A novel temperature control system based on model predictive control

Abstract

The application discloses a novel temperature control system based on model predictive control, which comprises a semiconductor refrigeration component and a sensing control component. A predictive control module is integrated in the single-chip microcomputer controller, and is cyclically executed through a prediction step, a rolling optimization step and a feedback correction step in each control cycle by establishing a thermal inertia physical model of the refrigeration sheet. The application utilizes the physical model to preview a future temperature evolution track, obtains an optimal control sequence by solving a quadratic programming problem, and introduces a control increment constraint in a target function to eliminate current impact in cooperation with a hardware LC filter circuit. The system has a predictive control capability, can compensate for power fluctuation and environmental disturbance, realizes zero overshoot and high-precision temperature control, and significantly improves the physical life of the semiconductor refrigeration sheet.

Description

Technical Field

[0001] This application relates to the field of temperature control technology, and more specifically, to a novel temperature control system based on model predictive control. Background Technology

[0002] With the continuous advancement of high-precision optoelectronic devices, precision chemical reaction laboratories, and high-performance sensor technologies, increasingly stringent requirements have been placed on the temperature control accuracy and dynamic response speed of thermoelectric coolers (TECs). As a solid-state refrigeration technology, thermoelectric coolers possess inherent advantages such as no refrigerant required, rapid response, and flexible switching between hot and cold temperatures, demonstrating broad application potential in precision temperature control scenarios. However, in practical engineering, achieving high-precision temperature control still encounters multiple technical bottlenecks. The primary problem stems from the inherent thermal inertia and hysteresis characteristics of thermoelectric cooling systems. Heat conduction and heat capacity effects cause significant delays in the system during temperature changes, making it difficult for traditional PID control algorithms to reconcile the contradiction between response speed and overshoot. Especially in the critical region of the target temperature, the cumulative effect of thermal inertia can easily induce continuous system oscillations, causing temperature fluctuations to exceed the allowable threshold, severely limiting the potential for improving temperature control accuracy.

[0003] Existing four-quadrant temperature control methods generally rely on hard-switching logic. While this logic can cover both cooling and heating operating conditions, at the boundary points of quadrant switching, control variables (such as duty cycle) often undergo a step-like change. Such abrupt changes not only subject the H-bridge drive circuit to a large instantaneous current surge, causing power supply Vcc voltage fluctuations and electromagnetic interference, but may also accelerate the fatigue of the semiconductor cooling chip material, significantly shortening its physical lifespan.

[0004] Furthermore, the actual operating environment is highly uncertain. Random fluctuations in power supply voltage, dynamic evolution of environmental heat dissipation conditions, and unpredictable disturbances in thermal load can all lead to continuous deviations between the preset control model and the actual system characteristics. Traditional feedback control mechanisms only make ex-post adjustments based on historical errors, lacking the ability to predict the future behavior of the system and failing to actively compensate for the impact of dynamic disturbances, resulting in a significant degradation in control performance under complex operating conditions.

[0005] Furthermore, the thermal efficiency of semiconductor refrigeration chips exhibits an inherent asymmetry between cooling and heating modes. This stems from the difference in the coupling effect between the Seebeck effect and Joule heating. A single linear control law is insufficient to adapt to the nonlinear characteristics across the entire temperature range, resulting in the system being unable to maintain consistent optimal performance under different operating conditions.

[0006] In summary, existing temperature control solutions have significant shortcomings in core aspects such as dynamic compensation for thermal inertia, long-term hardware protection mechanisms, environmental disturbance suppression capabilities, and smooth transitions across multiple quadrants. There is an urgent need for an innovative temperature control system that possesses predictive capabilities for system behavior, robust environmental adaptability, and hardware lifespan protection. Summary of the Invention

[0007] The purpose of this application is to provide a novel temperature control system based on model predictive control, which has the advantages of being able to offset the thermal inertia of semiconductor refrigeration systems, balancing temperature control accuracy and dynamic response speed, suppressing environmental disturbances, and improving temperature control stability.

[0008] The above-mentioned technical objective of this application is achieved through the following technical solution:

[0009] A novel temperature control system based on model predictive control includes:

[0010] Semiconductor cooling components: including semiconductor cooling chips, H-bridge drive circuits, high-voltage follower circuits, and LC filter circuits;

[0011] Sensing and control components: including a microcontroller, temperature sampling circuit, and power supply;

[0012] The feature is that the microcontroller controller integrates a predictive control module; the predictive control module is based on the established thermal inertia physical model of the cooling chip, and solves the optimal control sequence in each control cycle by collecting real-time data from the temperature sampling circuit, and outputs it to the H-bridge drive circuit.

[0013] Furthermore, the predictive control module cyclically executes the following steps within each control cycle to solve for the optimal control quantity in real time:

[0014] Prediction steps: Using the thermal inertia physical model of the cooling chip, starting from the corrected state estimate at the current moment, predict the temperature evolution trajectory for multiple future control cycles;

[0015] Rolling optimization steps: By solving a quadratic programming problem, find an optimal control sequence that minimizes the deviation between the temperature evolution trajectory and the set target temperature, while also minimizing the amplitude of the H-bridge drive circuit.

[0016] Feedback correction step: Data is read in real time through temperature sampling circuit, the residual between the actual temperature and the model prediction value after the execution of the control sequence is calculated, and the state estimate is updated accordingly to compensate for the impact of environmental disturbances on the prediction. The updated state estimate is used as the input benchmark for the prediction step of the next control cycle.

[0017] Furthermore, the thermal inertia physical model of the cooling chip adopts the following discretized state-space mathematical expression:

[0018]

[0019] Where x(k) is the state vector of the system at the k-th sampling time, including the temperature state quantity and the rate of temperature change;

[0020] u(k) is the control quantity applied to the H-bridge drive circuit at the k-th sampling time, and its value corresponds to the duty cycle value of the control sequence.

[0021] A is the system state matrix, which is composed of thermal inertia parameters determined by the specific heat capacity, mass, and heat dissipation coefficient of the cooling element;

[0022] B is the input matrix, which consists of the electrothermal conversion efficiency of the thermoelectric cooler and the voltage gain parameters of the high-voltage follower circuit.

[0023] Furthermore, in the prediction step, the mathematical expression for transforming the state estimate into a temperature evolution trajectory is as follows:

[0024]

[0025] in, Let $k$ be the state quantity predicted at time $k+i$ in the future at time $k$.

[0026] Let be the i-th power of the system state matrix A, representing the degree of influence of the system's inherent characteristics on the future state after i periods;

[0027] This is the current state estimate, output by the feedback correction step;

[0028] The combined gain represents the residual effect of the action performed at time j on the target time i after the remaining time has elapsed.

