ensuring a reasonable minimum search window is maintained even under low-fluctuation operating conditions, with a preferred value of 0.02. This dynamic mechanism can adapt to fluctuations caused by different batches or formulas, and avoid the lack of candidate entries due to excessively small tolerance, significantly improving the robustness and practicality of the inner layer controller. If candidate entries exist, the one with the highest energy marginal utility index Eu is selected, and its recorded rotor speed and top bolt pressure are output as the optimal combination command; if no candidate entries exist, the current actuator settings remain unchanged to ensure system stability. III. Adaptive Optimization of Energy Consumption through Dual-Layer Collaboration This invention, through the above dual-layer architecture, optimizes S(t) energy consumption. i ) and Eu(t iThe quantitative results are truly translated into energy-saving control: the outer layer dynamically generates the target power with optimal energy efficiency by utilizing the deviation between Eu and the reference function; the inner layer intelligently matches the optimal combination of process parameters to achieve this power using historical high-quality data, avoiding blind trial and error. The entire process requires no manual intervention, does not rely on offline models, and is entirely based on real and efficient data from the current batch under regular evolutionary conditions. This achieves closed-loop adaptive optimization of "perceiving energy efficiency—generating targets—matching instructions," significantly reducing ineffective energy consumption while ensuring mixing quality.

[0108] Furthermore, it also includes:

[0109] After a mixing operation is completed, cluster analysis is performed on all valid entries in the data window to extract one or more steady-state operation modes that represent the optimal efficiency of this mixing operation. These modes are then associated with the process identifier of this mixing operation and stored in a cross-batch global experience base to provide initial operation parameter recommendations for subsequent mixing operations with the same process identifier.

[0110] Specifically, after a mixing operation is completed, cluster analysis (such as K-means or density clustering) can be performed on all valid entries in the real-time high-quality data window. Entries with similar motor power, rotor speed, top bolt pressure, and high energy marginal utility indicators are grouped into one category; the center point or high-density area of ​​each category represents a steady-state, high-efficiency operating mode. After associating this mode with the process identifier of this mixing operation (such as formula number, rubber compound type), it is stored in a cross-batch global experience database to provide initial operating parameter recommendations when starting subsequent operations with the same process, accelerating the entry into a high-efficiency operating state.

[0111] The optimal combination of instructions is input to the actuator of the internal mixer to perform real-time control of the mixing process, optimize the marginal utility of energy, and achieve adaptive optimization of energy consumption in this mixing process.

[0112] Specifically, the optimal rotor speed command and top bolt pressure command are sent to the actuators of the internal mixer (including the main motor frequency converter and the hydraulic / pneumatic top bolt control system) to adjust the rotor shear strength and material compaction degree, thereby changing the energy input method and material response characteristics. Since the instantaneous power P of the motor is a function of the rotor speed and load torque, and the energy marginal utility index Eu reflects the process progress brought by unit power, adjusting the speed and pressure will directly change P and Eu at the next sampling moment, that is, drive the real-time state point in the energy state phase diagram to move to a new position. By selecting the optimal parameter combination corresponding to the target power from the historical high-efficiency data by the inner controller, the new state point can be made as close as possible to the characteristic trajectory defined by the energy efficiency reference function Lref(P), avoiding falling into the inefficient region (such as the high power low Eu region). As the control continues, the real-time state point evolves stably near the characteristic trajectory in the energy state phase diagram, indicating that the system always operates in the high-efficiency range under the current batch conditions, maximizing the effective mixing work converted from unit electrical energy. Therefore, while ensuring the quality of mixing, ineffective energy consumption is significantly reduced, achieving adaptive optimization of energy consumption in this mixing process.

[0113] The mixing process will end only if the following conditions are met after the learning phase:

[0114] (1) The current process state index S(t) i )≥Preset first threshold, where i=R+1, R+2, ..., N;

[0115] (2) In the energy state phase diagram, the real-time state point (P(t)) i Eu(t) i The shortest Euclidean distance to the feature trajectory is less than or equal to the preset end tolerance.

[0116] (3) Within X consecutive sampling periods, the current energy marginal utility index Eu(t) i The rate of change with respect to time satisfies:

[0117] Where X ≥ 3, The preset steady-state threshold value is preferred. Eu(t) i ) represents each sampling time t after the end of the learning phase. i The current marginal utility index of energy collected, Eu(t) i-1 ) represents the current energy marginal utility index collected at the previous sampling time, and Δt represents the sampling period.

