Sample pretreatment intelligent temperature control system based on edge computing

Abstract

The application discloses a sample pretreatment intelligent temperature control system based on edge computing and relates to the technical field of automatic control. Original sensing data of a heating device are collected to build a mapping relationship between a physical entity and a mathematical model and output thermal physical parameters. A core temperature estimation value is output by processing the thermal physical parameters by using an unscented Kalman filtering algorithm. Sample control output and temperature response are monitored, load mutation characteristics are extracted, and a feedforward compensation control amount is output. Deviation between a preset target temperature and the core temperature estimation value is calculated, basic PID control amount is output based on variable domain fuzzy control theory, and the feedforward compensation control amount and the basic PID control amount are superimposed and calculated to obtain a temperature driving control signal, so that the utilization efficiency of temperature control energy is improved, and the temperature fluctuation amplitude is prevented from being too large.

Description

Technical Field

[0001] This invention relates to the field of automatic control technology, and in particular to an intelligent temperature control system for sample pretreatment based on edge computing. Background Technology

[0002] Sample pretreatment is a crucial step, and the accuracy and stability of temperature control directly determine the repeatability and accuracy of experimental results. Traditional sample pretreatment equipment often uses standard PID control algorithms combined with thermocouple sensors for temperature control. However, with increasing demands for automation and intelligence, the limitations of traditional technologies are becoming increasingly apparent.

[0003] Sensors are typically placed on the surface of the heating block or near the heating element, rather than inside the sample. Because of the significant thermal resistance and heat capacity involved in heat transfer from the heating source to the sample core, a large time lag and spatial temperature difference exist between the surface temperature and the sample core temperature. Traditional temperature control systems can only achieve surface temperature control and cannot truly provide feedback and control of the sample's core temperature, easily leading to sample overheating, oxidation, or incomplete reactions. Secondly, the nonlinearity and time-varying nature of the system parameters pose a huge challenge. The type and volume of the sample, the material of the container, and the aging degree of the heating block will significantly change the equivalent thermal resistance and heat capacity of the system. Fixed-parameter PID controllers are difficult to adapt to such variable physical characteristics and often need to be manually tuned after changing the experimental scheme, which is inefficient and difficult to guarantee the global optimum. The effects of external disturbances and internal reaction heat are difficult to eliminate. The sample pretreatment process is often accompanied by intense exothermic or endothermic chemical reactions. Traditional feedback control has the nature of hysteresis regulation, that is, it only begins to correct after the temperature deviates. It cannot predict in advance and actively suppress, and is very prone to overshoot or fluctuation. Summary of the Invention

[0004] The technical problems solved by this invention are: existing technologies have difficulty in solving the problem of sensing the core temperature of a sample, which cannot be directly measured; changes in the real physical environment can easily lead to model mismatch; and traditional PID cannot resolve the trade-off between dynamic response and steady-state accuracy.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: an intelligent temperature control system for sample pretreatment based on edge computing, comprising a data fitting module, an intelligent correction module, a load identification and active suppression module, and a rule optimization module. The data fitting module is used to collect the original sensing data of the heating device, construct the mapping relationship between the physical entity and the mathematical model, and output thermophysical parameters; The intelligent correction module is used to process the thermophysical parameters using an unscented Kalman filter algorithm and output an estimated core temperature value. The load identification and active suppression module is used to monitor the sample control output and temperature response, extract load mutation characteristics, and output feedforward compensation control quantity. The rule optimization module is used to calculate the deviation between the preset target temperature and the core temperature estimate, output the basic PID control quantity based on the variable universe fuzzy control theory, and superimpose the feedforward compensation control quantity and the basic PID control quantity to obtain the temperature drive control signal.

[0006] Preferably, the data fitting module includes a data acquisition unit, a feature reconstruction unit, and a parameter identification unit; The data acquisition unit is used to synchronously acquire surface temperature data of the heating block, ambient temperature data and the drive control signal of the previous moment at a preset sampling frequency, and to perform moving average filtering on the data to obtain the original sensing data. The drive control signal of the previous moment includes the PWM duty cycle value, the heating power setting value, and the switching state. The PWM duty cycle value is used to represent the average electrical power applied to the heating block in the previous moment, the heating power setting value is used to represent the target output power set by the controller in the previous moment, and the switching state is used to represent the on / off state of the heating relay in the previous moment.

[0007] Preferably, the feature reconstruction unit is used to perform max-min normalization processing on the original sensing data to construct a normalized input feature vector of the current time and the original sensing data; The annular buffer region is set to cover N time points. Raw sensor data is input into the annular buffer region. After each sampling cycle, the surface temperature data of the heating block, the ambient temperature data, and the drive control signal from the previous time point are stored in the buffer. The oldest data outside the time window is automatically discarded, ensuring the buffer always stores the most recent N time points. Interpolation calculations are performed on the raw sensor data to obtain the effective temperature rise variable. The expression for calculating the effective temperature rise variable is: ; in, For effective temperature increase variables, This is the surface temperature data of the heating block. This refers to ambient temperature data. The effective temperature rise variable and the original sensor data are subjected to max-min normalization processing, and an input feature vector is constructed based on the controlled autoregressive model structure. The input feature vector is... ; in, For the normalized effective temperature rise variables of the first i time steps, The original sensor data for the first i time steps after normalization. This is the input feature vector.