[0029] Let be the control sequence, representing the control signal planned to be output at time k+j in the future.

[0030] Furthermore, in the optimization step, the optimal control sequence is obtained by minimizing the following objective function. :

[0031]

[0032] In the formula, the first term is the error penalty term, and the second term is the control constraint term;

[0033] Let $k$ be the state quantity predicted at time $k+i$ in the future at time $k$.

[0034] The target temperature at time k+i in the future;

[0035] P is the prediction time domain, and Q is the state weighting matrix;

[0036] It is the weighted square norm;

[0037] To control the increment, it represents the change in duty cycle between adjacent sampling times;

[0038] M represents the control time domain; R is the control weighting matrix, used to limit the control increment to work in conjunction with the LC filter circuit to suppress the thermal shock to the cooling chip.

[0039] It is the weighted square norm.

[0040] Furthermore, the feedback correction step adopts the following state correction formula:

[0041]

[0042] in, This is the current state estimate; The current time-priority predicted state is calculated by the prediction and control module of the previous cycle; L is the observation gain matrix, used to adjust the correction weights of the residuals. C represents the real-time sampled values, and C is the measurement matrix. This represents the residual between the actual temperature and the model prediction.

[0043] Furthermore, the predictive control module includes the following in its logical architecture:

[0044] State observation unit: Based on the real-time data from the temperature sampling circuit and the predicted data from the previous cycle, the state observer algorithm is used to eliminate environmental disturbance bias and output the corrected state estimate for the current moment.

[0045] Multi-step prediction unit: used to calculate the temperature evolution state at multiple future sampling points in the time domain, starting from the corrected state estimate and using a preset thermal inertia physical model of the cooling chip;

[0046] Rolling optimization unit: It is used to establish a cost function based on the preset temperature control accuracy target and hardware action constraints, and generate the optimal control sequence command by solving the quadratic programming problem online;

[0047] Execution drive unit: used to convert the first item of the optimal control sequence into the corresponding PWM signal pulse width and polarity control signal, and send it to the H-bridge drive circuit.

[0048] In summary, this application has the following beneficial effects:

[0049] By integrating a predictive control module into the microcontroller, the temperature change trend is predicted in advance based on the thermal inertia physical model of the thermoelectric cooler, and the optimal control output is solved. This can actively counteract the thermal inertia of the system, solving the problem of difficulty in balancing response speed and overshoot in traditional temperature control schemes. At the same time, it can compensate for control deviations caused by environmental disturbances, effectively improving temperature control accuracy and system operation stability. It has the advantages of being able to counteract the thermal inertia of the semiconductor refrigeration system, balancing temperature control accuracy and dynamic response speed, suppressing environmental disturbances, and improving temperature control stability. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the novel temperature control system based on model predictive control described in this application.

[0051] Figure 2 This is the control logic diagram of the predictive control module described in this application. Detailed Implementation

[0052] To make the technical means, creative features, objectives and effects of this application easier to understand, the following description, in conjunction with illustrations and specific embodiments, further elaborates on this application.

[0053] like Figure 1 As shown, this application proposes a novel temperature control system based on model predictive control. This system integrates a predictive control module within a microcontroller controller through the collaborative operation of a semiconductor refrigeration component and sensing and control components. Based on an established thermal inertia physical model of the refrigeration element, this module acquires real-time data from a temperature sampling circuit, solves for the optimal control sequence in each control cycle, and outputs the result to an H-bridge drive circuit. This effectively solves problems in existing technologies such as insufficient thermal inertia compensation, inadequate hardware protection, weak anti-disturbance capability, and unsmooth dynamic operating condition switching.

[0054] For ease of understanding, the following explains some key terms in this embodiment:

[0055] Semiconductor cooling module: This module is the execution part of the temperature control system, mainly responsible for realizing the cooling or heating function. It typically includes a semiconductor cooling chip, an H-bridge drive circuit, a high-voltage follower circuit, and an LC filter circuit. The semiconductor cooling chip is the core execution element, achieving temperature regulation through the Peltier effect; the H-bridge drive circuit controls the direction and magnitude of the current to the semiconductor cooling chip to achieve switching between cooling or heating modes and power regulation; the high-voltage follower circuit stabilizes and amplifies the drive voltage to ensure that the semiconductor cooling chip receives sufficient driving capability; the LC filter circuit smooths the pulse signal output by the H-bridge drive circuit, reducing thermal shock to the semiconductor cooling chip.

[0056] Sensing and Control Components: This component is the brain of the temperature control system, responsible for acquiring temperature data, running control algorithms, and outputting control signals. It typically includes a microcontroller controller, a temperature sampling circuit, and a power supply. The microcontroller controller is the core processing unit, carrying and executing the control algorithm; the temperature sampling circuit monitors the temperature of the controlled object in real time and converts analog signals into digital signals for processing by the microcontroller controller; the power supply provides stable operating power to the entire system.

[0057] Predictive Control Module: This module is a software or hardware logic unit integrated within the microcontroller controller. Its core idea is to use a system model to predict future states and optimize the control strategy accordingly. Through prediction, the system can anticipate potential temperature changes and disturbances, thereby achieving smoother and more precise temperature control.

[0058] Thermal inertia physical model of a thermoelectric cooler: This model is a mathematical description of the dynamic characteristics of a thermoelectric cooler and its related thermal system. It reflects the hysteresis and inertial characteristics of the cooler's temperature response after receiving electrical energy input. This model is usually based on thermodynamic principles and calibrated with experimental data to predict the temperature change trend of the cooler under different control inputs.

[0059] Temperature sampling circuit: This circuit is used to acquire the current temperature information of the controlled object in real time. It typically consists of a temperature sensor (such as a thermistor, thermocouple, or platinum resistance thermometer) and a signal conditioning circuit (such as an amplifier, filter, and analog-to-digital converter) to convert the physical temperature quantity into an electrical or digital signal that can be processed by the microcontroller controller.

[0060] Optimal control sequence: This sequence is a series of control commands calculated by the predictive control module in each control cycle. This sequence is obtained through an optimization algorithm, aiming to make the system temperature as close as possible to the target temperature over a future period, while also satisfying constraints on other performance indicators (such as energy consumption and hardware impact).

[0061] H-bridge driver circuit: This circuit is a commonly used motor drive or power conversion circuit, and in this system it is used to drive a thermoelectric cooler. By controlling the on / off state of the four switching transistors in the H-bridge, the direction and magnitude of the current flowing through the thermoelectric cooler can be changed, thereby achieving the switching between cooling and heating modes and the adjustment of output power.

