CN-122018603-A - High-precision online heating control method and system based on dynamic model prediction and multi-mode feedback

CN122018603ACN 122018603 ACN122018603 ACN 122018603ACN-122018603-A

Abstract

The invention discloses a high-precision online heating control method and a system based on dynamic model prediction and multi-mode feedback, which belong to the technical field of industrial precise heating control; the single sensor error and random interference are eliminated through multi-source sensor data fusion and environment interference dynamic compensation, and the control precision is improved to be within +/-0.1 ℃; the method and the device can be widely applied to industrial scenes with extremely high requirements on temperature precision and uniformity, such as semiconductor manufacture, precise die heat treatment and the like, and have high response speed, ultrahigh control precision and strong universality.

Inventors

XU YONGCHENG
LI HAODONG
ZHU YUANXIAN
LI HUI
LI ZHIHENG
ZHOU YUXING
PEI HONGWEI
ZHOU JINGYI

Assignees

上海神众智能科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (10)

1. A high-precision online heating control method based on dynamic model prediction and multi-mode feedback is characterized by comprising the following steps: s1, multi-mode sensing data acquisition and preprocessing, namely acquiring multi-dimensional temperature data of a heating object, running state data of heating equipment and environmental parameter data of a heating cavity in real time, and carrying out filtering denoising, time sequence alignment and outlier rejection preprocessing on the acquired original data to obtain a standardized sensing data set; S2, constructing a dynamic thermodynamic prediction model updated on line, namely constructing a lumped parameter dynamic thermodynamic model based on an energy conservation law and the heat transfer characteristic of a heating system and combining the preprocessed perception data, correcting the thermophysical parameters and heat loss coefficients of the model on line through the perception data acquired in real time, and predicting the temperature change trend of a heating object in N control periods in the future based on the corrected dynamic thermodynamic model; S3, generating a heating power pre-control instruction based on model prediction, namely solving a rolling optimization objective function with constraint in each control period by taking a preset target temperature curve as a control target and taking a predicted temperature change trend as a basis to obtain an optimal heating power pre-control instruction and realize advanced pre-control of temperature; S4, multi-mode feedback fusion and real-time compensation correction, namely, carrying out self-adaptive weighted fusion on multi-dimensional temperature data and running state data to obtain a high-precision actual temperature feedback value, constructing an environment interference compensation model, calculating real-time interference compensation quantity based on environment parameter data, and carrying out closed-loop correction on a heating power pre-control instruction by combining the deviation of the temperature feedback value and a target temperature to generate a final heating power control instruction; And S5, self-adapting online optimization of control parameters based on reinforcement learning, namely constructing a parameter optimization reinforcement learning intelligent body, taking temperature control deviation, overshoot and control stability as core rewarding indexes, taking dynamic thermodynamic model parameters, rolling optimization weight coefficients and environment compensation gain coefficients as optimization objects, carrying out online iterative optimization on full-flow control parameters, and adapting to the thermal characteristic differences of different heating objects.
2. The high-precision online heating control method based on dynamic model prediction and multi-mode feedback of claim 1 is characterized in that in S1, multi-dimensional temperature data comprise at least two paths of contact temperature measurement data and at least one path of non-contact temperature measurement data, wherein the contact temperature measurement data are temperature data collected by thermocouples and platinum resistors, the non-contact temperature measurement data are temperature data collected by infrared temperature sensors and thermal infrared imagers, the running state data comprise real-time output power, input voltage, input current and on-off state of a heating module, and the environment parameter data comprise airflow speed, environment humidity, environment temperature and cavity wall surface temperature in a heating cavity.
3. The high-precision online heating control method based on dynamic model prediction and multi-mode feedback of claim 1, wherein in S2, the construction and update of the dynamic thermodynamic model specifically comprises: s21, constructing an initial lumped parameter thermodynamic model based on the law of conservation of energy: wherein C is the total heat capacity of the heating object and the heating platform, T (T) is the real-time temperature of the heating object, eta is the heating power-heat energy conversion efficiency, pt is the heating power, h is the convection heat exchange coefficient, A is the heat exchange area, Epsilon is the emissivity of radiation and sigma is the steven-boltzmann constant for ambient temperature; S22, based on the preprocessed real-time perception data, adopting a recursive least square method to identify heat capacity C, conversion efficiency eta and convection heat transfer coefficient h in the correction model on line, and obtaining a real-time updated dynamic thermodynamic model; S23, based on a dynamic thermodynamic model, predicting a temperature sequence of N control periods in the future by taking the state of the current control period as an initial value, wherein the value range of N is 5-50, and the value range of the control period is 1ms-100ms.
