Continuous flow dynamic titration calorimeter combining microfluidic control and machine learning correction and use method of continuous flow dynamic titration calorimeter

Abstract

The invention provides a continuous flow dynamic titration calorimeter combining microfluidics and machine learning correction and a use method thereof, and the device comprises a heat preservation device which is used for placing a calorimetric structure, carrying out the heat preservation of the calorimetric structure, collecting the environment temperature of the calorimetric structure, and sending the environment temperature to a control and signal processing system; the calorimetric structure comprises a material preheating aluminum block and a constant-temperature aluminum block which are arranged left and right, a material inlet and outlet connector block A is arranged in a front groove of the material preheating aluminum block, a material inlet and outlet connector block B is arranged in a rear groove of the material preheating aluminum block, and a thermoelectric power generation piece A is arranged on the front portion of the constant-temperature aluminum block. A temperature difference power generation piece B is arranged on the rear portion of the constant-temperature aluminum block, and a film heating piece is arranged above the temperature difference power generation piece A. According to the technical scheme, the micro-fluidic technology, temperature difference power generation measurement and machine learning correction are integrated, and the continuous flow dynamic titration calorimetry method which is low in cost, small in size and high in precision is constructed.

Description

Technical Field

[0001] The present invention relates to the technical field of reaction heat measurement, and more particularly, to a continuous flow dynamic titration calorimeter combined with microfluidics and machine learning correction and its use method. Background Art

[0002] Isothermal Titration Calorimetry (ITC), as a powerful analytical tool, is widely used in fields such as biochemistry, drug development, and materials science. It measures the heat released or absorbed when two substances are mixed at a constant temperature to obtain thermodynamic parameters of the reaction, such as binding constant, enthalpy change of reaction (ΔH), entropy change (ΔS), and stoichiometry. Although this method is extremely important in the field of experimental analysis, existing isothermal titration calorimeters face various limitations.

[0003] Traditional isothermal titration calorimeters are large in volume, complex in structure, and expensive. They usually require precise temperature control systems and highly sensitive heat detectors. For example, US Patent US9103782B2 describes a calorimeter with an automatically adjusted titration speed. Although it improves the accuracy of the titration process, its complex mechanical and electronic systems make the device expensive and difficult to maintain, and it is not suitable for environments with limited resources. Additionally, US8827549B2 shows a calorimeter with multi-channel functions. Although it increases the diversity of experiments, its complex operation and large volume make it unsuitable for scenarios that require fast and simple operation. And US10254239B2 focuses on using advanced algorithms to improve the speed and accuracy of data processing, but fails to solve the detection accuracy problem of low-concentration samples or weak heat effects.

[0004] Furthermore, existing technologies usually rely on an intermittent operation mode, that is, waiting for the system to reach thermal equilibrium after each titration before performing the next titration, which has limited applicability in continuous processes. In terms of measurement accuracy, due to factors such as environmental temperature fluctuations, baseline drift, and instrument noise, there may be large errors in the test results. US10337933B2 and US10429328B2 improve the measurement accuracy by improving the detection technology and temperature control system, but the problems of high cost and complex operation still exist. At the same time, the calorimeter with automated sample preparation and analysis introduced in US8449175B2 reduces the need for manual operation, but its high cost and dependence on technical personnel still limit its wide application.

[0005] In recent years, the development of microfluidic technology has provided new possibilities for solving these problems. Microfluidic chips are widely used in various analytical instruments due to their small sample volume, high mass transfer efficiency, and easy integration. For example, the microscale dynamic tracking reference continuous calorimetry device and method introduced in Chinese Patent CN116818831A can perform rapid heat effect detection in a continuous flow state. However, this method is mainly applicable to processes with relatively large heat release, and its ability to accurately capture small heat releases is limited.

[0006] In view of the limitations of existing isothermal titration calorimetry techniques in terms of cost, complexity, test efficiency, and accuracy, there is an urgent need to develop a new calorimetry measurement method that is low-cost, high-precision, easy to operate, and applicable to continuous flow conditions. Summary of the Invention

[0007] In view of the technical problem that microfluidic technology cannot measure small heat releases in isothermal titration calorimetry as mentioned above, a continuous flow dynamic titration calorimeter combined with microfluidics and machine learning correction and its usage method are provided. The present invention mainly constructs a low-cost, miniaturized, and high-precision continuous flow dynamic titration calorimetry method. This method can replace the existing isothermal titration calorimetry method and expand the application scope of calorimetry technology.

[0008] The technical means adopted by the present invention are as follows:

[0009] A continuous flow dynamic titration calorimeter combined with microfluidics and machine learning correction, comprising:

[0010] A heat preservation device for placing the calorimetry structure, insulating the calorimetry structure, collecting the ambient temperature of the calorimetry structure, and sending it to the control and signal processing system;

[0011] The calorimetry structure includes a material preheating aluminum block and a constant temperature aluminum block arranged left and right. A material inlet and outlet interface block A is arranged in the front groove of the material preheating aluminum block, and a material inlet and outlet interface block B is arranged in the rear groove of the material preheating aluminum block. A thermoelectric generator A is arranged in the front part of the constant temperature aluminum block, and a thermoelectric generator B is arranged in the rear part of the constant temperature aluminum block. A thin film heating sheet is arranged above the thermoelectric generator A, a microfluidic chip A is arranged above the thin film heating sheet, and a microfluidic chip B is arranged above the thermoelectric generator B;

[0012] The rear part of the constant temperature aluminum block is connected to a constant temperature oil inlet, and the front part of the constant temperature aluminum block is connected to a constant temperature oil outlet. The lower part of the material inlet and outlet interface block A is respectively connected to a material inlet A1, a material inlet A2, and a material outlet A, and the lower part of the material inlet and outlet interface block B is respectively connected to a material inlet B1, a material inlet B2, and a material outlet B;

[0013] The first syringe pump A, the second syringe pump A, and the waste liquid tank are connected to the microfluidic chip A after passing through the material preheating aluminum block and the material inlet and outlet interface block A; the first syringe pump B, the second syringe pump B, and the waste liquid tank are connected to the microfluidic chip B after passing through the material preheating aluminum block and the material inlet and outlet interface block B;

[0014] A control and signal processing system for controlling an editable DC power supply and a syringe pump controller; for collecting the signals generated by the thermoelectric generator and outputting the calorimetry results.

[0015] Further, the heat preservation device includes a calorimeter outer cover, a calorimeter outer base, and a thermocouple;

[0016] The calorimeter outer cover and the calorimeter outer base are arranged outside the calorimetry structure, the heat preservation felt upper cover is arranged outside the calorimeter outer cover, the heat preservation felt base is arranged outside the calorimeter outer base, and the thermocouple is arranged inside the calorimeter outer cover.

[0017] Further, silicone rubber sealing rings are arranged on the material inlet and outlet interface block A and the material inlet and outlet interface block B.

[0018] Further, material inlet and outlet visual windows are arranged above the material inlet and outlet interface block A and the material inlet and outlet interface block B, and microfluidic visual windows are arranged above the microfluidic chip A and the microfluidic chip B.

[0019] Further, the microfluidic chip A and the microfluidic chip B have the same structure, and the thermoelectric generator A and the thermoelectric generator B are of the same model.