[0062] This application provides a novel temperature control system based on model predictive control, the specific implementation of which can be as follows:

[0063] The system comprises a thermoelectric cooling component and a sensing and control component. The thermoelectric cooling component can be composed of a thermoelectric cooler, an H-bridge driver circuit, a high-voltage follower circuit, and an LC filter circuit. The thermoelectric cooler, as the core actuator, raises or lowers the temperature through electrical input. The H-bridge driver circuit can employ a common four-switch topology to control the direction of current flowing through the thermoelectric cooler, thus switching between cooling and heating modes. The high-voltage follower circuit can be implemented using an operational amplifier or a dedicated driver chip to provide a stable drive voltage, ensuring the H-bridge driver circuit can effectively drive the thermoelectric cooler. The LC filter circuit, composed of inductors and capacitors, filters out high-frequency harmonics in the pulse signal output by the H-bridge driver circuit, providing a relatively smooth current and reducing thermal shock to the thermoelectric cooler. The sensing and control component can consist of a microcontroller controller, a temperature sampling circuit, and a power supply. The microcontroller controller can be a commercially available microcontroller, serving as the system's computation and control core. The temperature sampling circuit can be implemented using a thermistor in conjunction with an analog-to-digital converter (ADC) to convert analog temperature signals into digital signals. The power supply provides the required DC voltage for the entire system.

[0064] The microcontroller integrates a predictive control module. This predictive control module can be implemented as a software program running on the microcontroller, and its function is to calculate future control commands based on the current system state and the target temperature. For example, this module can contain a series of preset control rules or a simple proportional-integral-derivative (PID) controller to generate control outputs based on temperature deviations.

[0065] The predictive control module operates based on an established thermal inertial physical model of the thermoelectric cooler. This model can be a simplified linear model, such as a first-order inertial element model, used to describe the dynamic response of the thermoelectric cooler's temperature over time after receiving electrical energy input. This model can be obtained by experimentally testing the thermoelectric cooler and fitting its temperature response curve.

[0066] The predictive control module acquires real-time data from a temperature sampling circuit. This circuit periodically reads the temperature of the current environment or the controlled object, for example, every 100 milliseconds. This real-time data is input into the predictive control module as feedback information on the current system state.

[0067] Within each control cycle, the predictive control module solves for the optimal control sequence. This solution process can be a simple lookup table matching process, selecting a corresponding control sequence from a pre-defined control strategy table based on the deviation between the current temperature and the target temperature. Alternatively, a simple iterative algorithm, such as gradient descent, can be used to find a control sequence that minimizes the temperature deviation within a finite search space.

[0068] The optimal control sequence obtained is then output to the H-bridge driver circuit. For example, if a positive duty cycle sequence is obtained, the microcontroller will generate a corresponding pulse width modulation signal to drive the H-bridge circuit so that the current flows through the thermoelectric cooler in a specific direction, thereby achieving heating. Conversely, if a negative duty cycle sequence is obtained, the H-bridge circuit will be driven to reverse the current flow through the thermoelectric cooler, thereby achieving cooling.

[0069] The temperature control system of this application effectively overcomes the inherent thermal inertia and hysteresis of semiconductor refrigeration systems by introducing a predictive control strategy based on a physical model of the thermal inertia of the refrigeration chip. The system solves for the optimal control sequence in each control cycle, enabling prediction and early intervention of temperature evolution trends, thereby significantly reducing temperature overshoot and oscillations and improving temperature control accuracy. Simultaneously, the system avoids abrupt changes in control quantities caused by traditional hard switching, reducing instantaneous current surges to the H-bridge drive circuit and extending hardware lifespan. Furthermore, by combining real-time data feedback with model prediction, the system can effectively compensate for environmental disturbances, enhancing robustness under complex operating conditions.

[0070] In some of the solutions mentioned above in this application, a predictive control module is proposed to solve for the optimal control output based on the established thermal inertia physical model of the cooling chip and the collected real-time temperature data to achieve high-precision temperature control. However, in this process, the traditional method of solving the control output once cannot adapt to the dynamically changing temperature control state of the system. It cannot predict the temperature change trend in advance, nor can it correct the deviation caused by environmental disturbances to the model prediction in time. It is difficult to balance the temperature control response speed and temperature control accuracy, cannot offset the overshoot oscillation problem caused by thermal inertia, and cannot dynamically compensate for the model deviation caused by the changing environment, thus failing to meet the requirements of high-precision temperature control.

[0071] See Figure 2 This application further proposes that the predictive control module performs the following steps cyclically within each control cycle to solve for the optimal control quantity in real time: the prediction step uses the thermal inertia physical model of the cooling chip, starting from the corrected state estimate at the current moment, to predict the temperature evolution trajectory for multiple control cycles in the future; the rolling optimization step finds an optimal control sequence by solving a quadratic programming problem, so that the deviation between the temperature evolution trajectory and the set target temperature is minimized, and the action amplitude of the H-bridge drive circuit is minimized; the feedback correction step reads data in real time through the temperature sampling circuit, calculates the residual between the actual temperature after executing the control sequence and the model prediction value, and updates the state estimate value accordingly to compensate for the influence of environmental disturbances on the prediction, and uses the updated state estimate value as the input benchmark for the prediction step of the next control cycle.

[0072] The predictive control module executes cyclically within each control cycle to solve for the optimal control quantity in real time. This means that the control logic is not static or calculated only once, but is continuously re-evaluated and updated at preset time intervals (e.g., every 100 milliseconds or 1 second). This cyclic execution mechanism ensures that the system can continuously adapt to dynamically changing temperature control conditions, thereby achieving real-time optimization of the control quantity. For example, the periodic execution of the control algorithm can be triggered by a timer interrupt, or the consistency and real-time performance of the control cycle can be ensured by scheduling tasks through the operating system.

[0073] The prediction step is the core of model predictive control, which uses a thermoelectric thermal inertia physical model to simulate future temperature changes in the system. This step starts with a corrected state estimate at the current moment; for example, this state estimate may include the current temperature value and the rate of temperature change. Based on this, the temperature evolution trajectory over multiple control cycles is predicted, providing forward-looking information for subsequent control decisions. For example, iterative calculations can be performed using a discretized state-space model, or the evolution of a continuous-time model over discrete time steps can be simulated using numerical integration methods.

[0074] The rolling optimization step aims to determine the optimal control sequence by solving a quadratic programming problem based on the predicted temperature evolution trajectory. Quadratic programming is a mathematical optimization method applicable to problems with a quadratic objective function and linear constraints. In this step, the optimization objective is to find an optimal set of control sequences (e.g., the PWM duty cycle and polarity of the H-bridge drive circuit) that minimizes the deviation between the predicted temperature evolution trajectory and the set target temperature, while limiting the amplitude of the H-bridge drive circuit's operation to avoid sudden changes in control input that could impact the hardware. For example, standard quadratic programming algorithms such as the interior-point method or the effective set method can be used, or a QP solver specifically optimized for embedded systems can be employed.