[0118] Specifically, after the learning phase ends, the system continuously monitors the mixing process and only triggers the discharge command to end the mixing process when the following three conditions are met simultaneously: (1) Process completion meets the standard: the current process status index S(t) i (1) The temperature rise of the material is not less than the preset first threshold Sth (e.g., 0.95), indicating that the temperature rise and time progress of the material are close to the target endpoint; (2) The operation is in the high-efficiency range: the real-time state point (P(t) in the energy phase diagram) i Eu(t) i (2) The shortest Euclidean distance to the energy efficiency reference function Lref(P) does not exceed the preset end tolerance, ensuring that the system does not deviate from the high energy efficiency operating zone; (3) Energy efficiency enters a stable plateau period: within X consecutive sampling periods (X≥3), the energy marginal utility index Eu(t) i The absolute value of ) is not lower than the preset energy efficiency lower limit Eu. min Furthermore, the absolute value of the rate of change at adjacent moments is less than the preset steady-state threshold ε, indicating that the mixing effect has reached saturation, and the marginal benefit of continuing to invest energy is extremely low. min =α·E max E max The energy efficiency reference function is the maximum value within its domain, and α is a preset proportionality coefficient, typically ranging from 0.85 to 0.95, with α=0.90 being preferred. This termination strategy avoids the blindness of traditional fixed-time or single-temperature glue removal, ensuring precise termination under the triple conditions of process compliance, high energy efficiency, and gain saturation. It prevents both under-mixing and ineffective over-mixing, truly achieving the adaptive control goal of "mixing on demand and energy-saving glue removal".

[0119] Example 2, based on the same inventive concept as the adaptive optimization control method for the energy consumption of the internal mixer in the foregoing examples, such as... Figure 2 As shown, this application provides an adaptive optimization control system for the energy consumption of an internal mixer, including:

[0120] Multi-source synchronous sampling module 11 is used to collect the instantaneous power of the motor, the temperature of the material in the mixing chamber, the rotor speed and the pressure of the top plug at each sampling moment, with the time when the top plug is pressed down as the sampling start time.

[0121] The energy marginal calculation module 12 is used to calculate the process state index based on the material temperature in the mixing chamber at each sampling time, and to calculate the energy marginal utility index based on the rate of change between the instantaneous power of the motor and the process state index at adjacent sampling times. The process state index is a normalized process progress index, which represents the degree to which the mixing state approaches the preset target at the sampling time. The energy marginal utility index is used to quantify the process state progress caused by unit energy input, so as to characterize the energy utilization efficiency of the mixing process at the sampling time.

[0122] Phase diagram construction module 13 is used to construct an energy state phase diagram based on the instantaneous power of the motor and the energy marginal utility index, and map the two to real-time state points in the energy state phase diagram;

[0123] The energy state feature learning module 14 is used to enter the learning phase from the sampling start time. At each sampling time, the following operations are performed until the end of the learning phase: Based on the real-time state points obtained up to the sampling time, its consistency index is calculated, and based on the number of consistency indices that are not less than a preset consistency threshold, it is determined whether the mixing process has entered a regular evolution state; when it is determined that the regular evolution state has been entered, the sampling time is marked as the last sampling time of the learning phase, the learning phase ends, the real-time state points corresponding to the sampling times in the learning phase where the consistency index is not less than the preset consistency threshold are extracted, and an energy efficiency reference function is constructed with the instantaneous power of the motor as the independent variable and the energy marginal utility index as the dependent variable. The regular evolution state refers to the fact that the material in the mixing chamber has completed the main dispersion and its mixing state has entered a stable operating stage.

[0124] The dual-layer energy efficiency optimization control module 15 is used to start the dual-layer optimization controller in response to the last sampling moment of the learning phase. The outer layer controller obtains the real-time state point corresponding to each sampling moment after the end of the learning phase. Based on the instantaneous power of the motor in the state point, it obtains the feature trajectory corresponding to the energy efficiency reference function. Then, it searches for the reference point with the shortest Euclidean distance to the state point on the trajectory and records the power value corresponding to the reference point as the trajectory reference power. Based on the trajectory reference power, it generates the target power. The inner loop controller responds to the target power and solves the optimal combination command of rotor speed and upper bolt pressure with the goal of maximizing the current energy marginal utility index.