[0008] Preferably, the parameter identification unit is used to perform iterative calculations on the input feature vector using a recursive least squares method with a forgetting factor, identify the equivalent thermal resistance and equivalent heat capacity in real time, and output the identified equivalent thermal resistance and equivalent heat capacity as thermophysical parameters. A controlled autoregressive model is constructed, defining a linear relationship equation between the effective temperature rise variable at the current moment and the input feature vector, and initializing the parameter vector to be identified and the covariance matrix. The linear relationship equation is as follows: ; in, The normalized effective temperature rise variable at the current moment. The parameter vector to be identified, For fitting residuals; In each sampling period, the Kalman gain vector is calculated using the covariance matrix of the previous time step and the current input feature vector. The prior prediction error is calculated, and the parameter estimates of the previous time step are corrected using this error and the Kalman gain vector. The coefficients of the controlled autoregressive model updated at the current time step are obtained, and the covariance matrix is ​​updated. The identified coefficients are mapped and inverted using the Euler approximation method to calculate the equivalent thermal resistance and equivalent heat capacity. The equivalent thermal resistance and equivalent heat capacity are output as thermophysical parameters.

[0009] Preferably, the intelligent correction module includes a model building unit and a state update unit; The model building unit is used to receive thermophysical parameters, establish a second-order lumped parameter thermal network model containing surface temperature nodes of the heating block and core temperature nodes of the sample, and generate a set of sampling points through unscented transformation based on the current state estimate and covariance matrix.

[0010] Preferably, the state update unit includes: The sampling point set is substituted into a second-order lumped parameter thermal network model for time update to obtain a prior estimate of the core temperature. The surface temperature data of the original sensor data is used as the observation value to calculate the Kalman gain matrix. The Kalman gain matrix is ​​used to correct the prior estimate of the core temperature and output the core temperature estimate.

[0011] Preferably, the load identification and active suppression module includes a state observation unit and an active compensation unit; The state observation unit includes: Construct a linear extended state observer, define a state vector including the system output temperature and the rate of temperature change, introduce extended state variables, the initial values ​​of which are the sum of the rates of change of thermophysical parameters, and establish the state equation of the observer: ; in, The actual measured temperature output. To control the input, This is an estimate of the control gain. , , The gain coefficient of the observer. The system's output temperature, The rate of change of temperature, To expand the state variables, the state equations are iteratively calculated, and the expanded state variables are output in real time as the total disturbance estimate. The estimated total disturbance is received and frequency-domain filtered using a high-pass filter and a low-pass filter to obtain the load change characteristic signal.

[0012] Preferably, the active compensation unit includes: Real-time monitoring of load change characteristic signals, execution of threshold determination logic, and generation of feedforward compensation control quantity: If a negative load mutation characteristic signal is detected and its amplitude exceeds the first preset threshold, it is determined that the current system is connected to a heat-absorbing load, and a positive feedforward compensation control quantity is immediately generated through inversion calculation. If a positive load mutation characteristic signal is detected and its amplitude exceeds the second preset threshold, it is determined that an exothermic chemical reaction has occurred in the system, and a negative feedforward compensation control quantity is immediately generated. If the load mutation characteristic signal is between the first preset threshold and the second preset threshold, the system is determined to be in a steady state and outputs zero compensation.

[0013] The feedforward compensation control quantity is directly superimposed on the output of the feedback control loop.

[0014] Preferably, the rule optimization module includes an error calculation unit and a control quantity calculation unit; The error calculation unit includes: Based on the core temperature estimate and the preset target temperature, calculate the temperature deviation and the rate of change of deviation, preset a deviation threshold, and compare the absolute value of the temperature deviation with the deviation threshold: When the absolute value of the temperature deviation is greater than or equal to the deviation threshold, the universe of discourse scaling factor is adjusted to 1. When the absolute value of the temperature deviation is less than the deviation threshold, the universe of discourse scaling factor is calculated using the exponential decay function. The formula for the exponential decay function is: ; in, The current domain scaling factor, The deviation threshold, This is the sensitivity adjustment coefficient. This is the minimum resolution protection value; The current temperature deviation and the rate of change of deviation are calculated by comparing them with the current universe of discourse scaling factor to obtain the first fuzzy value. The first fuzzy value includes the fuzzy temperature deviation and the fuzzy deviation rate of change. The first fuzzy value is mapped to a preset fuzzy universe of discourse interval. The fuzzy rule base is queried to obtain the proportional coefficient correction, integral coefficient correction and derivative coefficient correction of the PID controller.

[0015] Preferably, the control quantity calculation unit includes: The proportional, integral, and derivative coefficients of the PID controller are added together to obtain the current PID parameters. The basic PID control quantity is calculated using the incremental PID algorithm. The temperature drive control signal is obtained by superimposing the basic PID control quantity with the feedforward compensation control quantity.