[0020] Further, the control and signal processing system includes a central control and data acquisition system, a signal amplifier A, a signal amplifier B, a low-pass filter A, a low-pass filter B, an analog-to-digital converter, an editable DC power supply, and a syringe pump controller;

[0021] The positive and negative electrodes of the editable DC power supply are connected to the thin film heating sheet through wires. The signal generated by the thermoelectric generator A is sequentially connected to the signal amplifier A, the low-pass filter A, and the analog-to-digital converter through signal transmission lines; the signal generated by the thermoelectric generator B is sequentially connected to the signal amplifier B, the low-pass filter B, and the analog-to-digital converter through signal transmission lines; the analog-to-digital converter transmits the signal to the central control and data acquisition system through a signal transmission line; the central control and data acquisition system controls the editable DC power supply and the syringe pump controller through a communication line.

[0022] The present invention also provides a method for using a continuous flow dynamic titration calorimeter combining microfluidics and machine learning correction, which is realized based on any one of the continuous flow dynamic titration calorimeters combining microfluidics and machine learning correction, and includes the following steps:

[0023] S1. Continuously add materials into microfluidic chip A and microfluidic chip B at the same flow rate simultaneously; among them, in the test part, the material continuously introduced into the microchannel reference chip A by the first syringe pump A is the titrant in the test titration process, and the material continuously introduced into the microchannel reference chip A by the second syringe pump A is the solution to be titrated in the test titration process; in the reference part, the materials continuously introduced into the microfluidic chip B by the first syringe pump B and the second syringe pump B are both solutions to be titrated; the flow rates of the first syringe pump A and the first syringe pump B are the same, and the flow rates of the second syringe pump A and the second syringe pump B are the same;

[0024] S2. Control the target concentration of the solution in microfluidic chip A by controlling the flow rates of each syringe pump; simulate the titration process by adjusting the flow rate ratio of the syringe pumps, and introduce constant-temperature oil into the material preheating aluminum block and the constant-temperature aluminum block through the constant-temperature oil inlet, and the temperature of the constant-temperature oil is the set test temperature; the signal values generated by the thermoelectric power generation chip A and the thermoelectric power generation chip B are amplified 100 times respectively by the signal amplifier A and the signal amplifier B, and then the high-frequency noise is filtered out by the low-pass filter A and the low-pass filter B respectively;

[0025] S3. Collect the signal values of the two channels through the analog-to-digital converter and transmit them to the central control and data acquisition system;

[0026] S4. The central control and data acquisition system collects the temperature signals generated by the thermocouple, cleans the data of the signal values, and obtains an effective data set;

[0027] S41. Continuously collect the temperature signals generated by the thermocouple under the same test conditions through the central control and data acquisition system for 10 minutes, and store the collected temperature signals in the database, denoted as where T(k) represents the temperature signal value at the k-th sampling moment, and N is the total number of sampling points;

[0028] S42. Perform a fast Fourier transform on the collected temperature signal data to obtain the energy distribution of the signal in the frequency domain. Transforming the time-domain signal {T(k)} to the frequency domain can be expressed as:

[0029]

[0030] where ω is the frequency sampling point index (0 ≤ ω < N);

[0031] Remove the DC component: that is, set the component corresponding to ω = 0 (DC quantity) to 0; find the maximum peak: within the range of ω > 0, find the frequency component ω with the largest amplitude max , and convert it to the corresponding frequency where f sdenotes the signal sampling frequency; set the cut-off frequency: take 1.5×f max as the cut-off frequency of the Butterworth low-pass filter, denoted as f cut = 1.5·f max ;

[0032] S43. Filter the temperature signal {T(k)} using a Butterworth low-pass filter to remove high-frequency noise. The transfer function of the filter can be expressed as:

[0033]

[0034] where ω c = 2πf cut is the angular cut-off frequency of the filter. Map this transfer function to the z-domain through discretization to achieve digital filtering; the filtered data is denoted as {T flt1 (k)}, obtaining the preliminary filtering result;

[0035] S44. Based on the preliminary filtering result {T flt1 (k)}, further use two-dimensional Kalman filtering to estimate and correct the measurement signal; the system state model is:

[0036] x k = Ax k-1 + Bu k + w k

[0037] where x k represents the state vector at the k-th moment, u k is the optional control quantity vector, w k is the process noise, and A and B are the system state transition matrix and the control matrix respectively; the observation model is:

[0038] z k = Hx k + v k

[0039] where z k is the observed value, v k is the observation noise, and H is the observation matrix; in the Kalman filter recurrence equation, the prediction stage is:

[0040]

[0041] P k|k-1 = AP k|k-1 A T + Q

[0042] where P k|k-1 is the state covariance prediction, Q is the process noise covariance matrix, and the update stage is:

[0043] K k = P k|k-1 H T (HP k|k-1 H T + R) -1

[0044]

[0045] P k|k = (I - K k H)P k|k-1

[0046] where K k is the Kalman gain matrix, R is the observation noise covariance matrix, and finally the temperature signal {T flt2 (k)} corrected by two-dimensional Kalman filtering is obtained;

[0047] S45. Record the heat release power data of the thin film heating sheet as and expand it into a one-dimensional data sequence in chronological order to maintain the comparability of the heat release power at different time points; P(k) corresponds one-to-one with the sampling time T flt2 (k) of the temperature signal;

[0048] S46. Perform segmented analysis on {P(k)} using a window function; set the window width W = 60 and the window moving step size s = 2; statistically analyze the power data within the m-th window, and the data within the window is {P(k)|k ∈ [k start , k end}, calculate the first quartile Q1 and the third quartile Q3 within this window, where Q1 represents the value of the data at the 25% position after being arranged in ascending order, and Q3 represents the value of the data at the 75% position after being arranged in ascending order;

[0049] S47. Make the following judgment on the data {P(k)} within each window, and calculate the interquartile range IQR within this window:

[0050] IQR = Q3 - Q1

[0051] Judge whether the data is an outlier. If P(k) exceeds the interval [Q1 - 1.5 × IQR, Q3 + 1.5 × IQR], it is regarded as an outlier; discard all the marked abnormal heat release power values and the corresponding measurement signal data {T flt2 (k)} together, and only retain the normal data and its temperature signal pairs; through the above window function traversal and outlier marking, remove the extreme noise points that appear during the observation or acquisition process, so as to obtain the effective data set {(T flt2 (k), P(k))};

[0052] S5. Adopt a machine learning-based method to establish a meta-model based on the effective data set;

[0053] S6. Use the meta-model for reaction heat prediction.