[0125] The optimal instruction execution module 16 is used to input the optimal combination of instructions to the actuator of the internal mixer to perform real-time control of the mixing process, optimize the marginal utility of energy, and achieve adaptive optimization of energy consumption in this mixing process.

[0126] In embodiment three, this application provides a storage medium that stores one or more programs that can be executed by one or more processors to implement the steps in the adaptive optimization control method for energy consumption of the internal mixer described in any of the above embodiments.

[0127] The storage media described herein include RAM, memory, ROM, EEPROM, registers, hard disks, removable disks, or any other form of storage media known in the art.

[0128] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications or additions should fall within the protection scope of the present invention.

Claims

1. An adaptive optimization control method for energy consumption of an internal mixer, characterized in that, The method includes the following steps: The sampling start time is the moment when the top plug is pressed down. At each sampling moment, the instantaneous power of the motor, the temperature of the material in the mixing chamber, the rotor speed and the pressure of the top plug are collected synchronously during the mixing process. For each sampling time, based on the material temperature in the mixing chamber at that sampling time, a process state index is calculated, and based on the rate of change of the instantaneous power of the motor and the process state index between adjacent sampling times at that sampling time, an energy marginal utility index is calculated. The process state index is a normalized process progress index, which represents the degree to which the mixing state approaches the preset target at that sampling time. The energy marginal utility index is used to quantify the process state progress caused by a unit energy input, so as to characterize the energy utilization efficiency of the mixing process at that sampling time. Based on the instantaneous power of the motor and the energy marginal utility index, an energy state phase diagram is constructed, and the two are mapped to real-time state points in the energy state phase diagram; The learning phase begins from the sampling start time. At each sampling time, the following operations are performed until the learning phase ends: Based on the real-time state points obtained up to the sampling time, calculate their consistency index. Based on the number of consistency indices that are not less than a preset consistency threshold, determine that the learning phase has entered a regular evolution state. Then, mark the sampling time as the last sampling time of the learning phase and end the learning phase. Extract the real-time state points corresponding to the sampling times in the learning phase where the consistency index is not less than the preset consistency threshold. Construct an energy efficiency reference function with the instantaneous power of the motor as the independent variable and the energy marginal utility index as the dependent variable. The regular evolution state refers to the fact that the material in the mixing chamber has completed the main dispersion and its mixing state has entered a stable operating stage. In response to the last sampling moment of the learning phase, a two-layer optimization controller is started. The outer controller obtains the real-time state point corresponding to each sampling moment after the end of the learning phase. Based on the instantaneous power of the motor in the state point, the characteristic trajectory corresponding to the energy efficiency reference function is obtained. Then, the reference point with the shortest Euclidean distance to the state point on the trajectory is searched, and the power value corresponding to the reference point is recorded as the trajectory reference power. Based on the trajectory reference power, the target power is generated. The inner loop controller responds to the target power and solves the optimal combination command of rotor speed and top bolt pressure with the goal of maximizing the current energy marginal utility index. The optimal combination of instructions is input to the actuator of the internal mixer to perform real-time control of the mixing process, optimize the marginal utility of energy, and achieve adaptive optimization of energy consumption in this mixing process. The process state index is calculated based on the material temperature inside the mixing chamber at the sampling time, including: A preset sampling period is used, with the moment the top bolt is pressed down as the sampling start time t0. A series of discrete sampling times t are generated according to the preset sampling period. i i = 0, 1, 2, ...; At each sampling time t i Synchronously collect the instantaneous power P(t) of the motor during the mixing process i ), material temperature T(t) in the mixing chamber i ), rotor speed V(t) i ) and the pressure of the top bolt F(t) i ); Based on the sampling time t i The collected material temperature T(t) in the mixing chamber i ), calculate sampling time t i The process state index S(t) i The formula is as follows: ； Where T0 is the initial temperature of the material, T(t) i (t) represents the sampling time. i The material temperature in the mixing chamber, T target Based on the preset temperature reference endpoint for this mixing process, t max The maximum mixing time preset for this mixing process is t, where t is the time from t0 to t... i The mixing time, α(t) i ) and β(t i ) represents the dynamically adjusted weighting coefficients, and satisfies α(t) i )+β(t i )=1; α(t i ) and β(t i The weighting coefficients are dynamically adjusted, including: ， ； Where κ is the sensitivity coefficient, with a value ranging from [0.5, 1.5]. To normalize the absolute temperature difference; Based on the rate of change of the instantaneous power of the motor and the process state index at the sampling time between adjacent sampling times, the energy marginal utility index is calculated, including: ； Among them, Eu(t) i (t) represents the sampling time. i The marginal utility index of energy, P(t) i (t) represents the sampling time. i Instantaneous power of the motor under the following conditions, t i-1 Sampling time t i The previous sampling time, t i -t i-1 Let S(t) be the sampling period Δt. i-1 The current process status index corresponding to the previous sampling time.