[0016] The beneficial effects of this invention are as follows: Through the deep coupling of unscented Kalman filtering and second-order thermal network model, the system can estimate the core temperature inside the sample in real time, which is significantly better than the traditional control scheme that relies solely on surface temperature measurement. Combined with the localized processing of edge computing, the real-time performance of the temperature control process is ensured. This invention lowers the barrier to entry and maintenance costs. Through the LESO observer, it achieves transparent monitoring of total disturbances. For dangerous conditions such as violent exothermic reactions commonly encountered in sample pretreatment, this invention can identify and execute maximum negative compensation in advance through reaction heat determination logic, effectively avoiding the risk of thermal runaway. Simultaneously, the feedforward compensation technology reduces temperature fluctuation amplitude when facing external disturbances such as opening the lid. The variable universe of discourse fuzzy PID algorithm dynamically adjusts the scaling factor, enabling the control logic to use an expanded universe of discourse for full-power propulsion during the heating phase, shortening experimental time. During the plateau phase, a contracted universe of discourse is used for fine current maintenance, avoiding energy consumption and electromagnetic interference caused by frequent switching. Attached Figure Description

[0017] Figure 1 A schematic diagram of the basic process of an intelligent temperature control system for sample pretreatment based on edge computing, provided as an embodiment of the present invention; Figure 2 System response curves for different set temperatures. Detailed Implementation

[0018] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0019] Example, refer to Figure 1-2This paper presents an intelligent temperature control system for sample pretreatment based on edge computing, including a data fitting module, an intelligent correction module, a load identification and active suppression module, and a rule optimization module. The data fitting module is used to collect raw sensor data of the heating device, construct the mapping relationship between the physical entity and the mathematical model, and output thermophysical parameters. The intelligent correction module is used to process the thermophysical parameters using the unscented Kalman filter algorithm and output the core temperature estimate. The load identification and active suppression module is used to monitor the sample control output and temperature response, extract load mutation characteristics, and output feedforward compensation control quantity; The rule optimization module is used to calculate the deviation between the preset target temperature and the core temperature estimate. Based on the variable universe fuzzy control theory, it outputs the basic PID control quantity. The feedforward compensation control quantity and the basic PID control quantity are superimposed to calculate the temperature drive control signal.

[0020] The data fitting module includes a data acquisition unit, a feature reconstruction unit, and a parameter identification unit; The data acquisition unit is used to synchronously acquire surface temperature data of the heating block, ambient temperature data and the drive control signal of the previous moment at a preset sampling frequency, and to perform moving average filtering on the data to obtain the raw sensing data. The drive control signal of the previous moment includes the PWM duty cycle value, the heating power setting value, and the switching state. The PWM duty cycle value is used to represent the average electrical power applied to the heating block in the previous moment, the heating power setting value is used to represent the target output power set by the controller in the previous moment, and the switching state is used to represent the on / off state of the heating relay in the previous moment.

[0021] In this embodiment, the drive control signal output in the k-1th cycle is read in the kth sampling cycle. The drive control signal is the PWM comparison value stored in the MCU timer register, with a value range of 0 to 1000, corresponding to 0% to 100% heating power output. The switch state is 1 and 0, where 1 indicates power-on heating and 0 indicates power-off stopping.

[0022] The feature reconstruction unit is used to perform max-min normalization on the original sensing data to construct the normalized input feature vector of the current time and the original sensing data. The annular buffer region is set to cover N time points. Raw sensor data is input into the annular buffer region. After each sampling cycle, the surface temperature data of the heating block, the ambient temperature data, and the drive control signal from the previous time point are stored in the buffer. The oldest data outside the time window is automatically discarded, ensuring the buffer always stores the most recent N time points. Interpolation calculations are performed on the raw sensor data to obtain the effective temperature rise variable. The expression for the effective temperature rise variable is: ; in, For effective temperature increase variables, This is the surface temperature data of the heating block. This refers to ambient temperature data. The effective temperature rise variable and the raw sensor data are subjected to max-min normalization, and an input feature vector is constructed based on the controlled autoregressive model structure. The input feature vector is... ; in, For the normalized effective temperature rise variables of the first i time steps, The original sensor data for the first i time steps after normalization. This is the input feature vector.

[0023] In this embodiment, the annular buffer region is set to N=10, and operates with a sampling period of 100ms. Whenever a sampling interruption is triggered, the data acquisition unit acquires the raw sensor data at the current time k, and the feature reconstruction unit performs an enqueue operation, writing the heating block surface temperature data, ambient temperature data, and the drive control signal of the previous cycle into the head of the annular buffer region. At the same time, the buffer pointer is moved to overwrite or discard the oldest data at time k−N. This mechanism ensures that the buffer always stores the dynamic data sequence within the most recent second, providing a data foundation for constructing a feature vector with historical memory characteristics. After interpolation calculation and normalization processing, the normalized data is extracted from the annular buffer based on the second-order controlled autoregressive model structure to construct the input feature vector.

[0024] The parameter identification unit is used to perform iterative calculations on the input feature vector using the recursive least squares method with a forgetting factor, identify the equivalent thermal resistance and equivalent heat capacity in real time, and output the identified equivalent thermal resistance and equivalent heat capacity as thermophysical parameters. Construct a controlled autoregressive model, define the linear relationship equation between the effective temperature rise variable at the current time and the input feature vector, and initialize the parameter vector to be identified and the covariance matrix. The linear relationship equation is as follows: ; in, The normalized effective temperature rise variable at the current moment. The parameter vector to be identified, For fitting residuals; In each sampling period, the Kalman gain vector is calculated using the covariance matrix of the previous time step and the current input feature vector. The prior prediction error is calculated, and the parameter estimates of the previous time step are corrected using this error and the Kalman gain vector. The coefficients of the controlled autoregressive model updated at the current time step are obtained, and the covariance matrix is ​​updated. The identified coefficients are mapped and inverted using the Euler approximation method to calculate the equivalent thermal resistance and equivalent heat capacity. The equivalent thermal resistance and equivalent heat capacity are output as thermophysical parameters.