[0075] The feedback correction step is crucial for ensuring the robustness of the control system. This step involves real-time reading of the actual temperature data via a temperature sampling circuit, comparing it with the model's predicted value, and calculating the residual between the two. Based on this residual, the system updates the current state estimate, compensating for the impact of environmental disturbances, model uncertainties, or unmodeled dynamics on the prediction results. The updated state estimate serves as the input benchmark for the prediction step in the next control cycle, ensuring that each prediction is based on the most accurate current system state.

[0076] Through the above technical solution, the predictive control module of this application can overcome the limitations of traditional single-solution control methods. The prediction, rolling optimization, and feedback correction steps, executed cyclically within each control cycle, form a dynamic and adaptive closed-loop control process. The prediction step enables the system to predict future temperature trends, allowing control actions to be planned in advance and effectively offsetting overshoot and oscillation problems caused by the thermal inertia of the thermoelectric cooler. The rolling optimization step, while ensuring temperature control accuracy, also considers the smoothness of the H-bridge drive circuit's operation, avoiding the impact of sudden changes in control quantities on the hardware and extending the service life of the thermoelectric cooler and drive circuit. The feedback correction step corrects the state estimate using real-time data, enabling the system to dynamically compensate for environmental disturbances and model biases, ensuring the accuracy of prediction and the robustness of control. This allows for high-precision, high-response temperature control even in complex and changing environments, meeting the stringent requirements of applications such as high-precision optoelectronic devices, precision chemical reaction laboratories, and high-performance sensors.

[0077] In some of the solutions mentioned above in this application, a thermal inertia physical model of the thermoelectric cooler is proposed to support the predictive control module in predicting the temperature evolution trajectory and solving the optimal control sequence. However, in this process, there is no accurate mathematical model that adapts to the discrete calculation requirements of predictive control and combines the characteristics of the semiconductor refrigeration system itself. The existing model cannot accurately reflect the thermal inertia characteristics of the semiconductor refrigeration system and the conversion relationship between control input and temperature change, which will lead to deviations in the prediction of future temperature changes by the predictive control module. It cannot effectively compensate for the overshoot problem caused by thermal inertia and the deviation caused by environmental disturbances, and it is difficult to meet the requirements of high-precision temperature control.

[0078] In this regard, this application further proposes that the thermal inertia physical model of the cooling chip adopts the following discretized state-space mathematical expression:

[0079]

[0080] Where x(k) is the state vector of the system at the k-th sampling time, including the temperature state quantity and the rate of temperature change;

[0081] u(k) is the control quantity applied to the H-bridge drive circuit at the k-th sampling time, and its value corresponds to the duty cycle value of the control sequence.

[0082] A is the system state matrix, which is composed of thermal inertia parameters determined by the specific heat capacity, mass, and heat dissipation coefficient of the cooling element;

[0083] B is the input matrix, which consists of the electrothermal conversion efficiency of the thermoelectric cooler and the voltage gain parameters of the high-voltage follower circuit.

[0084] Specifically, this discretized state-space mathematical expression is a standard method for describing dynamic systems in modern control theory, especially suitable for digital controllers. It represents the dynamic behavior of the system as a set of first-order difference equations, concisely describing the changes in the system state over time through matrix operations. Discretization allows the model to be precisely matched with the digital sampling and computation cycle, facilitating rolling prediction and optimization within each control cycle. This can be achieved through discretization methods of continuous-time models (e.g., zero-order hold discretization, Tustin transform, etc.) or directly through system identification based on discrete-time data.

[0085] The x(k) mentioned here serves as the system's state vector at the k-th sampling time, representing a complete description of the system at that moment and containing all the necessary information to predict its future behavior. Here, it includes not only the current temperature value (temperature state variable) but also the temperature change trend (temperature rate of change). This design provides more comprehensive system dynamic information, enabling the predictive control module to more accurately capture the rate of temperature rise or fall, thus more effectively addressing thermal inertia. For example, the temperature state variable can specifically refer to the actual temperature of the cooler surface, while the temperature rate of change can be the temperature gradient per unit time.

[0086] The variable u(k), applied to the H-bridge drive circuit at the k-th sampling time, is the system's input command to the H-bridge drive circuit and directly affects the cooling or heating power of the thermoelectric cooler. Defining it as the duty cycle value of the control sequence means that the control variable can be directly mapped to the pulse width and polarity of the PWM signal of the H-bridge drive circuit. For example, a positive duty cycle corresponds to the heating mode, and a negative duty cycle corresponds to the cooling mode, while the absolute value of the duty cycle determines the output power. This direct correspondence simplifies the conversion process of the control variable and improves control efficiency.

[0087] The matrix A, representing the system state, describes how the internal state of the system evolves over time when there is no external input. By incorporating thermal inertia parameters such as the specific heat capacity, mass, and heat dissipation coefficient of the thermoelectric cooler into the A matrix, the model accurately reflects the inherent thermal inertia characteristics of the thermoelectric cooler. For example, specific heat capacity and mass together determine the system's response speed to changes in heat, while the heat dissipation coefficient affects the efficiency of heat exchange between the system and the environment. This configuration allows the model to realistically simulate the hysteresis effects in the heat transfer and accumulation process.

[0088] The input matrix B describes how the external control input u(k) affects the system state x(k). By incorporating the electrothermal conversion efficiency of the thermoelectric cooler and the voltage gain parameter of the high-voltage follower circuit into the B matrix, the model can accurately reflect the conversion relationship from the control signal to the actual thermal power output. For example, the electrothermal conversion efficiency determines the proportion of electrical energy converted into heat or cold energy, while the voltage gain of the high-voltage follower circuit affects the magnitude of the voltage actually applied to the thermoelectric cooler by the H-bridge drive circuit. This configuration ensures that the model can accurately simulate the direct effect of the control quantity on temperature changes.

[0089] Through the above technical solution, this application constructs a physical model of the thermal inertia of the thermoelectric cooler using discretized state-space mathematical expressions, solving the problem that existing models cannot accurately reflect the thermal inertia characteristics of the semiconductor cooling system and the conversion relationship between control input and temperature change. Specifically, the system state vector x(k) is defined as including temperature state quantities and the rate of temperature change, enabling the model to comprehensively capture the current temperature level and trend of the system, providing more complete and accurate starting point information for predictive control. Simultaneously, the control quantity u(k) is directly mapped to the duty cycle value of the control sequence, simplifying the conversion between control commands and actual drive signals and improving control efficiency. Furthermore, by incorporating thermal inertia parameters such as the specific heat capacity, mass, and heat dissipation coefficient of the thermoelectric cooler into the system state matrix A, the model can realistically reflect the inherent thermal inertia characteristics of the semiconductor cooler and accurately simulate the hysteresis effect in the heat transfer and accumulation process. In addition, the electrothermal conversion efficiency of the thermoelectric cooler and the voltage gain parameters of the high-voltage follower circuit are incorporated into the input matrix B, ensuring that the model can accurately simulate the direct effect of the control quantity on temperature changes. This accurate mathematical model, adapted to the discrete computational requirements of predictive control and incorporating the inherent characteristics of the semiconductor refrigeration system, provides a solid foundation for temperature prediction and optimization solutions in the predictive control module. Therefore, the predictive control module can more accurately predict future temperature changes, effectively compensate for overshoot caused by thermal inertia, and improve robustness to environmental disturbances, thereby significantly enhancing the accuracy and stability of temperature control.