[0054] Furthermore, S5 specifically includes the following steps:

[0055] S51. Respectively use three algorithms, namely XGBoost, LightGBM, and random forest, to perform modeling training on the training data, and perform parameter tuning through methods such as grid search, random search, or Bayesian optimization to obtain a relatively stable parameter configuration, and then form three groups of base models; the objective function of XGBoost is:

[0056]

[0057] where y i represents the true label, is the cumulative prediction of the first m - 1 trees, f m represents the residual function learned by the mth tree, and Ω(·) is the tree complexity regularization term;

[0058] For the mth tree, the output of the LightGBM model can be approximately expressed as:

[0059] F m (x) = F m-1 (x) + α m T m (x)

[0060] where α m is the learning rate of the mth tree, and T m (x) represents the estimation of the mth tree for the input x; the random forest obtains the final prediction output by constructing K CART decision trees and voting or taking the average, and the model can be expressed as:

[0061]

[0062] where T k (x) is the output of the kth decision tree;

[0063] S52. After completing parameter tuning for XGBoost, LightGBM, and random forest respectively, use the finally locked optimal hyperparameter combination to train and obtain three groups of base models, denoted as:

[0064] M XGB , M LGB , M RF

[0065] S53. Input the same training set or validation set data into the three groups of base models in sequence to obtain three groups of predicted values:

[0066]

[0067] Among them, X represents the input feature matrix, respectively represent the vectors of heat release prediction values output by XGBoost, LightGBM, and random forest; compare the above vectors of heat release prediction values with the corresponding true value Y to obtain the error and residual information generated by each of the three models;

[0068] S54. Use the true value Y as the training label of the meta-model, and concatenate or combine them into a new feature vector, denoted as:

[0069]

[0070] S55. Adopt support vector machine regression as the meta-model, denoted as M SVM ; its objective function can adopt the ∈-insensitive loss form:

[0071]

[0072] Among them, w and b are the weight vector and bias term of the support vector machine respectively, C is the regularization coefficient, and ξ i represents the slack variable exceeding the ∈-insensitive interval; the kernel function K(x i , x j ) can select the RBF kernel, linear kernel or polynomial kernel form according to actual needs;

[0073] S56. Use cross-validation, holdout validation or other validation techniques to adjust and optimize the parameters of the SVM. The parameters include the kernel function type, regularization coefficient C, and kernel function parameter γ; specifically, it can be expressed as:

[0074]

[0075] Among them, CVLoss represents the comprehensive loss value obtained by measuring and averaging the prediction error during cross-validation; through this process, the meta-model fully learns the mapping relationship between the output of the base model and the true label;

[0076] S57. When the prediction error of the meta-model on the validation set meets the preset accuracy requirement, complete the parameter determination work of the meta-model to obtain the final meta-model

[0077] Furthermore, when the environmental temperature during actual measurement exceeds the temperature range of the machine learning training set, adopt a method of improving the robustness of the model output in the extrapolation scenario through linear mapping to map the temperature outside the temperature range to the temperature range. The steps are as follows:

[0078] P1. Based on the correspondence between temperature T and measurement signal M in the existing training dataset, use the least squares method or other linear regression methods to obtain its linear fitting model, denoted as:

[0079] M = k·T + b

[0080] where M represents the measurement signal, and k and b are the slope and intercept of the linear model respectively, and the formulas are as follows:

[0081]

[0082] where T i and M i are the temperature and measurement signal of the i-th sample in the training set, and are the average values of the temperature and measurement signal respectively;

[0083] P2. Set the temperature range of the training data as [T min , T max , and this range can be regarded as the core area where the prediction reliability of the model in the temperature dimension is relatively high; store {k, b} and T min , T max in the parameter library for subsequent calls;

[0084] P3. For the new test dataset where is the temperature value of the j-th test sample, represents other feature vectors or measurement signals of this sample. First, it is necessary to judge whether falls within the range of [T min , T max ; if then it is regarded as normal temperature and no correction is required; if or then it is marked as a sample with out-of-range temperature and recorded in the set O;

[0085] P4. For the samples marked as out-of-range call the linear relationship to correct their corresponding measurement signals; assume the original value of the measurement signal of the out-of-range sample is then the mapping correction formula is:

[0086]

[0087] where represents the corrected measurement signal, and together with ensures that the new test sample is closer to the training interval [T min , T maxThe corresponding feature distribution;

[0088] P5. Replace all the test samples after linear correction with the original out-of-range data to form a new test data set At this time, the existing meta-model can be used to predict the updated test data.

[0089] Further, the specific steps of S6 are as follows:

[0090] S61. Determine whether the process is an endothermic process or an exothermic process. When it is determined that the process is an exothermic process, execute S62 - S65; when it is determined that the process is an exothermic process, execute S66, S67, S63, S64, and S65 in sequence;

[0091] S62. Input the measurement signal that has been pre-processed by collecting, filtering, and removing outliers in the early stage into the machine learning model, and the meta-model outputs the predicted value of the process exothermic power, and record this predicted power as P 预测 = P;

[0092] S63. Substitute the predicted exothermic power value P output by the machine learning model into the following formula to calculate the process heat release Q:

[0093]

[0094] Among them, V is the liquid storage volume in the microfluidic chip, indicating the volume of the liquid stored in the chip during the titration process or other test processes; q is the flow rate of the material in the microfluidic channel, usually set and monitored through a microfluidic pump or other precision flow rate control devices; c is the concentration of the target substance in the test material, which is used for the subsequent calculation of the reaction heat ΔH;

[0095] S64. Given that the target concentration of a part of the test material is c, under the condition that the volume of the microfluidic chip is V, the total feed amount or the reactant content can be characterized; convert μL to L;

[0096] S65. Calculate the reaction heat ΔH according to the measured heat release Q and the concentration c:

[0097]

[0098] S66. If it is determined that the process is an endothermic process, an additional exothermic power P needs to be provided to the thin film heating sheet through an editable DC power supply 补偿 and ensure that this compensation value can cover the power absorbed by the endothermic process to ensure that the entire system is in the desired temperature or energy balance state;

[0099] S67. Calculate the actually required set power of the heating sheet from the predicted endothermic power P and the compensation power P 补偿 Find out:

[0100] P = P 补偿 -P 预测

[0101] where P 预测 is the magnitude of the endothermic power predicted by the machine learning model; P 补偿 needs to be greater than |P 预测 | to meet the energy compensation for the external endothermic process.

[0102] Compared with the prior art, the present invention has the following advantages:

[0103] 1. Cost reduction and structure simplification: The present invention uses a microfluidic chip, a thermoelectric power generation chip, and a signal amplifier as core components, greatly reducing the manufacturing cost and volume of the instrument. The detachable design of the microfluidic chip enables it to flexibly adjust the mixing structure according to actual needs, facilitating later maintenance and replacement. This makes the calorimeter more suitable for use in laboratories or on-site environments with limited resources, broadening its application scope.

[0104] 2. Measurement accuracy improvement and error reduction: Aiming at the complex error factors (such as flow rate, test temperature, ambient temperature, signal values at the reference end and the test end, etc.) that affect accuracy during the test process, the present invention introduces a machine learning method and establishes an accurate calibration model. Through the accurate calibration of the calorimeter, the test error is significantly reduced, and the detection accuracy is improved. Especially in the detection of low-concentration samples and weak thermal effects, the problem of difficult accurate measurement by traditional methods is solved.

[0105] 3. Dynamic titration under continuous flow conditions: The microfluidic chip is used as the carrier for continuous flow mixing of materials, combined with a high-precision injection pump to accurately adjust the flow rate ratio, and the titration process is dynamically simulated under continuous flow conditions. The present invention can obtain the thermodynamic parameters of substances in a continuous state, making up for the deficiencies of traditional batch methods and meeting the requirements of specific application scenarios such as continuous process control and real-time monitoring.