[0025] In this embodiment, the parameter vector to be identified in the controlled autoregressive model Based on the first-order backward Euler difference discretization method, the discrete model coefficients and the equivalent thermal resistance of the continuous-time physical parameters are... and equivalent heat capacity The mapping inversion relationship is as follows: ; ; in, For each sampling period, the parameter identification unit updates the coefficient matrix using the recursive least squares method based on the aforementioned algebraic relationships. In real time, it is converted into thermal resistance and heat in the physical domain; By constructing a real-time updated physical thermal model through the data fitting module and combining it with the variable universe of discourse fuzzy control technology in the rule tuning module, this invention can automatically adjust PID parameters for different loads. In the heating stage, it uses the expanded universe of discourse to achieve full-speed heating and shorten the preheating time; in the isothermal stage, it uses the contracted universe of discourse to achieve fine adjustment, effectively solving the overshoot and oscillation problems that traditional PID control is prone to when facing large thermal inertial loads.

[0026] The intelligent correction module includes a model building unit and a state update unit; The model building unit is used to receive thermophysical parameters, establish a second-order lumped-parameter thermal network model containing surface temperature nodes of the heating block and core temperature nodes of the sample, and generate a set of sampling points through unscented transformation based on the current state estimate and covariance matrix.

[0027] In this embodiment, the equivalent thermal resistance received from the data fitting module is... and equivalent heat capacity Equivalent thermal resistance and equivalent heat capacity describe the dynamic heat transfer characteristics from the heating power input to the surface temperature of the heating block, and obtain the thermal resistance from the surface of the heating block to the sample core. and the heat capacity of the sample itself In this embodiment, and Instead of real-time online identification, the thermophysical properties of the samples are stored as preset parameters. These preset parameters are based on calibration values ​​from common sample types and labeled consumables and are stored in a parameter library. The system reads these parameters from the library. Common sample types include aqueous solutions, organic solvents, and heated samples, based on equivalent thermal resistance. Equivalent heat capacity Thermal resistance of the sample core and the heat capacity of the sample itself Establish the state equations for a second-order total parameter heat network model: ; in, For heating power, For ambient temperature, This is the surface temperature node of the heating block. This is the core temperature node of the sample. To determine the thermal resistance between the heating block and the environment, the state equations are discretized, resulting in a discrete state-space model: ; ; Wherein, the state vector is Control input The normalized heating power can be measured with ambient temperature as the disturbance, and the output can be observed. The readings are from the surface temperature sensor of the heating block. and These are process noise and observation noise, respectively. and The state estimate is obtained from the thermal resistance, thermal capacity, and sampling period using the Euler method; the initial state estimate is 0, i.e., the state vector is 0, and the covariance matrix is ​​set as a diagonal matrix. The state estimate and covariance matrix of the previous time step are received, and then an unscented transformation is performed to generate a set of sampling points. By establishing a second-order lumped-parameter thermal network model that includes surface temperature nodes of the heating block and core temperature nodes of the sample, the model building unit can accurately characterize the dynamic process of heat transfer from the heating source to the sample center based on the real-time identified thermal resistance and thermal capacity parameters. This breaks the limitation of traditional temperature control relying solely on surface sensors and can accurately calculate the core temperature inside the sample, which cannot be directly measured, without the need for invasive temperature measurement.

[0028] The state update unit includes: The sampling point set is substituted into the second-order lumped parameter thermal network model for time update to obtain the prior estimate of the core temperature. The surface temperature data of the original sensor data is used as the observation value to calculate the Kalman gain matrix. The Kalman gain matrix is ​​used to correct the prior estimate of the core temperature and output the core temperature estimate.

[0029] In this embodiment, during each sampling period k, the state update unit receives a set of sampling points, which includes a Sigma point set and corresponding mean weights and covariance weights, used to reflect the characteristics of the state distribution. The Sigma point set... Substituting the discretized second-order lumped-parameter thermal network state equations, the predicted Sigma point is calculated: ; Where n=2, is the number of state dimensions. The current control input (normalized heating power) is used. and It is a discrete system matrix determined by thermal resistance, thermal capacity, and sampling period. Based on the predicted Sigma points, the prior state estimate and prior covariance matrix are calculated. The predicted observations are calculated using the observation equations through the predicted Sigma points. Then, the mean, innovation covariance, and cross-covariance between the state and the observations are calculated. The prior state estimate is corrected using the actual surface temperature observation y(k+1) obtained from the data fitting module at the current moment, resulting in the posterior state estimate and posterior covariance matrix. The second component in the corrected state vector is the sample core temperature estimate at the current moment. This value is output to the rule tuning module to calculate the temperature deviation. At the same time, the posterior state estimate and posterior covariance matrix will be used as the input to the model building unit at the next moment, thereby realizing recursive estimation. When generating the sampling point set using the unscented transformation, the state vector dimension is set to n=2, and the scaling parameter of the Sigma point set is set as follows: β=2 is used to control the distribution range of Sigma points, and β=2 is used to introduce prior distribution information. Under a Gaussian distribution, 2 is optimal, and κ=0. The weight calculation formula is: ; ; ; in, , Weighted by mean, For covariance weights, As the weights for the remaining points, the initial process noise covariance matrix Q is set as a diagonal matrix diag(0.01,0.01) to reflect the uncertainty of the physical model; the observation noise covariance matrix R is set to a fixed value of 0.1 based on the nominal accuracy of the temperature sensor.