[0090] In some of the solutions described above in this application, a predictive step based on the thermal inertia physical model of the thermoelectric cooler is proposed to predict the temperature evolution trajectory for multiple future control cycles, in order to solve the problem that traditional temperature control schemes lack predictability and cannot compensate for thermal inertia and dynamic disturbances. However, in its implementation, it is not clear how to use the current corrected state estimate as a starting point and combine it with the established thermal inertia physical model to calculate the temperature evolution state at multiple future sampling times, nor can it accurately quantify the impact of control actions at each time on the future temperature state. Without an accurate method for calculating the evolution trajectory, it is impossible to provide a reliable predictive basis for subsequent rolling optimization, making it difficult to guarantee the accuracy of the optimal control quantity solution, and failing to achieve the temperature control objectives of suppressing overshoot and compensating for disturbances.

[0091] In this regard, this application further proposes a mathematical expression for transforming the state estimate into a temperature evolution trajectory in the prediction step:

[0092]

[0093] in, Let $k$ be the state quantity predicted at time $k+i$ in the future at time $k$.

[0094] Let be the i-th power of the system state matrix A, representing the degree of influence of the system's inherent characteristics on the future state after i periods;

[0095] This is the current state estimate, output by the feedback correction step;

[0096] The combined gain represents the residual effect of the action performed at time j on the target time i after the remaining time has elapsed.

[0097] Let be the control sequence, representing the control signal planned to be output at time k+j in the future.

[0098] The above technical solution clearly presents a calculation method for deriving the temperature evolution trajectory at multiple future moments from the current corrected state estimate. This provides an accurate predictive basis for subsequent rolling optimization to solve for the optimal control quantity, ensuring the accuracy of model predictive control in predicting temperature changes. Specifically, the expression uses the current state estimate output after the feedback correction step as the starting point for calculation. It relies on the corrected state value, which eliminates environmental disturbance bias, to carry out predictions. This effectively avoids carrying over early disturbance errors into the subsequent prediction process, improving the accuracy of the predicted trajectory from the starting point. Simultaneously, the i-th power of the system state matrix A is introduced to quantify the impact of the system's inherent thermal inertia characteristics on the future state after a corresponding number of control cycles. This accurately reflects the long-term effect of the inherent thermal inertia of the semiconductor refrigeration system on temperature changes, making the prediction results more consistent with the actual physical characteristics of semiconductor refrigeration temperature control. Furthermore, a combined gain is set to quantify the residual impact of the control actions already executed or planned for each moment on the target temperature state after the remaining time evolution. This clearly distinguishes the impact of the system's inherent characteristics and control actions on future temperature changes, making the prediction results more consistent with the actual temperature control process. By incorporating the control quantity sequence to correspond to the planned output control signals at various future moments, and fully integrating future temperature changes under different control schemes into the pre-simulation process, the system can provide accurate calculation basis for subsequently finding the optimal control sequence, ensuring the reliability of the subsequent optimal control quantity solution. Through this precise prediction of future temperature evolution trajectory, the system can anticipate temperature change trends in advance, effectively suppress temperature overshoot, and compensate for environmental disturbances, thereby achieving high-precision and high-stability temperature control.

[0099] In some of the solutions mentioned above in this application, a method is proposed to predict the future temperature state based on the thermal inertia physical model of the thermoelectric cooler, and then solve for the optimal control quantity to achieve high-precision temperature control. However, in this process, if the optimization objective is not set reasonably when solving for the optimal control sequence, it is impossible to simultaneously take into account the temperature control accuracy requirements and the hardware protection requirements. On the one hand, it is impossible to effectively constrain the deviation between the temperature and the target temperature, making it difficult to meet the requirements of high-precision temperature control. On the other hand, without reasonable constraints on the changes in the control quantity, sudden changes in the control quantity are likely to occur, which will generate thermal shock, affecting the service life of the thermoelectric cooler and the H-bridge drive circuit, and it is also impossible to cooperate with the LC filter circuit to suppress the shock. Therefore, it is necessary to reasonably construct the optimization objective to obtain the optimal control sequence.

[0100] In this regard, this application further proposes that, in the optimization step, the optimal control sequence is obtained by minimizing the following objective function. :

[0101]

[0102] In the formula, the first term is the error penalty term, and the second term is the control constraint term;

[0103] Let $k$ be the state quantity predicted at time $k+i$ in the future at time $k$.

[0104] The target temperature at time k+i in the future;

[0105] P is the prediction time domain, and Q is the state weighting matrix;

[0106] It is the weighted square norm;

[0107] To control the increment, it represents the change in duty cycle between adjacent sampling times;

[0108] M represents the control time domain; R is the control weighting matrix, used to limit the control increment to work in conjunction with the LC filter circuit to suppress the thermal shock to the cooling chip.

[0109] It is the weighted square norm.

[0110] In model predictive control (MMC), minimizing the objective function is a mathematical expression used to quantify control performance and constraint violation. By minimizing the objective function, the system can weigh different control objectives, such as temperature tracking accuracy, control stability, and energy consumption, to obtain the optimal control sequence within the current prediction time domain. This minimization process can be efficiently handled using a quadratic programming (QP) solver, which can handle linear systems and quadratic objective functions and integrate linear inequalities and equality constraints; or optimization algorithms such as the interior-point method or the active-set method can be used. These algorithms are suitable for solving large-scale convex optimization problems and can quickly converge to the global optimum.

[0111] The first term, error penalty, is a component of the objective function. It specifically measures the deviation between the system output (e.g., temperature) and the set target value, and applies a "penalty" to this deviation. Its purpose is to guide the optimization process, making the system output as close as possible to the target value, thereby improving control accuracy. This term is achieved by calculating the difference between the predicted state and the target temperature at each sampling point in the prediction time domain, and then summing these differences by weighted squares. The second term, control constraint, is another component of the objective function. It limits the magnitude or absolute value of changes in the control input to ensure the smoothness of the control action and protect the hardware. Its purpose is to avoid drastic changes in the control quantity, thereby reducing the impact on the H-bridge drive circuit, extending hardware lifespan, and contributing to system stability. This term penalizes the rate of change of the control quantity by calculating the weighted sum of squares of the control increments at adjacent sampling times.