[0106] 4. Simple operation, high efficiency and speed: The present invention simplifies the detection process through direct energy conversion and measurement, reducing the dependence on complex mechanical and electronic systems. Without a precise temperature control system and a highly sensitive traditional heat detector, the requirements for professional operators are reduced, and the test efficiency is improved, making it suitable for scenarios that require fast and simple operation.

[0107] 5. Expand the application scope of calorimetry technology: Through the series processing of a signal amplifier and a low-pass filter, the present invention converts an originally extremely weak thermal signal into available information with a high signal-to-noise ratio. On the one hand, the signal amplifier can significantly amplify an extremely small thermal signal, avoiding the loss of effective information caused by insufficient quantization accuracy during subsequent digitization. On the other hand, the low-pass filter removes high-frequency noise or jitter from the amplified signal, making the signal input to the analog-to-digital converter more pure and stable, thereby greatly improving the accuracy and resolution of micro-thermal change detection. By creatively applying machine learning technology and linear mapping correction methods, high prediction accuracy and stability can still be maintained when extrapolating the model and the temperature deviates from the training interval. The overall device combines a microfluidic chip and a thermoelectric power generation chip, with the advantages of a compact structure, miniaturization, and low energy consumption. At the same time, combined with an intelligent machine learning algorithm, it can flexibly adapt to various types of chemical systems or reaction conditions, and has wide applicability and expandable value for various titration calorimetry requirements.

[0108] 6. A thermal insulation device provided by the present invention is used to place a calorimetry structure and insulate it; the thermal insulation device can also collect the ambient temperature of the calorimetry structure and send the collected temperature signal to the control and signal processing system in real time. By placing the calorimetry structure in the thermal insulation device, the interference caused by external environmental temperature fluctuations to the main body of the calorimetry device can be effectively reduced, thereby significantly improving the accuracy and stability of micro-calorimetric change measurement.

[0109] In summary, by combining microfluidic technology, thermoelectric power generation measurement, and machine learning correction, the present invention constructs an innovative continuous flow dynamic titration calorimetry method. This method not only effectively replaces the existing isothermal titration calorimetry technology, but also comprehensively improves in terms of cost, accuracy, efficiency, and applicability, and has significant practical application value and promotion prospects. Brief Description of the Drawings

[0110] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0111] Figure 1 It is a schematic diagram of the structure of the calorimeter of the present invention.

[0112] Figure 2 It is an exploded view of the calorimeter of the present invention.

[0113] Figure 3 It is a calorimetry structure diagram of the present invention.

[0114] Figure 4 This is a diagram of the control and signal processing system of the present invention.

[0115] Figure 5 This is a flowchart of the signal value processing of the present invention.

[0116] Figure 6 This is a diagram of the machine learning assisted calorimetric correction of the present invention.

[0117] Figure 7 This is a distribution diagram of the test error of the calorimeter of the present invention.

[0118] Figure 8 This is a diagram of the fitting process of the embodiment of the present invention.

[0119] Figure 9 This is a data plot of Embodiment 1 of the present invention.

[0120] Figure 10 This is a data plot of Embodiment 2 of the present invention.

[0121] Figure 11 This is a data plot of Embodiment 3 of the present invention.

[0122] Figure 12 This is a data plot of Embodiment 4 of the present invention.

[0123] In the figure: 11, upper heat insulation felt cover; 12, heat insulation felt base; 21, upper calorimeter housing cover; 22, calorimeter housing base; 3, thermocouple; 411, material inlet and outlet viewing window; 412, microfluidic viewing window; 421, microfluidic chip A; 422, microfluidic chip B; 43, thin film heating sheet; 441, thermoelectric power generation sheet A; 442, thermoelectric power generation sheet B; 451, material inlet and outlet interface block A; 452, material inlet and outlet interface block B; 453, silicone rubber sealing ring; 461, material preheating aluminum block; 462, constant temperature aluminum block; 463, constant temperature oil inlet; 464, constant temperature oil outlet; 471, material inlet A1; 472, material inlet A2; 473, material outlet A; 474, material inlet B1; 475, material inlet B2; 476, material outlet B; 481, first injection pump A; 482, second injection pump A; 483, first injection pump B; 484, second injection pump B; 49, waste liquid tank; 51, central control and data acquisition system; 521, signal amplifier A; 522, signal amplifier B; 531, low-pass filter A; 532, low-pass filter B; 54, analog-to-digital converter; 55, programmable DC power supply; 56, injection pump controller. Detailed implementation manners

[0124] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the drawings and in combination with the embodiments.

[0125] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part rather than all of the embodiments of the present invention. The description of at least one exemplary embodiment is actually only illustrative and in no way restricts the present invention and its application or use. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0126] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments of the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0127] Unless otherwise specifically stated, the relative arrangements of components and steps, numerical expressions and values set forth in these embodiments do not limit the scope of the present invention. At the same time, it should be clear that, for the convenience of description, the dimensions of the various parts shown in the drawings are not drawn in actual proportional relationships. Technologies, methods and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, the said technologies, methods and devices should be regarded as part of the authorized specification. In all the examples shown and discussed herein, any specific values should be construed as merely exemplary and not as a limitation. Therefore, other examples of the exemplary embodiments may have different values. It should be noted that like reference numerals and letters denote like items in the following drawings, and thus, once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0128] In the description of the present invention, it should be understood that the orientation or positional relationships indicated by orientation words such as "front, rear, upper, lower, left, right", "lateral, vertical, perpendicular, horizontal" and "top, bottom" are generally based on the orientation or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description. Without contrary description, these orientation words do not indicate and imply that the device or element referred to must have a specific orientation or be constructed and operated in a specific orientation, and thus should not be construed as limiting the scope of protection of the present invention. The orientation words "inside, outside" refer to the inside and outside relative to the contour of each component itself.

[0129] For ease of description, spatial relative terms such as "above", "over", "on the upper surface", "upper" etc. can be used here to describe the spatial positional relationship between a device or feature shown in the figure and other devices or features. It should be understood that the spatial relative terms are intended to encompass different orientations in use or operation in addition to the orientation depicted in the figure for the device. For example, if the device in the figure is inverted, the device described as "above" or "over" other devices or structures will then be positioned "below" or "under" the other devices or structures. Thus, the exemplary term "above" can include both the orientations of "above" and "below". The device can also be positioned in other different ways (rotated 90 degrees or in other orientations), and the corresponding explanations for the spatial relative descriptions used here will be made accordingly.

[0130] In addition, it should be noted that the use of terms such as "first", "second" etc. to define components is only for the convenience of differentiating the corresponding components. Without further statement, the above terms have no special meaning, and thus should not be construed as limiting the scope of protection of the present invention.

[0131] As Figures 1-4 shown, the present invention provides a continuous flow dynamic titration calorimeter combining microfluidics and machine learning correction, comprising:

[0132] A heat preservation device for placing the calorimetric structure, insulating the calorimetric structure and collecting the ambient temperature of the calorimetric structure and sending it to the control and signal processing system; the heat preservation device includes the calorimeter outer shell upper cover 21, the calorimeter outer shell base 22 and the thermocouple 3;

[0133] The calorimeter outer shell upper cover 21 and the calorimeter outer shell base 22 are arranged outside the calorimetric structure, the heat preservation felt upper cover 11 is arranged outside the calorimeter outer shell upper cover 21, the heat preservation felt base 12 is arranged outside the calorimeter outer shell base 22, and the thermocouple 3 is arranged inside the calorimeter outer shell upper cover 21.