4. The high-precision online heating control method based on dynamic model prediction and multi-mode feedback of claim 1, wherein in S3, the rolling optimization objective function is: The constraint conditions are as follows: Wherein, the For the predicted temperature of the i-th control period, For the target temperature of the ith control period, For the amount of power variation of adjacent control periods, As the weight coefficient of the light-emitting diode, For the upper and lower limits of the heating power, Is the maximum allowable temperature change rate.
5. The high-precision on-line heating control method based on dynamic model prediction and multi-mode feedback of claim 1, wherein in S4, the specific method of self-adaptive weighted fusion is as follows: for the measurement data of M paths of temperature measuring sensors, respectively calculating the measurement variance of each sensor in a sliding time window Self-adaptive distribution of weight of each sensor according to variance : The final fusion temperature values were: Wherein, the Is a preprocessed temperature measurement value of the ith sensor.
6. The high-precision online heating control method based on dynamic model prediction and multi-mode feedback of claim 1, wherein in S4, the environmental disturbance compensation model comprises an airflow heat loss compensation module, a power supply fluctuation compensation module and a sensor temperature drift compensation module; The airflow heat loss compensation module calculates the variable quantity of the convection heat exchange coefficient based on the airflow speed acquired in real time to obtain the convection heat loss compensation power; the power supply fluctuation compensation module calculates the output deviation of the heating power based on the deviation of the input voltage and the rated voltage acquired in real time to obtain the power supply fluctuation compensation quantity; the sensor temperature drift compensation module carries out drift correction on the measured value of the temperature measuring sensor based on the difference value between the ambient temperature and the calibration temperature, and eliminates the system error caused by the ambient temperature.
7. The high-precision online heating control method based on dynamic model prediction and multi-mode feedback of claim 1, wherein in S5, the reinforcement learning agent adopts a near-end strategy to optimize PPO algorithm, and a state space comprises current temperature deviation, temperature change rate, model prediction error, environmental parameters and heating object thermal characteristic parameters; The reward function is: Wherein, the For the deviation of the current fusion temperature from the target temperature, As a variance of the temperature deviation within the sliding window, As the amount of change in the power control command, Is a weight coefficient when In the time-course of which the first and second contact surfaces, A positive prize value, otherwise 0.
8. The method for high-precision online heating control based on dynamic model prediction and multi-mode feedback of claim 1, wherein S4 further comprises a sensor fault diagnosis and fault tolerance step of monitoring the deviation of each sensor measured value and the fusion temperature value in real time, judging that a sensor fails when the deviation of a certain sensor continuously exceeds a preset fault threshold value, automatically eliminating fault sensor data, reassigning the fusion weight of the rest sensors, and ensuring continuous and stable operation of a control closed loop.
9. The method for high-precision online heating control based on dynamic model prediction and multi-mode feedback according to claim 1, wherein the method further comprises a starting pre-calibration step, namely after starting, executing a preset heating-insulating-cooling calibration process, collecting full-temperature-range thermal response data of a heating object, identifying to obtain initial thermodynamic model parameters and initial control parameter values, and entering an online control mode after the pre-calibration is completed.
10. A high-precision online heating control system based on dynamic model prediction and multi-mode feedback is characterized by comprising the following components: the multi-mode sensing unit is used for collecting multi-dimensional temperature data of a heating object, running state data of heating equipment and environmental parameter data of a heating cavity in real time and outputting a preprocessed standardized sensing data set; The dynamic model prediction unit is in communication connection with the multi-mode sensing unit and is used for constructing and online updating a dynamic thermodynamic model and predicting the temperature change trend of the heating object in N control periods in the future; the power pre-control generation unit is in communication connection with the dynamic model prediction unit and is used for solving a rolling optimization objective function based on the predicted temperature trend and the target temperature curve to generate an optimal heating power pre-control instruction; The multi-mode feedback compensation unit is respectively in communication connection with the multi-mode sensing unit and the power pre-control generation unit and is used for fusing multi-source data to obtain a high-precision temperature feedback value, calculating the real-time compensation quantity of environmental interference, and performing closed-loop correction on the power pre-control instruction to generate a final heating power control instruction; The self-adaptive parameter optimization unit is respectively in communication connection with the dynamic model prediction unit, the power pre-control generation unit and the multi-mode feedback compensation unit, is internally provided with a reinforcement learning intelligent body, is used for on-line iterative optimization of the whole-flow control parameters and is used for self-adapting to the thermal characteristics of different heating objects; And the heating execution unit is in communication connection with the multi-mode feedback compensation unit and is used for receiving the heating power control instruction and outputting corresponding power to the heating module to complete heating control.