2. The adaptive optimization control method for energy consumption of a mixer according to claim 1, characterized in that, Constructing an energy state phase diagram includes: Define a two-dimensional coordinate system, with the horizontal axis representing the instantaneous power of the motor and the vertical axis representing the energy marginal utility index; At each sampling moment, the collected instantaneous power of the motor and the corresponding energy marginal utility index are mapped to a real-time state point in the coordinate system; As the mixing process progresses, the real-time state points are continuously accumulated to form a dynamically evolving energy state phase diagram, which is used to characterize the evolution trajectory of the energy efficiency state.

3. The adaptive optimization control method for energy consumption of an internal mixer according to claim 2, characterized in that, Based on the real-time state points obtained up to the sampling time, its consistency index is calculated, including: The learning phase begins at the start of the sampling period, and the following loop is executed at each sampling time until the learning phase ends: (a) Up to the sampling time, record the most recent M consecutive real-time state points to form a sliding state sequence {Q1, Q2, ..., Q...} M }, M≥3, the time period covered by the sliding state sequence constitutes a local time window; (b) Based on the sliding state sequence {Q1, Q2, ..., Q...} M }, calculate the displacement vector between adjacent real-time state points within it, and normalize each displacement vector to obtain multiple unit vectors; (c) Calculate the magnitude of the average vector of the unit vector as the consistency index C, where 0≤C≤1, and store the calculated consistency index C in the pre-constructed consistency index sequence in chronological order. (d) In the consistency index sequence, take the most recent K consecutive consistency indices to form a decision window, and count the number of indices that satisfy C≥ preset consistency threshold. The value of K should be such that the total time covered by the decision window is greater than twice the local time window. (e) If the ratio of the number to K is greater than or equal to the preset confidence ratio, it is determined that the mixing process has entered a regular evolution state. Then, the sampling time is marked as the last sampling time of the learning phase, and the learning phase ends. After the learning phase is completed, the energy efficiency reference function is obtained by fitting a quadratic polynomial regression method.

4. The adaptive optimization control method for energy consumption of an internal mixer according to claim 3, characterized in that, The optimal combination of rotor speed and upper jack pressure is determined by commands including: Once the mixing process is determined to have entered a regular evolutionary state, in response to the last sampling moment, the dual-layer optimization controller is activated, and the outer controller executes the following steps: At each sampling time after the end of the learning phase, the real-time state point in the energy state phase diagram at that sampling time is obtained. The real-time state point includes the instantaneous power of the motor and the energy marginal utility index at that sampling time. Based on the instantaneous power of the motor at the sampling moment, the characteristic trajectory of the energy efficiency reference function is obtained. The reference point on the trajectory with the shortest Euclidean distance to the real-time state point at the sampling moment is searched, and the power value at the reference point is recorded as the trajectory reference power. Based on the trajectory reference power, calculate the energy efficiency deviation between it and the instantaneous power of the motor at the sampling moment; A target power is generated based on the energy efficiency deviation. The target power represents the magnitude of adjustment towards the trajectory reference power direction. The adjustment magnitude is determined by a preset gain coefficient, which has a value range of (0, 1]. After the learning phase is completed, the inner loop controller constructs a real-time, high-quality data window, including the following steps: At each sampling time after the learning phase ends, the instantaneous power of the motor, the material temperature in the mixing chamber, the rotor speed, and the pressure of the top plug corresponding to that sampling time will only be recorded as a valid entry if the following conditions are met simultaneously: (1) The Euclidean distance calculated by the outer controller is less than the preset deviation threshold; (2) The absolute value of the material temperature change rate in the mixing chamber calculated based on adjacent sampling times is less than the preset temperature change rate threshold. The valid entries are stored in a real-time high-quality data window in chronological order, wherein the data window retains all valid entries within the most recent γ time period, where γ is the preset time window length; In response to the instantaneous target power, one or more candidate entries that satisfy the energy efficiency deviation being within a preset tolerance range are retrieved from the data window; If there are candidate entries, the entry with the highest energy marginal utility index is selected, and the rotor speed and upper bolt pressure recorded in that entry are output as the optimal combination command. If no candidate entry is found, maintain the current rotor speed and the pressure of the top bolt.