[0030] By using the sampling point set generated by the unscented transform for nonlinear filtering, the state update unit can integrate the prior knowledge of the physical model with the measured data of the sensor. By calculating the Kalman gain matrix, it dynamically adjusts the weight distribution of the model prediction value and the sensor observation value. When the sensor is subjected to electromagnetic interference or instantaneous jumps, it automatically reduces the weight of the observation value and outputs a smooth and stable core temperature estimate, preventing the PID controller from malfunctioning due to noise. In the early stage of heating or when the load changes suddenly, it can quickly correct the model state using the observation residuals, so that the core temperature estimate can quickly approach the true value within 3-5 sampling periods, shortening the system's settling time.

[0031] The load identification and active suppression module includes a state observation unit and an active compensation unit; The state observation unit includes: Construct a linear extended state observer, define a state vector including the system output temperature and the rate of temperature change, introduce extended state variables with initial values ​​of the sum of the rates of change of thermophysical parameters, and establish the state equation of the observer: ; ; ; in, The actual measured temperature output. To control the input, This is an estimate of the control gain. , , The gain coefficient of the observer. The system's output temperature, The rate of change of temperature, To expand the state variables, the state equations are iteratively calculated, and the expanded state variables are output in real time as the total disturbance estimate. In this embodiment, the observer gain coefficient is tuned based on the bandwidth parameterization method, letting... , , ,in The observer bandwidth is typically set to 3-5 times the closed-loop system bandwidth. The estimated total received disturbance is processed by frequency domain filtering using a high-pass filter and a low-pass filter to obtain the load change characteristic signal.

[0032] In a preferred embodiment, a first-order forward Euler method is used for differential discretization. The discretization step size is consistent with the system sampling period, which is set to 100ms. The recursive equation of the discretized linear extended state observer is: ; ; ; in, The system output temperature estimated by the observer at the current moment. This is an estimate of the rate of temperature change. This is for the extended state variables (i.e., the total disturbance estimate). The core temperature estimate is provided by the intelligent correction module. This is the control input for the current moment, i.e., the temperature drive control signal output from the previous cycle. For each sampling period, at each sampling interruption, the MCU updates the state observer sequentially according to the recursive equation of the discretized linearly extended state observer described above. , and The total disturbance estimate is obtained in real time and output to the filter; The gain coefficient is determined based on the bandwidth parameterization method: , , , The observer bandwidth is determined as follows: Under no-load conditions, a step power input is applied to the heating device, and the rise curve of the estimated core temperature is recorded. The system's principal time constant τ is obtained by fitting a first-order inertial plus pure time delay model. The closed-loop bandwidth can then be taken as ωc≈(5~10) / τ. The closed-loop bandwidth is taken as 3 to 5 times, i.e., ωo = 3ωc to 5ωc. In this embodiment... It is set to 2 rad / s; In the cascaded filter, the low-pass filter is used to suppress high-frequency noise, and its cutoff frequency is set to 0.1~0.2 times the sampling frequency. In this embodiment, the sampling frequency is 10Hz, and the low-pass cutoff frequency is 2Hz. The high-pass filter is used to filter out low-frequency components such as slow temperature drift. Its cutoff frequency is determined according to the rate of temperature change caused by load abrupt changes, and is set to half of the slowest abrupt change frequency to be detected. In this embodiment, it is set to 0.1Hz. Through this frequency domain separation, the characteristics of load abrupt changes caused by exothermic or endothermic chemical reactions or mechanical operations can be effectively extracted. In the recursive least squares method with a forgetting factor, the forgetting factor λ ranges from 0.95 to 0.995. When the thermophysical parameters of the system change slowly, such as the slow aging of the heating block, λ is set to 0.995. When it is necessary to quickly track parameter changes caused by sample replacement, λ is set to 0.95 to 0.98. In this embodiment, the initial value of λ is set to 0.98. This embodiment innovatively employs a linear extended state observer, using thermophysical parameters as extended state variables for real-time observation. This greatly simplifies the structure of the control system while ensuring universal observation capability for various types of disturbances, improving the robustness of temperature control observation. Through cascaded high-pass and low-pass filters, the observed total disturbance signal undergoes fine frequency domain processing, effectively eliminating low-frequency interference caused by slow changes in ambient temperature or sensor zero-point drift, preventing the system from accumulating errors due to long-term operation. It also filters out high-frequency electromagnetic noise, ensuring that subsequent active compensation actions are triggered only for real load mutations, avoiding frequent control jitter caused by noise, and extending the service life of the heating actuator.

[0033] The active compensation unit includes: Real-time monitoring of load change characteristic signals, execution of threshold determination logic, and generation of feedforward compensation control quantity: If a negative load mutation characteristic signal is detected and its amplitude exceeds the first preset threshold, it is determined that the current system is connected to a heat-absorbing load, and a positive feedforward compensation control quantity is immediately generated through inversion calculation. If a positive load mutation characteristic signal is detected and its amplitude exceeds the second preset threshold, it is determined that an exothermic chemical reaction has occurred in the system, and a negative feedforward compensation control quantity is immediately generated. If the load change characteristic signal is between the first preset threshold and the second preset threshold, the system is determined to be in a steady state and outputs zero compensation.