Description

High-precision online heating control method and system based on dynamic model prediction and multi-mode feedback Technical Field The invention relates to the technical field of semiconductor manufacturing, in particular to a high-precision on-line heating control method and system based on dynamic model prediction and multi-mode feedback. Background In the field of semiconductor manufacturing and precise die heat treatment, the temperature uniformity and the temperature control precision in the heating process directly determine the molding quality and the performance stability of products, are key links for limiting the high-end manufacturing yield and the core process landing, and particularly aim at high-precision workpieces such as semiconductor photoetching dies, wafer carriers, micro structural parts and the like, and small temperature fluctuation and temperature difference can lead to deformation, dimension out-of-tolerance and performance attenuation of the workpieces, so that severe standards of chip manufacturing process and precise molding are difficult to be met. In the existing industrial heating control scheme, most of the conventional PID control algorithm is adopted, feedback adjustment is carried out based on the deviation between the current temperature and the target temperature, and the system has the advantages of simple principle and easiness in realization, but the system has the inherent defects that firstly, the heating system belongs to a typical large inertia and large hysteresis system, PID control belongs to post error adjustment, the temperature change trend cannot be prejudged in advance, temperature overshoot and oscillation are extremely easy to occur, the ultra-high precision control requirement of +/-0.1 ℃ is difficult to meet, secondly, the control parameters of the conventional PID are fixed, when a heating object is replaced and the environmental parameters are dynamically changed, the control performance is greatly reduced, professional personnel are required to manually reset the parameters, the suitability is poor, and the requirements of various flexible production cannot be met. In order to solve the defects of PID control, a model prediction control scheme is introduced in part of the prior art, but a thermodynamic model with fixed parameters is mostly adopted, the real-time change of the thermal characteristics of a heating object cannot be adapted on line, the influence of dynamic environment interference such as airflow, humidity and power supply fluctuation is not considered, the model prediction precision is insufficient, the control precision cannot break through, meanwhile, the prior art mostly adopts a single thermocouple temperature measurement scheme, the problems that the whole temperature of the heating object cannot be reflected by temperature measurement delay and single-point measurement is easy to generate temperature drift due to environmental interference exist, the upper limit of the control precision is directly limited by the measurement error of a single sensor, the control precision of +/-0.1 ℃ cannot be stably realized, and in addition, the compensation of the environmental interference in the prior art is mostly the compensation of the fixed parameters, the real-time change interference factor cannot be adapted dynamically, the compensation effect is poor, and the complex industrial field environment cannot be dealt with. Therefore, developing an ultra-high precision online heating control scheme capable of fundamentally solving the problem of hysteresis of a heating system, eliminating multi-source measurement errors, dynamically adapting to environmental interference and heating object characteristics becomes a technical problem to be solved in the field. Disclosure of Invention Technical problem to be solved Aiming at the defects of the prior art, the invention aims to provide a high-precision online heating control method and a high-precision online heating control system based on dynamic model prediction and multi-mode feedback, which solve the technical problems of hysteresis overshoot, insufficient measurement precision, poor environment interference adaptability and weak universality of heating objects of the traditional control scheme, stably realize ultra-high precision temperature control within +/-0.1 ℃, have strong self-adaption capability and high reliability, and meet the severe control requirements of high-end industrial scenes. Technical proposal The high-precision online heating control method based on dynamic model prediction and multi-mode feedback comprises the following steps: S1, multi-mode sensing data acquisition and preprocessing, namely acquiring multi-dimensional temperature data of a heating object, running state data of heating equipment and environmental parameter data of a heating cavity in real time, and carrying out filtering denoising, time sequence alignment and outlier rejection preprocessing on the ac