[0134] Calorimetric structure, including a preheating aluminum block 461 for materials and a constant-temperature aluminum block 462 arranged left and right. A material inlet and outlet interface block A451 is arranged in the front slot of the preheating aluminum block 461 for materials, and a material inlet and outlet interface block B452 is arranged in the rear slot of the preheating aluminum block 461 for materials. A thermoelectric generator A441 is arranged at the front part of the constant-temperature aluminum block 462, and a thermoelectric generator B442 is arranged at the rear part of the constant-temperature aluminum block 462. A thin-film heating sheet 43 is arranged above the thermoelectric generator A441, and a microfluidic chip A421 is arranged above the thin-film heating sheet 43. A microfluidic chip B422 is arranged above the thermoelectric generator B442; the left end is the reference part, and the right end is the test part. Silicone rubber sealing rings 453 are arranged on the material inlet and outlet interface block A451 and the material inlet and outlet interface block B452. A material inlet and outlet viewing window 411 is arranged above the material inlet and outlet interface block A451 and the material inlet and outlet interface block B452, and a microfluidic viewing window 412 is arranged above the microfluidic chip A421 and the microfluidic chip B422. The microfluidic chip A421 and the microfluidic chip B422 have the same structure, and the thermoelectric generator A441 and the thermoelectric generator B442 are of the same model.

[0135] A constant-temperature oil inlet 463 is connected to the rear part of the constant-temperature aluminum block 462, and a constant-temperature oil outlet 464 is connected to the front part of the constant-temperature aluminum block 462. The lower part of the material inlet and outlet interface block A451 is respectively connected to a material inlet A1471, a material inlet A2472, and a material outlet A473. The lower part of the material inlet and outlet interface block B452 is respectively connected to a material inlet B1474, a material inlet B2475, and a material outlet B476;

[0136] The first syringe pump A481, the second syringe pump A482, and the waste liquid tank 49 are connected to the microfluidic chip A421 after passing through the preheating aluminum block 461 for materials and the material inlet and outlet interface block A451; the first syringe pump B483, the second syringe pump B484, and the waste liquid tank 49 are connected to the microfluidic chip B422 after passing through the preheating aluminum block 461 for materials and the material inlet and outlet interface block B452;

[0137] A control and signal processing system is used to control an editable DC power supply 55 and a syringe pump controller 56; it is used to collect the signals generated by the thermoelectric generators and output calorimetric results.

[0138] The control and signal processing system includes a central control and data acquisition system 51, a signal amplifier A521, a signal amplifier B522, a low-pass filter A531, a low-pass filter B532, an analog-to-digital converter 54, an editable DC power supply 55, and a syringe pump controller 56;

[0139] The positive and negative electrodes of the editable DC power supply 55 are connected to the thin-film heating sheet 43 through wires. The signals generated by the thermoelectric power generation sheet A 441 are sequentially connected to the signal amplifier A 521, the low-pass filter A 531, and the analog-to-digital converter 54 through signal transmission lines. The signals generated by the thermoelectric power generation sheet B 442 are sequentially connected to the signal amplifier B 522, the low-pass filter B 532, and the analog-to-digital converter 54 through signal transmission lines. The analog-to-digital converter 54 transmits the signals to the central control and data acquisition system 51 through signal transmission lines. The central control and data acquisition system 51 controls the editable DC power supply 55 and the injection pump controller 56 through communication lines.

[0140] The present invention also provides a method for using a continuous flow dynamic titration calorimeter combining microfluidics and machine learning correction, including:

[0141] S1. The structures of the microfluidic chip A and the microfluidic chip B are exactly the same. Materials are continuously added to the microfluidic chip A and the microfluidic chip B at the same flow rate simultaneously. Among them, the materials continuously introduced into the microchannel reference chip A through the first injection pump A and the second injection pump A in the test part are the titrant and the solution to be titrated in the titration process respectively. The materials continuously introduced into the microfluidic chip B through the first injection pump B and the second injection pump B in the reference part are both the solution to be titrated. The flow rates of the first injection pump A and the first injection pump B are the same, and the flow rates of the second injection pump A and the second injection pump B are the same.

[0142] S2. Control the target concentration of the solution in the microfluidic chip A by controlling the flow rates of each injection pump. Simulate the titration process by adjusting the injection pump flow rate ratio. Pass the constant-temperature oil into the material preheating aluminum block and the constant-temperature aluminum block through the constant-temperature oil inlet 463. The temperature of the constant-temperature oil is the set test temperature. As Figure 5 shown, the signal values generated by the thermoelectric power generation sheet A and the thermoelectric power generation sheet B are amplified by 100 times by the signal amplifier A and the signal amplifier B respectively, and then the high-frequency noise is filtered out by the low-pass filter A and the low-pass filter B respectively.

[0143] S3. Then collect the signal values of the two channels through the analog-to-digital converter and transmit them to the central control and data acquisition system.

[0144] S4. At the same time, the central control and data acquisition system collects the temperature signals generated by the thermocouple; clean the data of the signal values to obtain an effective data set;

[0145] S41. Continuously collect the temperature signals generated by the thermocouple by the central control and data acquisition system under the same test conditions for 10 minutes, and store the collected temperature signals in the database, denoted as Among them, T(k) represents the temperature signal value at the k-th sampling moment, and N is the total number of sampling points. To ensure the consistency of test conditions, this step can be carried out multiple times under the same ambient temperature and the same operation process, so as to obtain several groups of temperature signal data for subsequent analysis.

[0146] S42. Perform a fast Fourier transform (FFT) on the collected temperature signal data to obtain the energy distribution of the signal in the frequency domain. Specifically, the transformation of the time-domain signal {T(k)} to the frequency domain can be expressed as:

[0147]

[0148] where ω is the frequency sampling point index (0 ≤ ω < N). Remove the DC component: that is, set the component corresponding to ω = 0 (DC quantity) to 0; find the maximum peak: within the range of ω > 0, find the frequency component ω with the largest amplitude max , and convert it to the corresponding frequency where f s represents the signal sampling frequency; set the cut-off frequency: take 1.5×f max as the cut-off frequency of the Butterworth low-pass filter, denoted as f cut = 1.5·f max .

[0149] S43. Use a Butterworth low-pass filter to filter the temperature signal {T(k)} to remove high-frequency noise. The transfer function of the filter can be expressed as:

[0150]

[0151] where ω c = 2πf cut is the angular cut-off frequency of the filter. In this invention, the cut-off frequency is 0.5. Through discretization (such as bilinear transformation), this transfer function can be mapped to the z-domain to achieve digital filtering. The filtered data is denoted as {T flt1 (k)}, and the preliminary filtering result is obtained.