[0034] The feedforward compensation control quantity is directly superimposed on the output of the feedback control loop.

[0035] In this embodiment, the threshold for determining sudden changes in heat absorption and release load is obtained using the following calibration method: Under standard no-load conditions, after the system stabilizes at the target temperature, the fluctuation range of three times the estimated total disturbance value per unit time is recorded, and its peak-to-peak value is taken as the reference noise level N0. The first preset threshold is set as three times the change amplitude corresponding to 20% of the theoretical heating rate under full-scale heating power; the second preset threshold is set as three times the reference noise level N0. In this embodiment, the three-times amplitude corresponding to the first preset threshold is -0.5, and the second preset threshold corresponds to +0.2. Based on the amplitude and polarity of the characteristic signal, a feedforward compensation control quantity is generated directly before the feedback loop, realizing active suppression of interference. Before the temperature sensor detects a significant deviation, the heating power is adjusted in advance to keep the temperature fluctuation within a minimum range. For violent exothermic reactions, negative compensation can be quickly identified and triggered to prevent safety accidents from occurring in the sample.

[0036] The rule optimization module includes an error calculation unit and a control quantity calculation unit; The error calculation unit includes: Based on the core temperature estimate and the preset target temperature, calculate the temperature deviation and the rate of change of deviation, preset the deviation threshold, and compare the absolute value of the temperature deviation with the deviation threshold: When the absolute value of the temperature deviation is greater than or equal to the deviation threshold, the universe of discourse scaling factor is adjusted to 1. When the absolute value of the temperature deviation is less than the deviation threshold, the universe of discourse scaling factor is calculated using the exponential decay function. The formula for the exponential decay function is:

[0037] in, The current domain scaling factor, The deviation threshold, This is the sensitivity adjustment coefficient. This is the minimum resolution protection value; The current temperature deviation and the rate of change of deviation are calculated by comparing them with the current universe of discourse scaling factor to obtain the first fuzzy value. The first fuzzy value includes the fuzzy temperature deviation and the fuzzy deviation rate of change. The first fuzzy value is mapped to the preset fuzzy universe of discourse interval. The fuzzy rule base is queried to obtain the proportional coefficient correction, integral coefficient correction and derivative coefficient correction of the PID controller.

[0038] In this embodiment, the preset fuzzy universe of discourse interval is divided into 7 fuzzy sets: {Negative Large (NB), Negative Medium (NM), Negative Small (NS), Zero (ZO), Positive Small (PS), Positive Medium (PM), Positive Large (PB)}; The fuzzy rule base is built based on the following physical logic: When the temperature deviation is large, in order to speed up the response, a larger proportional coefficient correction amount and a smaller differential coefficient correction amount are used; When the temperature deviation is small and the rate of change of the deviation is large, in order to prevent overshoot, a smaller proportional coefficient correction amount and a larger differential coefficient correction amount should be used. The integral coefficient correction is moderately increased when the temperature deviation is moderate or small to eliminate steady-state error, and is zero when the deviation is too large to prevent integral saturation. After inputting the first fuzzy value into the fuzzy rule base for fuzzy inference, the centroid method is used for defuzzification, and the specific PID controller proportional coefficient correction, integral coefficient correction and derivative coefficient correction are output. The preset deviation threshold is 5 degrees Celsius, and is determined based on the maximum allowable dynamic error range of the experiment; the sensitivity adjustment coefficient γ is used to control the nonlinear curvature of the universe of discourse shrinkage, and its value ranges from 0.5 to 2.0; the minimum resolution protection value ϵ is an engineering constant (taken as 10). -4This is solely for preventing overflow where the denominator is zero in microprocessor program calculations. By setting a clear deviation threshold and dynamically adjusting the universe of discourse scaling factor in conjunction with the exponential decay function, the control strategy is piecewise optimized to maintain the maximum universe of discourse, enabling the system to output saturated or large control quantities. This allows the heating mechanism to be driven at maximum power, shortening the rise time from room temperature to the target temperature. Furthermore, it automatically maintains optimal control performance without the need for manual parameter readjustment to address changes in heating efficiency caused by temperature rise during sample pretreatment, overcoming the poor robustness of traditional fixed-parameter PID controllers under complex operating conditions.

[0039] The control quantity calculation unit includes: The proportional, integral, and derivative coefficients of the PID controller are added together to obtain the current PID parameters. The basic PID control quantity is calculated using the incremental PID algorithm. The temperature drive control signal is obtained by superimposing the basic PID control quantity with the feedforward compensation control quantity.