[0112] Through the above technical solution, this application constructs a dual objective function containing an error penalty term and a control constraint term, which can simultaneously meet the dual requirements of temperature control accuracy and hardware protection when solving for the optimal control sequence. The error penalty term is based on the temperature evolution trajectory of multiple future control cycles predicted by the predictive control module using the thermal inertia physical model of the thermoelectric cooler. By minimizing the weighted squared deviation between the predicted state and the target temperature, the system can adjust for potential future temperature deviations in advance, effectively suppressing temperature overshoot and oscillations, thereby significantly improving temperature control accuracy. Simultaneously, the control constraint term limits the change in duty cycle between adjacent sampling times, i.e., the control increment, and, combined with the control weighting matrix R and the LC filter circuit, avoids instantaneous current surges and thermal shocks to the thermoelectric cooler caused by sudden changes in control quantity, thus effectively protecting the H-bridge drive circuit and the thermoelectric cooler, and extending the hardware's lifespan. In summary, the design of this objective function enables the system to meet high-precision temperature control requirements while also ensuring long-term stable operation and lifespan of the hardware, achieving an optimized balance between performance and reliability.

[0113] In some of the schemes mentioned above in this application, a feedback correction step is proposed to update the state estimate, compensate for the impact of environmental disturbances on the prediction, and provide an input benchmark for the prediction of the next control cycle. However, in this process, without a clear and reasonable state correction calculation method, it is impossible to accurately combine the actual sampled temperature data and the model's prior prediction results to correct the state estimate, effectively adjust the correction weight of the residual, accurately eliminate the deviation between the model prediction and the actual temperature, and effectively compensate for the model deviation caused by environmental disturbances. This will affect the accuracy of subsequent prediction steps, thereby reducing the temperature control accuracy and the system's anti-disturbance capability.

[0114] In response, this application further proposes that the feedback correction step adopt the following state correction formula:

[0115]

[0116] in, This is the current state estimate; The current time-priority predicted state is calculated by the prediction and control module of the previous cycle; L is the observation gain matrix, used to adjust the correction weights of the residuals. C represents the real-time sampled values, and C is the measurement matrix. This represents the residual between the actual temperature and the model prediction.

[0117] Specifically, the current state estimate refers to the optimal state representation of the system at the current moment after feedback correction. It integrates model prediction information and actual measurement data, serving as the input benchmark for the prediction step in the next control cycle. This state estimate can be a posterior state estimate obtained through the Kalman filter algorithm, or an estimate obtained by processing the nonlinear system using the extended Kalman filter algorithm. The prior predicted state at the current moment, derived from the previous cycle's predictive control module, refers to the system state predicted by the model at the current moment based on the system model and the state estimate from the previous moment. It represents the model's expectation of the system's future evolution. For example, it can be obtained by substituting the state estimate from the previous moment into the discretized state-space model for one-step prediction, or by considering a known system disturbance model for prediction.

[0118] The observation gain matrix L is used to adjust the weighting of the influence of the actual measurement residuals on the state estimation correction. Its role is to balance the reliability of the model prediction with the accuracy of the actual measurement data. Real-time sampled values ​​refer to the actual temperature data at the current moment, acquired in real time through a temperature sampling circuit. For example, this value can be directly derived from the output of a temperature sensor (such as a thermistor, platinum resistance thermometer, or thermocouple) after analog-to-digital conversion, or, to improve measurement accuracy and anti-interference capability, it can be the result of averaging data from multiple sensors or digitally filtering sampled data over a period of time.

[0119] The measurement matrix C is used to map the system state vector to the observable measurement space, so that the state variables predicted by the model can be directly compared with the actual sampled values. For example, if the system state vector contains temperature and the rate of temperature change, and the measured value is only temperature, then C can be a selection matrix, such as

[10] , used to extract the temperature component from the state vector; or, if the measured value is a linear combination of state variables, then C is the corresponding linear transformation matrix. The residual between the actual temperature and the model prediction value refers to the difference between the real-time sampled value and the predicted measured value after the prior predicted state is mapped to the measurement space through the measurement matrix C. This residual directly quantifies the deviation between the model prediction and the actual system behavior. For example, it can be obtained directly by calculation, or in some cases, the residual can be weighted or filtered to improve robustness.

[0120] Through the above technical solution, this application provides a clear and executable state correction calculation method for the feedback correction step. This method can effectively combine the pre-obtained prediction results from the model with the actual collected temperature data to accurately correct the system state. Specifically, by introducing the observation gain matrix L, the system can flexibly adjust the influence weight of the actual measurement residuals on the state correction, thereby achieving an optimal balance between model prediction and actual measurement, effectively eliminating model deviations caused by uncertainties such as environmental disturbances. The measurement matrix C ensures accurate matching between the model-predicted state and the actual sampled values, enabling the residuals to truly reflect system deviations. Ultimately, the corrected current state estimate can more accurately reflect the actual operating state of the system, providing a precise input benchmark for the prediction step in the next control cycle, significantly improving the prediction accuracy of the predictive control module, and thus ensuring the temperature control accuracy and anti-disturbance capability of the entire temperature control system, enabling the system to maintain stable and efficient temperature control performance even in complex and changing environments.

[0121] In some of the solutions mentioned above in this application, a predictive control module is proposed to realize model predictive control based on a thermal inertial physical model and to perform high-precision temperature control on a semiconductor refrigeration chip. However, in this process, if the logical architecture of the predictive control module is not clearly functionally divided, it will lead to chaotic control logic, unreasonable allocation of controller computing resources, difficulty in completing a series of calculation processes such as multi-step prediction and rolling optimization required for model predictive control online in real time, and inability to stably and accurately output the optimal control result obtained from the calculation to the downstream drive circuit. It is also not conducive to the debugging of the control module and subsequent function updates.

[0122] In this regard, this application further proposes that the predictive control module includes a state observation unit, a multi-step prediction unit, a rolling optimization unit, and an execution drive unit in its logical architecture.

[0123] The state observation unit is used to eliminate environmental disturbance biases based on real-time data from the temperature sampling circuit and predicted data from the previous cycle, using a state observer algorithm to output a corrected state estimate for the current moment. This state observation unit can employ a Kalman filter algorithm, which integrates real-time measurement data from the temperature sampling circuit and predicted data from the system dynamic model (i.e., the thermal inertia physical model of the cooling element). Through iterative calculation, it performs optimal estimation of the system state, effectively handling measurement and process noise. Alternatively, the state observation unit can use a Luenberger observer. By constructing a mathematical model parallel to the actual system and using the error between the system output and the observer output to correct the observer's state, it asymptotically converges to the actual system state, thus ensuring the observer's robustness to environmental disturbances.