[0152] S44. Based on the preliminary filtering result {T flt1 (k)}, further use two-dimensional Kalman filtering to estimate and correct the measurement signal. Two-dimensional Kalman filtering jointly estimates the signal in the time dimension and an additional spatial or parameter dimension, which can effectively reduce the influence of system noise and observation noise, so as to obtain a measurement signal closer to the true value. The system state model is:

[0153] x k = Ax k-1 + Bu k + w k

[0154] Among them, x k represents the state vector at the k-th moment (which may include temperature values and additional states), u k is the optional control quantity vector, w k is the process noise, and a and B are the system state transition matrix and the control matrix respectively. The observation model is:

[0155] z k = Hx k + v k

[0156] Among them, z k is the observed value (here it refers to the filtered temperature measurement value {T flt1 (k)} and the corresponding spatial / parameter information), v k is the observation noise, and H is the observation matrix. In the Kalman filter recurrence equation, the prediction stage is:

[0157]

[0158] P k|k-1 = AP k|k-1 A T + Q

[0159] Among them, P k|k-1 is the state covariance prediction, and Q is the process noise covariance matrix. The update stage is:

[0160] K k = P k|k-1 H T (HP k|k-1 H T + R) -1

[0161]

[0162] P k|k = (I - K k H)P k|k-1

[0163] Among them, K k is the Kalman gain matrix, and R is the observation noise covariance matrix. Finally, the temperature signal {T flt2 (k)} corrected by two-dimensional Kalman filter is obtained.

[0164] S45. Record the heat release power data of the thin film heating sheet as and expand it into a one-dimensional data sequence in chronological order to maintain the comparability of the heat release power at different time points. Here, P(k) and the sampling moment T flt2 (k) of the temperature signal correspond one by one.

[0165] S46. Segment and analyze {P(k)} using a window function. Set the window width W = 60 and the window moving step size s = 2. Statistically analyze the power data within the m-th window, and the data within the window is {P(k)|k ∈ [k start ,k end}. Calculate the first quartile Q1 and the third quartile Q3 within this window, where Q1 represents the value of the data at the 25% position after being sorted in ascending order, and Q3 represents the value of the data at the 75% position after being sorted in ascending order.

[0166] S47. Make the following judgments on the data {P(k)} within each window. Calculate the interquartile range (IQR) within this window:

[0167] IQR = Q3 - Q1

[0168] Judge whether the data is an outlier. If P(k) exceeds the interval [Q1 - 1.5×IQR, Q3 + 1.5×IQR], it is regarded as an outlier. Eliminate all the marked abnormal heat release power values and the corresponding measurement signal data {T flt2 (k)} together, and only retain the normal data and its temperature signal pairs. Through the above window function traversal and outlier marking, extreme noise points that appear during observation or acquisition can be effectively removed, thereby obtaining a new effective data set {(T flt2 (k), P(k))}. This data set contains the signal values of both the reference part and the test part, as well as information such as the ambient temperature of the device, providing a more accurate and clean basis for subsequent feature extraction and modeling.

[0169] During each test, signal values are continuously collected for 10 minutes under the same conditions. The collected data is further processed through the central control and data acquisition system. First, use the interquartile range test method (IQR) to eliminate outliers, and eliminate the data of the first 5 minutes. Perform Kalman filtering on the stable signal values of the last 5 minutes to further eliminate noise interference. Then use the interquartile range test method to eliminate outliers, and take the average value of the data of the last 2 minutes. In this way, the signal values of the reference part and the test part under this test condition, as well as the ambient temperature of the device, are obtained.

[0170] S5. Adopt a machine learning-based method to establish a meta-model based on the effective data set;

[0171] S51. After completing data preprocessing, three algorithms, namely XGBoost, LightGBM, and Random Forest, are used to perform modeling training on the training data, and parameter tuning (hereinafter referred to as "tuning") is carried out through methods such as grid search, random search, or Bayesian optimization to obtain a relatively stable parameter configuration, thereby forming three groups of base models. XGBoost is an efficient implementation based on the gradient-boosted decision tree (GBDT) framework, and its objective function can be expressed as:

[0172]

[0173] where y i represents the true label, is the cumulative prediction of the first m - 1 trees, f m represents the residual function learned by the m-th tree, and Ω(·) is the tree complexity regularization term. By iteratively training multiple decision trees in sequence, XGBoost can effectively capture the non-linear structure in the data. LightGBM is also based on the GBDT framework, but it adopts an efficient data structure based on gradient histograms and an incremental learning strategy based on Leaf-wise. For the m-th tree, the output of the LightGBM model can be approximately expressed as:

[0174] F m (x) = F m-1 (x) + α m T m (x)

[0175] where α m is the learning rate of the m-th tree, and T m (x) represents the estimation of the m-th tree for the input x. Random Forest obtains the final prediction output by constructing K CART decision trees and performing voting or taking the average (usually taking the mean in the regression scenario), and the model can be expressed as:

[0176]

[0177] where T k (x) is the output of the k-th decision tree. Since each decision tree is independent and the training data undergoes random sampling and random feature selection, Random Forest has strong generalization ability and anti-overfitting ability.

[0178] S52. After completing the tuning of XGBoost, LightGBM, and Random Forest respectively, the finally locked best hyperparameter combinations are used to train three groups of base models, denoted as:

[0179] M XGB , M LGB , M RF

[0180] S53. Input the same training set (or validation set) data into the three base models in sequence to obtain three sets of predicted values:

[0181]

[0182] where X represents the input feature matrix, respectively represent the predicted heat release value vectors output by XGBoost, LightGBM, and Random Forest. By comparing the above predicted values with the corresponding true value Y, the error and residual information generated by each of the three models can be obtained, providing a basis for subsequent meta-model training.

[0183] In order to fully explore the differences and complementarities in data feature learning among different base models, the present invention re-learns and fuses the three sets of prediction results by constructing a meta-model. This meta-model can, to a certain extent, capture the biases, residual distributions, and learning preferences of the three base models for different input features, thereby further improving the overall prediction accuracy and stability.

[0184] S54. Use the true value Y as the training label (true value) of the meta-model, and concatenate or combine them into a new feature vector, denoted as:

[0185]

[0186] S55. Adopt support vector machine regression (SVM Regression) as the meta-model, denoted as M SVM . Its objective function can adopt the ∈-insensitive loss form:

[0187]

[0188] In this formula, w and b are the weight vector and bias term of the support vector machine respectively, C is the regularization coefficient, and ξ i represents the slack variable exceeding the ∈-insensitive interval. The kernel function K(x i , x j ) can select forms such as RBF kernel, linear kernel, or polynomial kernel according to actual needs.

[0189] S56. Use cross-validation, holdout validation, or other validation techniques to adjust and optimize the parameters of the SVM (including kernel function type, regularization coefficient C, kernel function parameter γ, etc.), which can be specifically expressed as:

[0190]

[0191] Among them, CVLoss represents the comprehensive loss value obtained by measuring and averaging the prediction error (such as mean squared error MSE) during the cross-validation process (such as K-fold cross-validation). Through this process, the meta-model fully learns the mapping relationship between the output of the base model and the true label.

[0192] S57. When the prediction error of the meta-model on the validation set meets the preset accuracy requirement (such as MSE ≤ δ), the parameter determination of the meta-model is completed, and the final meta-model is obtained. This meta-model performs function approximation on the outputs of three groups of base models in the high-dimensional feature space, which can effectively improve the accuracy and stability of the overall prediction.

[0193] Through the secondary learning of the meta-model on the outputs of the three base models, the differences and complementarities in local features and residual information of different models are fully captured, improving the overall prediction accuracy. The three base models learn for different feature distributions or training preferences, and the meta-model fuses them on this basis, thus significantly enhancing the adaptability of the system to diverse scenarios.