[0040] In this embodiment, the deviation threshold is set according to the temperature control accuracy level, generally 2 to 3 times the allowable fluctuation range of the target temperature, and is set to 5℃ in this embodiment. The sensitivity adjustment coefficient γ ranges from 1.0 to 2.0. The larger γ is, the faster the universe of discourse shrinks and the higher the steady-state accuracy, but it may cause slight fluctuations. In this embodiment, it is set to 1.5. The minimum resolution protection value ε is used to prevent the universe of discourse from shrinking excessively and causing dead zones, and is set to 0.05. All of the above parameters can be fine-tuned on-site according to the actual temperature control effect. By linearly superimposing the basic PID control quantity and the feedforward compensation control quantity, the steady-state zero steady-state error advantage of feedback control and the fast anti-disturbance advantage of feedforward control are combined. When a sudden load change occurs, the feedforward channel responds quickly to provide coarse adjustment compensation to offset most of the interference energy; subsequently, the PID feedback channel performs fine adjustment to eliminate the remaining residual error, ensuring that the temperature curve does not drop or surge significantly when the system is subjected to strong interference, which significantly improves the reliability and safety of the fully automatic resolution process. The underlying hardware of the system includes a microprocessor (MCU), a PWM solid-state relay, a metal heating block for placing the sample, and a PT100 high-precision temperature sensor attached to the surface of the heating block. The MCU runs the aforementioned data fitting module, intelligent correction module, load identification and active suppression module, and rule tuning module. Furthermore, to clarify the feasibility of this system on embedded edge devices, the following hardware configuration and computation cycle allocation examples are provided: The edge computing node required by this system is a microcontroller based on the ARM Cortex-M4 core, with a main frequency of no less than 72MHz, and equipped with at least 64KB SRAM and 256KB Flash storage. The temperature sensor communicates with the MCU via an SPI interface, and the PWM output controls a solid-state relay to drive the heating wire. The above hardware resources are sufficient to handle all algorithm calculations, enabling local real-time closed-loop control without cloud collaboration. The system adopts a multi-task timed scheduling mechanism with a basic control cycle of 100ms. Within each control cycle, the following tasks are executed sequentially: 0~5ms: The data acquisition unit reads the temperature sensor data through an interrupt and completes the moving average filtering; 5~15ms: The feature reconstruction unit updates the circular buffer, completing normalization and input feature vector construction; 15~30ms: The parameter identification unit runs the recursive least squares method and outputs thermophysical parameters; 30~50ms: The intelligent correction module performs unscented Kalman filtering time update and measurement update, and outputs the core temperature estimate; 50~70ms: The load identification and active suppression module first runs LESO discrete recursion to obtain the total disturbance, then extracts the load change characteristics through IIR high-pass / low-pass filters, performs threshold determination and generates feedforward compensation amount; 70~90ms: The rule optimization module calculates the temperature deviation and the rate of change of the deviation, updates the universe of discourse scaling factor, performs fuzzy inference by looking up the table and calculates the incremental PID output, which is then superimposed with the feedforward compensation to obtain the final temperature drive control signal. 90~100ms: Output the updated PWM duty cycle to complete one cycle of control.

[0041] Under the above timing arrangement, the single-cycle operation time of each module is less than 20ms, and the total computing load does not exceed 50% of the MCU's processing capacity, leaving sufficient margin to ensure long-term stable real-time operation of the system.

[0042] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0043] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A sample pretreatment intelligent temperature control system based on edge computing, characterized in that, It includes a data fitting module, an intelligent correction module, a load identification and active suppression module, and a rule tuning module: The data fitting module is used to collect the original sensing data of the heating device, construct the mapping relationship between the physical entity and the mathematical model, and output thermophysical parameters; The intelligent correction module is used to process the thermophysical parameters using an unscented Kalman filter algorithm and output an estimated core temperature value. The load identification and active suppression module is used to monitor the sample control output and temperature response, extract load mutation characteristics, and output feedforward compensation control quantity. The load identification and active suppression module includes a state observation unit and an active compensation unit; The state observation unit includes: Construct a linear extended state observer, define a state vector including the system output temperature and the rate of temperature change, introduce extended state variables, the initial values ​​of which are the sum of the rates of change of thermophysical parameters, and establish the state equation of the observer: ； in, The core temperature estimate is obtained by unscented Kalman filtering. To control the input, This is an estimate of the control gain. , , The gain coefficient of the observer. The system's output temperature, The rate of change of temperature, To expand the state variables, the state equations are iteratively calculated, and the expanded state variables are output in real time as the total disturbance estimate. The estimated value of the total disturbance is received and frequency domain filtered using a high-pass filter and a low-pass filter to obtain the load change characteristic signal; The active compensation unit includes: Real-time monitoring of load change characteristic signals, execution of threshold determination logic, and generation of feedforward compensation control quantity: If a negative load mutation characteristic signal is detected and its amplitude exceeds the first preset threshold, it is determined that the current system is connected to a heat-absorbing load, and a positive feedforward compensation control quantity is immediately generated through inversion calculation. If a positive load mutation characteristic signal is detected and its amplitude exceeds the second preset threshold, it is determined that an exothermic chemical reaction has occurred in the system, and a negative feedforward compensation control quantity is immediately generated. If the load mutation characteristic signal is between the first preset threshold and the second preset threshold, the system is determined to be in a steady state and outputs zero compensation. The feedforward compensation control quantity is directly superimposed on the output of the feedback control loop; The rule optimization module is used to calculate the deviation between the preset target temperature and the core temperature estimate, output the basic PID control quantity based on the variable universe fuzzy control theory, and superimpose the feedforward compensation control quantity and the basic PID control quantity to obtain the temperature drive control signal.