5. The adaptive optimization control method for energy consumption of an internal mixer according to claim 4, characterized in that, Also includes: After a mixing operation is completed, cluster analysis is performed on all valid entries in the data window to extract one or more steady-state operation modes that represent the optimal efficiency of this mixing operation. These modes are then associated with the process identifier of this mixing operation and stored in a cross-batch global experience base to provide initial operation parameter recommendations for subsequent mixing operations with the same process identifier.

6. An adaptive optimization control system for energy consumption of an internal mixer, characterized in that, The system executes the adaptive optimization control method for energy consumption of the internal mixer according to any one of claims 1 to 5, including: The multi-source synchronous sampling module is used to collect the instantaneous power of the motor, the temperature of the material in the mixing chamber, the rotor speed and the pressure of the top plug at each sampling moment, with the moment when the top plug is pressed down as the sampling start time. The energy marginal calculation module is used to calculate the process state index based on the material temperature in the mixing chamber at each sampling time, and to calculate the energy marginal utility index based on the rate of change between the instantaneous power of the motor and the process state index at adjacent sampling times. The process state index is a normalized process progress index, which represents the degree to which the mixing state approaches the preset target at the sampling time. The energy marginal utility index is used to quantify the process state progress caused by a unit energy input, so as to characterize the energy utilization efficiency of the mixing process at the sampling time. A phase diagram construction module is used to construct an energy state phase diagram based on the instantaneous power of the motor and the energy marginal utility index, and map the two to real-time state points in the energy state phase diagram; The energy state feature learning module is used to enter the learning phase from the sampling start time. At each sampling time, the following operations are performed until the end of the learning phase: Based on the real-time state points obtained up to the sampling time, its consistency index is calculated, and based on the number of consistency indices that are not less than a preset consistency threshold, it is determined that the system has entered a regular evolution state. Then, the sampling time is marked as the last sampling time of the learning phase, and the learning phase ends. The real-time state points corresponding to the sampling times in the learning phase where the consistency index is not less than the preset consistency threshold are extracted, and an energy efficiency reference function is constructed with the instantaneous power of the motor as the independent variable and the energy marginal utility index as the dependent variable. The regular evolution state refers to the fact that the material in the mixing chamber has completed the main dispersion and its mixing state has entered a stable operating stage. A dual-layer energy efficiency optimization control module is used to respond to the last sampling moment of the learning phase by starting the dual-layer optimization controller. The outer controller obtains the real-time state point corresponding to each sampling moment after the end of the learning phase. Based on the instantaneous power of the motor in the state point, it obtains the feature trajectory corresponding to the energy efficiency reference function. Then, it searches for the reference point with the shortest Euclidean distance to the state point on the trajectory and records the power value corresponding to the reference point as the trajectory reference power. Based on the trajectory reference power, a target power is generated. The inner loop controller responds to the target power to maximize the current energy marginal utility index and solves the optimal combination command of rotor speed and top bolt pressure. The optimal instruction execution module is used to input the optimal combination of instructions to the actuator of the internal mixer to perform real-time control of the mixing process, optimize the marginal utility of energy, and achieve adaptive optimization of energy consumption in this mixing process.

7. A computer-readable storage medium, characterized in that, The storage medium stores one or more programs, which can be executed by one or more processors to implement the steps in the adaptive optimization control method for energy consumption of the internal mixer as described in any one of claims 1-5.

Images