[0194] The present invention aims to address the problem of limited temperature range of training data, and proposes a method to improve the robustness of the model output in the extrapolation scenario through linear mapping. In traditional modeling, when the temperature value of the test data exceeds the training temperature range, the model is prone to overfitting or mismatch, resulting in obvious deviations in the prediction results. Therefore, the present invention constructs a linear fitting model of temperature and measurement signal based on the existing training data in advance, and uses this linear relationship as a "mapping function" to correct the test samples with out-of-range temperature, thereby effectively improving the stability and accuracy of the model during extrapolation.

[0195] P1. Based on the correspondence between temperature T and measurement signal M in the existing training dataset, use the least squares method or other linear regression methods to obtain its linear fitting model, denoted as

[0196] M = k·T + b

[0197] Among them, M represents the measurement signal (or other relevant parameters can be used), and k and b are the slope and intercept of the linear model respectively, given by the following formula:

[0198]

[0199] Among them, T i and M i are the temperature and measurement signal of the i-th sample in the training set, and are the average values of temperature and measurement signal respectively.

[0200] P2. Set the temperature range of the training data as [T min , Tmax , this interval can be regarded as the core area where the model has a relatively high prediction reliability in the temperature dimension. {k, b} and T min , T max are stored in the parameter library for subsequent calls.

[0201] P3. For the new test data set where is the temperature value of the j-th test sample, represents other feature vectors or measurement signals of this sample. First, it is necessary to judge whether it falls within the range of [T min , T max . Specifically, if then it is regarded as normal temperature and no correction is required; if or then it is marked as a "sample with out-of-range temperature" and recorded in the set O.

[0202] P4. For the samples marked as out of range call the previously recorded linear relationship to correct their corresponding measurement signals. Let the original value of the measurement signal of the out-of-range sample be (if the measurement signal is not yet clear, it can be obtained from the existing model or sensor readings), then the mapping correction formula is

[0203]

[0204] where represents the corrected measurement signal. Cooperating with can ensure that the new test sample is closer to the feature distribution corresponding to the training interval [T min , T max .

[0205] P5. Replace all the linearly corrected test samples with the original out-of-range data to form a new test data set At this time, the existing prediction model can be used to predict the updated test data, so as to maintain a relatively high prediction accuracy and stability in the temperature extrapolation scenario.

[0206] When the temperature of the test data falls outside the training range, the measured signal is corrected through a linear fitting relationship, which can significantly reduce the prediction fluctuations and mismatches of the model in the temperature extrapolation region, thereby improving the prediction stability and accuracy. The linear mapping required by the present invention only needs to be calculated once and can be corrected by {k, b} rows of simple operations, which can greatly reduce the cost in terms of algorithm complexity and time overhead. This linear mapping strategy can be combined with various models (such as tree-based methods, neural networks, linear regression, etc.) and can also be extended to other extrapolation variable scenarios (such as pressure, humidity, etc.), with strong versatility and expandable value.

[0207] S6. Use the meta-model to predict the heat of reaction.

[0208] S61. Input the measured signal that has been pre-processed by data collection, filtering, outlier removal, etc. in the early stage into the machine learning model, and the model outputs the predicted value of the process heat release power. Denote this predicted power as P. 预 Measure.

[0209] S62. Substitute the predicted value P of the heat release power output by the machine learning model 预测 into the following formula to calculate the process heat release Q,

[0210]

[0211] where V is the liquid storage volume (μL) in the microfluidic chip, representing the volume of the liquid stored in the chip during the titration process or other test processes; q is the flow rate (μL / min) of the material in the microfluidic channel, usually set and monitored by a microfluidic pump or other precision flow rate control devices; c: the concentration (mol / L) of the target substance in the test material, which is used for the subsequent calculation of the heat of reaction ΔH.

[0212] S63. Given that the target concentration of the test part of the material is c, under the condition that the volume of the microfluidic chip is V, the total feed amount or the reactant content can be characterized. To facilitate the calculation of the chemical heat effect related to energy, it is usually necessary to convert μL to L.

[0213] S64. Calculate the heat of reaction ΔH based on the measured heat release Q and concentration c.

[0214]

[0215] S65. If it is determined that the process is an endothermic process, an additional heat release power P needs to be provided to the thin film heating sheet through an editable DC power supply 补偿 , and ensure that this compensation value can cover the power absorbed during the endothermic process to ensure that the entire system is in the desired temperature or energy balance state.

[0216] S66. From the predicted endothermic power P and the compensation power P 补偿Calculate the actual required power of the heating sheet:

[0217] P = P 补偿 - P 预测

[0218] Wherein, P 预测 is the magnitude of the endothermic power predicted by the machine learning model; P 补偿 needs to be greater than |P 预测 | to meet the energy compensation for the external endothermic process.

[0219] Whether the reaction is exothermic or endothermic, the dynamic balance can be achieved by calculating the exothermic power P 预测 and matching with the DC power supply power compensation measures, ensuring that the measurement process is more stable and accurate. With the prediction ability of the machine learning model and the precise controllability of the microfluidic system in terms of flow rate, volume, etc., the resolution and robustness of the exothermic / endothermic heat measurement can be significantly improved, while meeting the experimental requirements of various types of chemical systems or materials. The method of the present invention is not only applicable to the test scenario of the titration calorimeter, but also can be extended to other continuous flow reaction systems; it also has good adaptability to different flow conditions, different temperature or pressure environments.

[0220] In order to correct the signal value, the present invention adopts a machine learning-based method to precisely calibrate the calorimeter.

[0221] Specifically, the present invention controls the heating power of the thin film heating sheet through an adjustable DC power supply to simulate the exothermic situation of the test part in the microfluidic chip A. During the calibration stage, the substances injected into the two chips are both water to obtain the signal values under different exothermic powers. First, as Figure 6 shown, the present invention collects the signal values of the test part under different exothermic powers and according to Figure 5The method preprocessed the data. Then, the model was initialized and parameters were set, and a Gradient Boosting Regressor with powerful modeling ability was created to handle the non-linear relationships in the data. Next, the present invention started iterative training. In each iteration, the data was divided into a training set and a test set at a ratio of 80% and 20%. The model was trained and predicted on the training set, and the residuals were calculated. To improve the robustness of the model, the present invention dynamically determined the threshold of outliers according to the standard deviation of the residuals, filtered out the samples with absolute residuals greater than the threshold, obtained the filtered training set, and retrained the model on it. Subsequently, the model was evaluated: predictions were made on the test set, the mean square error (MSE) was calculated, and convergence was judged. If the current mean square error was less than the set tolerance, it was considered that the model had achieved the expected accuracy and the iteration was stopped; otherwise, the data set was updated and the next iteration was entered. Finally, the present invention made predictions on the complete data set, calculated the residuals and relative errors, and saved the final model. By this method, the present invention successfully achieved high-precision calibration of the calorimeter signal, significantly improving the reliability and prediction accuracy of the model. As Figure 7 shown, it is the error distribution of the calorimeter test after calibration. It can be seen that in the test part, the exothermic power is between 0.3 - 1000 microwatts, and the error can basically be controlled below 5%, and the error is below 3% for 30 - 1000 microwatts.