2. The intelligent temperature control system for sample pretreatment based on edge computing as described in claim 1, characterized in that, The data fitting module includes a data acquisition unit, a feature reconstruction unit, and a parameter identification unit; The data acquisition unit is used to synchronously acquire surface temperature data of the heating block, ambient temperature data and the drive control signal of the previous moment at a preset sampling frequency, and to perform moving average filtering on the data to obtain the original sensing data. The drive control signal of the previous moment includes the PWM duty cycle value, the heating power setting value, and the switching state. The PWM duty cycle value is used to represent the average electrical power applied to the heating block in the previous moment, the heating power setting value is used to represent the target output power set by the controller in the previous moment, and the switching state is used to represent the on / off state of the heating relay in the previous moment.

3. The intelligent temperature control system for sample pretreatment based on edge computing as described in claim 2, characterized in that, The feature reconstruction unit is used to perform maximum and minimum normalization processing on the original sensing data to construct the normalized input feature vector of the current time and the original sensing data. The annular buffer region is set to cover N time points. Raw sensor data is input into the annular buffer region. After each sampling cycle, the surface temperature data of the heating block, the ambient temperature data, and the drive control signal from the previous time point are stored in the buffer. The oldest data outside the time window is automatically discarded, ensuring the buffer always stores the most recent N time points. Interpolation calculations are performed on the raw sensor data to obtain the effective temperature rise variable. The expression for calculating the effective temperature rise variable is: ； in, For effective temperature increase variables, This is the surface temperature data of the heating block. This refers to ambient temperature data. The effective temperature rise variable and the original sensor data are subjected to max-min normalization processing, and an input feature vector is constructed based on the controlled autoregressive model structure. The input feature vector is... ； in, For the normalized effective temperature rise variables of the first i time steps, The original sensor data for the first i time steps after normalization. This is the input feature vector.

4. The intelligent temperature control system for sample pretreatment based on edge computing as described in claim 3, characterized in that, The parameter identification unit is used to perform iterative calculations on the input feature vector using a recursive least squares method with a forgetting factor, identify the equivalent thermal resistance and equivalent heat capacity in real time, and output the identified equivalent thermal resistance and equivalent heat capacity as thermophysical parameters. A controlled autoregressive model is constructed, defining a linear relationship equation between the effective temperature rise variable at the current moment and the input feature vector, and initializing the parameter vector to be identified and the covariance matrix. The linear relationship equation is as follows: ； in, The normalized effective temperature rise variable at the current moment. The parameter vector to be identified, For fitting residuals; In each sampling period, the Kalman gain vector is calculated using the covariance matrix of the previous time step and the current input feature vector. The prior prediction error is calculated, and the parameter estimates of the previous time step are corrected using this error and the Kalman gain vector. The coefficients of the controlled autoregressive model updated at the current time step are obtained, and the covariance matrix is ​​updated. The identified coefficients are mapped and inverted using the Euler approximation method to calculate the equivalent thermal resistance and equivalent heat capacity. The equivalent thermal resistance and equivalent heat capacity are output as thermophysical parameters.

5. The intelligent temperature control system for sample pretreatment based on edge computing as described in claim 4, characterized in that, The intelligent correction module includes a model building unit and a state update unit; The model building unit is used to receive thermophysical parameters and, in combination with preset sample thermophysical parameters, establish a second-order lumped parameter thermal network model containing surface temperature nodes of the heating block and core temperature nodes of the sample. Based on the current state estimate and covariance matrix, it generates a set of sampling points through unscented transformation.

6. The intelligent temperature control system for sample pretreatment based on edge computing as described in claim 5, characterized in that, The state update unit includes: The sampling point set is substituted into a second-order lumped parameter thermal network model for time update to obtain a prior estimate of the core temperature. The surface temperature data of the original sensor data is used as the observation value to calculate the Kalman gain matrix. The Kalman gain matrix is ​​used to correct the prior estimate of the core temperature and output the core temperature estimate.

7. The intelligent temperature control system for sample pretreatment based on edge computing as described in claim 6, characterized in that, The rule optimization module includes an error calculation unit and a control quantity calculation unit; The error calculation unit includes: Based on the core temperature estimate and the preset target temperature, calculate the temperature deviation and the rate of change of deviation, preset a deviation threshold, and compare the absolute value of the temperature deviation with the deviation threshold: When the absolute value of the temperature deviation is greater than or equal to the deviation threshold, the universe of discourse scaling factor is adjusted to 1. When the absolute value of the temperature deviation is less than the deviation threshold, the universe of discourse scaling factor is calculated using the exponential decay function. The formula for the exponential decay function is: ； in, The current domain scaling factor, The deviation threshold, This is the sensitivity adjustment coefficient. This is the minimum resolution protection value; Divide the current temperature deviation and deviation change rate by the current universe of discourse scaling factor to obtain the scaled fuzzy temperature deviation and fuzzy deviation change rate. Save the fuzzy temperature deviation and fuzzy deviation change rate as the first fuzzy value. Map the first fuzzy value to the preset fuzzy universe of discourse interval. Query the fuzzy rule base to obtain the proportional coefficient correction, integral coefficient correction and derivative coefficient correction of the PID controller.

8. The intelligent temperature control system for sample pretreatment based on edge computing as described in claim 7, characterized in that, The control quantity calculation unit includes: The proportional, integral, and derivative coefficients of the PID controller are added together to obtain the current PID parameters. The basic PID control quantity is calculated using the incremental PID algorithm. The temperature drive control signal is obtained by superimposing the basic PID control quantity with the feedforward compensation control quantity.

Images