[0124] The multi-step prediction unit is used to calculate the temperature evolution at multiple future sampling points in the time domain, starting from the corrected state estimate and utilizing a pre-defined thermoelectric thermal inertia physical model. This multi-step prediction unit can use the corrected state estimate as the initial condition of the thermoelectric thermal inertia physical model, combined with future control input sequences, to perform iterative simulations in the prediction time domain. It uses a discretized state-space model to calculate the temperature state at the next moment, gradually deducing the temperature evolution trajectory at multiple future sampling points. Alternatively, the multi-step prediction unit can directly use matrix exponentiation to calculate the future state. By pre-calculating the powers of the system state matrix A and combining the current corrected state estimate with the expected control input sequence, it can calculate the temperature evolution at all future sampling points in the prediction time domain in one go, thereby improving computational efficiency.

[0125] The rolling optimization unit establishes a cost function based on preset temperature control accuracy targets and hardware action constraints, and generates the optimal control sequence instructions by solving a quadratic programming problem online. This rolling optimization unit can use the interior-point method to solve the quadratic programming problem online. The interior-point method transforms the quadratic programming problem into solving a series of linear equations, quickly finding the optimal control sequence that satisfies the temperature control accuracy target and hardware action constraints within each control cycle. Alternatively, the rolling optimization unit can also use the active-set method to solve the quadratic programming problem. By iteratively identifying and processing constraints that are active at the optimal solution, the original problem is decomposed into a series of unconstrained or equality-constrained subproblems, thereby effectively generating the optimal control sequence instructions.

[0126] The execution drive unit converts the first term of the optimal control sequence into a corresponding PWM signal pulse width and polarity control signal, and sends it to the H-bridge drive circuit. This execution drive unit can integrate a hardware PWM generator module. This module receives the first term of the optimal control sequence output by the rolling optimization unit, directly converts it into a PWM signal with the corresponding pulse width and polarity, and controls the switching state of the H-bridge drive circuit according to the polarity information. Alternatively, in resource-constrained microcontrollers, the execution drive unit can also simulate the PWM signal through software. Using timer interrupts or loop control, the software can calculate the high-level duration based on the first term of the optimal control sequence and precisely control the level toggling of the GPIO pins, thereby generating the required PWM signal.

[0127] By functionally dividing the predictive control module into four independent logical units, each responsible for a specific step in the model predictive control flow, the overall predictive control logic becomes clearer. Specifically, the state observation unit, by fusing and correcting multi-source information, overcomes the limitation of relying solely on real-time sampled data to fully acquire system state information. It also effectively compensates for model biases caused by environmental disturbances, providing an accurate starting point for subsequent prediction steps, thus ensuring the adaptability of the entire control system to dynamic environmental changes and the accuracy of predictions. The multi-step prediction unit starts with the current corrected accurate state, ensuring that the multi-step prediction results are free from initial bias. Simultaneously, it utilizes a pre-established physical model for rapid calculation, ensuring the real-time nature of the prediction process and providing a timely and accurate predictive basis for subsequent optimization decisions. The rolling optimization unit not only ensures that the final temperature control effect meets accuracy requirements but also effectively avoids control outputs exceeding the capacity of the H-bridge drive circuit and the thermoelectric cooler by considering hardware action constraints, thereby protecting the hardware circuitry and extending the lifespan of the thermoelectric cooler. The online solution method allows the system to dynamically adjust the control strategy based on the updated state information in each control cycle, enhancing the adaptability of the control. The execution drive unit follows the principle of model predictive control rolling optimization, which only executes the first control variable. This ensures that the control output can be updated with the latest state information in each control cycle. This significantly improves the anti-disturbance capability of the control system, enabling it to better adapt to dynamically changing environmental conditions and stably transmit the optimal control command to the H-bridge drive circuit, thereby achieving high-precision and high-efficiency drive of the semiconductor cooling chip.

[0128] In summary, the above technical solution provides a clear functional division and reasonable allocation of computing resources for the predictive control module. It can complete the multi-step prediction and rolling optimization processes required for model predictive control online in real time, and stably and accurately output the optimal control result to the downstream drive circuit. This not only solves the problems of chaotic control logic and low computational efficiency, but also greatly facilitates the debugging of the control module and subsequent function updates, improving the robustness, accuracy, and reliability of the entire temperature control system.

[0129] The following example will provide a more detailed explanation of the above technical solution:

[0130] In a high-precision optics laboratory, a critical laser requires its operating cavity temperature to be precisely stabilized at 25.0°C to ensure the stability of the output wavelength. This temperature control system is integrated within the laser cavity.

[0131] The system first includes a semiconductor cooling assembly, in which a semiconductor cooler is directly attached to the laser cavity wall to provide cooling or heating capabilities. An H-bridge drive circuit, a high-voltage follower circuit, and an LC filter circuit are responsible for precisely driving the semiconductor cooler. Simultaneously, a temperature sampling circuit in the sensing and control assembly (e.g., a high-precision platinum resistance temperature sensor PT1000) continuously monitors the real-time temperature of the laser cavity and transmits the data to the microcontroller. A power supply provides stable energy to the entire system.

[0132] When the laser is started or the ambient temperature changes, the cavity temperature may deviate from the set value of 25.0°C. At this time, the predictive control module integrated in the microcontroller controller begins to function. This module first utilizes a pre-established thermal inertia physical model of the thermoelectric cooler, which describes the thermal inertia parameters of the thermoelectric cooler, such as specific heat capacity, mass, heat dissipation coefficient, and electrothermal conversion efficiency, using discretized state-space mathematical expressions (as described in claim 4).

[0133] Within each control cycle, the predictive control module iteratively performs the following steps to solve for the optimal control variable in real time:

[0134] The first step is the prediction process. Starting with the current state estimate (including temperature state variables and the rate of temperature change) after feedback correction, the predictive control module uses a thermal inertia physical model of the cooling chip to predict the temperature evolution trajectory over several future control cycles. For example, if the current cavity temperature is 26.0°C, the prediction module will predict how the cavity temperature will change under different control variables over the next 10 control cycles. This predictive capability effectively solves the problems of response lag and overshoot in traditional PID control when dealing with large inertia systems.

[0135] Next comes the rolling optimization step. The predictive control module finds an optimal control sequence by solving a quadratic programming problem (the objective function as described in claim 6). This sequence aims to minimize the deviation between the predicted temperature evolution trajectory and the target temperature of 25.0°C, while limiting the amplitude of the H-bridge drive circuit to work in conjunction with the LC filter circuit to suppress thermal shock to the thermoelectric cooler, thereby protecting the hardware and extending its lifespan. For example, if the predicted display cavity temperature may fall below 25.0°C at some point in the future under the current cooling trend, resulting in temperature overshoot, the predictive control module will adjust the polarity of the control quantity in advance before the temperature reaches 25.0°C, outputting a negative area control sequence (i.e., reducing cooling or starting heating). This utilizes the heating capacity of the thermoelectric cooler to offset the thermal inertia of the system, achieving a smooth transition and avoiding the sudden changes in control quantity and instantaneous current surges to the H-bridge drive circuit caused by traditional hard switching logic.

[0136] Finally, there is the feedback correction step. After executing the first control variable of the optimal control sequence, the temperature sampling circuit reads the actual temperature data of the laser cavity in real time. The predictive control module calculates the residual between the actual temperature and the model prediction value, and updates the state estimate accordingly (the state correction formula as described in claim 7). This updated state estimate serves as the input benchmark for the prediction step of the next control cycle, effectively compensating for the impact of environmental disturbances such as power supply voltage fluctuations, changes in the environmental heat dissipation coefficient, and random thermal load disturbances on the prediction. This solves the problem that traditional feedback control lacks the ability to predict the future trend of the system and can only adjust after the fact.

[0137] Through the aforementioned collaborative operation, the system can achieve high-precision, fast-response, and smooth control of the laser cavity temperature, effectively overcoming the challenges inherent in semiconductor cooling systems such as thermal inertia, hardware shocks, and environmental disturbances, and ensuring the stable operation of the laser under various operating conditions.

[0138] In this document, the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inner", "outer", "vertical", and "horizontal" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the purpose of clarifying the technical solution and for the convenience of description, and therefore should not be construed as limiting this application.

[0139] In this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, which includes not only the elements listed but also other elements not expressly listed.

[0140] The foregoing has shown and described the basic principles, main features, and advantages of this application. Those skilled in the art should understand that this application is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this application. Various changes and modifications can be made to this application without departing from the spirit and scope thereof, and all such changes and modifications fall within the scope of this application as claimed. The scope of protection of this application is defined by the appended claims and their equivalents.

Claims

1. A novel temperature control system based on model predictive control, characterized in that, include: Semiconductor cooling components: including semiconductor cooling chips, H-bridge drive circuits, high-voltage follower circuits, and LC filter circuits; Sensing and control components: including a microcontroller, temperature sampling circuit, and power supply; The feature is that the microcontroller controller integrates a predictive control module; the predictive control module is based on the established thermal inertia physical model of the cooling chip, and solves the optimal control sequence in each control cycle by collecting real-time data from the temperature sampling circuit, and outputs it to the H-bridge drive circuit.

2. The novel temperature control system based on model predictive control according to claim 1, characterized in that, The predictive control module executes the following steps cyclically within each control cycle to solve for the optimal control quantity in real time: Prediction steps: Using the thermal inertia physical model of the cooling chip, starting from the corrected state estimate at the current moment, predict the temperature evolution trajectory for multiple future control cycles; Rolling optimization steps: By solving a quadratic programming problem, find an optimal control sequence that minimizes the deviation between the temperature evolution trajectory and the set target temperature, while also minimizing the amplitude of the H-bridge drive circuit. Feedback correction step: Data is read in real time through temperature sampling circuit, the residual between the actual temperature and the model prediction value after the execution of the control sequence is calculated, and the state estimate is updated accordingly to compensate for the impact of environmental disturbances on the prediction. The updated state estimate is used as the input benchmark for the prediction step of the next control cycle.

3. The novel temperature control system based on model predictive control according to claim 2, characterized in that, The thermal inertia physical model of the cooling element adopts the following discretized state-space mathematical expression: Where x(k) is the state vector of the system at the k-th sampling time, including the temperature state quantity and the rate of temperature change; u(k) is the control quantity applied to the H-bridge drive circuit at the k-th sampling time, and its value corresponds to the duty cycle value of the control sequence. A is the system state matrix, which is composed of thermal inertia parameters determined by the specific heat capacity, mass, and heat dissipation coefficient of the cooling element; B is the input matrix, which consists of the electrothermal conversion efficiency of the thermoelectric cooler and the voltage gain parameters of the high-voltage follower circuit.

4. The novel temperature control system based on model predictive control according to claim 3, characterized in that, In the prediction step, the mathematical expression for transforming the state estimate into a temperature evolution trajectory is as follows: in, Let $k$ be the state quantity predicted at time $k+i$ in the future at time $k$. Let be the i-th power of the system state matrix A, representing the degree of influence of the system's inherent characteristics on the future state after i periods; This is the current state estimate, output by the feedback correction step; The combined gain represents the residual effect of the action performed at time j on the target time i after the remaining time has elapsed. Let be the control sequence, representing the control signal planned to be output at time k+j in the future.

5. The novel temperature control system based on model predictive control according to claim 4, characterized in that, In the optimization step, the optimal control sequence is obtained by minimizing the following objective function. : In the formula, the first term is the error penalty term, and the second term is the control constraint term; Let $k$ be the state quantity predicted at time $k+i$ in the future at time $k$. The target temperature at time k+i in the future; P is the prediction time domain, and Q is the state weighting matrix; It is the weighted square norm; To control the increment, it represents the change in duty cycle between adjacent sampling times; M represents the control time domain; R is the control weighting matrix, used to limit the control increment to work in conjunction with the LC filter circuit to suppress the thermal shock to the cooling chip. It is the weighted square norm.

6. The novel temperature control system based on model predictive control according to claim 5, characterized in that, The feedback correction step uses the following state correction formula: in, This is the current state estimate; The current time-priority predicted state is calculated by the prediction and control module of the previous cycle; L is the observation gain matrix, used to adjust the correction weights of the residuals. C represents the real-time sampled values, and C is the measurement matrix. This represents the residual between the actual temperature and the model prediction.

7. The novel temperature control system based on model predictive control according to claim 1, characterized in that, The predictive control module includes the following in its logical architecture: State observation unit: It is used to eliminate environmental disturbance deviations and output the corrected state estimate at the current moment based on the real-time data of the temperature sampling circuit and the prediction data of the previous cycle through the state observer algorithm. Multi-step prediction unit: used to calculate the temperature evolution state at multiple future sampling points in the time domain, starting from the corrected state estimate and using a preset thermal inertia physical model of the cooling chip; Rolling optimization unit: It is used to establish a cost function based on the preset temperature control accuracy target and hardware action constraints, and generate the optimal control sequence command by solving the quadratic programming problem online; Execution drive unit: used to convert the first item of the optimal control sequence into the corresponding PWM signal pulse width and polarity control signal, and send it to the H-bridge drive circuit.

Images