[0222] In summary, the present invention can measure the data that can be measured by a conventional isothermal titration calorimeter, and is used to study the dynamic process and thermodynamic parameters of the interaction between molecules, including but not limited to measuring the micelle concentration and micelle formation enthalpy of surfactants, the interaction between proteins and proteins, the interaction between drugs and DNA, the interaction between proteins and nucleic acids, etc. It can be used as an alternative research method for isothermal titration calorimeters.

[0223] Example 1

[0224] In this example, the micelle concentration and micelle formation enthalpy of the anionic surfactant sodium dodecylbenzenesulfonate (SDBS) were measured by continuous flow titration calorimetry.

[0225] The first syringe pump A was filled with CTAB at a concentration of 0.01 mol / L, and the second syringe pump A, the first syringe pump B, and the second syringe pump B were all filled with water. The test temperature was set at 25 °C. Table 1 shows the parameters set for each pump during the test and the corresponding target concentration of CTAB.

[0226] Table 1

[0227]

[0228] According to the set pump parameters, the injection pump is simultaneously turned on and continuously fed through the central control and data collector, and the signals of thermoelectric generator A and thermoelectric generator B and the thermocouple signal are continuously acquired through the analog-to-digital converter. Each pump parameter runs for 10 minutes. Through Figure 5 Signal value processing is carried out. The exothermic power of the test data is predicted using the model established by machine learning. The liquid storage volume of the microfluidic chip is 154 microliters. The heat release and reaction heat are calculated by Formula 1 and Formula 2. The calculation results are shown in Table 2.

[0229] Table 2

[0230]

[0231]

[0232] Draw a graph based on the obtained data ( Figure 9 ), use the Boltzmann function to fit the data, and use Python to fit by the least squares method. The fitting process is as Figure 8 shown. The defined model function is:

[0233]

[0234] Among them, A max is the maximum value of the reaction heat, A min is the minimum value of the reaction heat, and CMC is the micelle concentration. Through fitting, A max = 14.36 ± 0.15 kJ / mol, A min = 2.00 ± 0.10 kJ / mol, CMC = 0.99 ± 0.01 mmol. Calculate the micelle formation enthalpy ΔH mic ,

[0235] -ΔH mic = A max -A min = 12.37 kJ / mol (5)

[0236] Example 2

[0237] In this example, the micelle concentration and micelle formation enthalpy of the anionic surfactant sodium dodecylbenzenesulfonate (SDBS) are tested by continuous flow titration calorimetry.

[0238] The first injection pump A is filled with SDBS with a concentration of 0.01 mol / L, and the second injection pump A, the first injection pump B, and the second injection pump B are all filled with water. The test temperature is set at 25 °C. Table 3 shows the parameters set for each pump during the test and the corresponding target concentration of SDBS.

[0239] Table 3

[0240]

[0241]

[0242] According to the set pump parameters, the injection pump is simultaneously started and continuously fed through the central control and data collector, and the signals of the thermoelectric power generation chip A and the thermoelectric power generation chip B and the thermocouple signal are continuously obtained through the analog-to-digital converter. Each pump parameter runs for 10 minutes. Through Figure 5 Signal value processing is carried out. The exothermic power of the test data is predicted using the model established by machine learning. The liquid storage volume of the microfluidic chip is 154 microliters. The heat release and reaction heat are calculated by Formula 1 and Formula 2. The calculation results are shown in Table 4.

[0243] Table 4

[0244]

[0245] Plot the graph according to the obtained data ( Figure 10 ), and use the Boltzmann function to fit the data. The fitting process is as Figure 8 shown. The defined model function is Formula 3. Through fitting, A max = 16.19 ± 0.55 kJ / mol, A min = 2.13 ± 0.42 kJ / mol, CMC = 1.31 ± 0.05 mmol. Calculate the micelle formation enthalpy ΔH mic ,

[0246] -ΔH mic = A max - A min = 14.06 kJ / mol (6)

[0247] Example 3

[0248] In this example, the micelle concentration and micelle formation enthalpy of the non-ionic surfactant Triton X-100 (TX-100) are tested by continuous flow titration calorimetry.

[0249] The first injection pump A is filled with TX-100 with a concentration of 0.002 mol / L, and the second injection pump A, the first injection pump B, and the second injection pump B are all filled with water. The test temperature is set at 25 °C. Table 5 shows the parameters set for each pump during the test and the corresponding target concentration of TX-100.

[0250] Table 5

[0251]

[0252] It was found in the experiment that the process of TX-100 forming micelles is an endothermic process. Through an editable DC power supply, the heating power of the thin film heating element was set to 5 microwatts. Therefore, the endothermic power P(TX-100) (microwatts) measured was,

[0253] P(TX-100) = 5 - P 预测 (7)

[0254] where P 预测 is the predicted value of the exothermic power of the test data predicted by the model established using machine learning, with the unit of microwatts.

[0255] According to the set pump parameters, the syringe pump was simultaneously started and continuously fed through the central control and data collector, and the signals of the thermoelectric power generation chip A and the thermoelectric power generation chip B and the thermocouple signal were continuously obtained through the analog-to-digital converter. Each pump parameter was run for 10 minutes. Through Figure 5 signal value processing was carried out. The exothermic power of the test data was predicted using the model established by machine learning. The liquid storage capacity of the microfluidic chip was 154 microliters. The heat release and reaction heat were calculated by Formula 1 and Formula 2. The calculation results are shown in Table 6.

[0256] Table 6

[0257]

[0258] Based on the obtained data, a graph was plotted ( Figure 11 ), and the data was fitted using the Boltzmann function. The fitting process is as Figure 8 shown. The defined model function is Formula 3. Through fitting, A max = -2.70 ± 0.06 kJ / mol, A min = -8.40 ± 0.06 kJ / mol, CMC = 0.32 ± 0.00 mmol. The enthalpy of micelle formation ΔH mic ,

[0259] ΔH mic = 5.70 kJ / mol (8)

[0260] Comparative Example

[0261] The data measured using this method was compared with the data in the literature, as shown in Table 7.

[0262] By comparing with the experimental data from different sources in the literature, it can be seen that the critical micelle concentration (CMC) and the enthalpy change of micelle formation (ΔHmic) measured using the method of the present invention show high consistency in terms of accuracy and repeatability. The present invention utilizes continuous flow titration combined with microfluidics and machine learning correction, which can effectively reduce the operation error while shortening the test time, thereby achieving precise detection of thermal effects.

[0263] Table 7 Comparison of measured values ​​and literature values

[0264]

[0265] Embodiment 4

[0266] In this example, the enzymatic reaction kinetics and thermal effect of trypsin and N-benzoyl-L-arginine ethyl ester hydrochloride (BAEE) were measured by continuous flow dynamic titration calorimetry.

[0267] Trypsin was dissolved in 0.05 mol / L phosphate buffer solution (PBS) with a pH of 7.8 to prepare an enzyme solution with a concentration of 10 nmol / L; BAEE was dissolved in the same buffer to prepare a substrate solution of 800 μmol / L. The first injection pump A, the second injection pump A, and the first injection pump B were filled with the above trypsin solution, and the second injection pump B was loaded with the substrate solution. The test temperature was set to 25°C, and four injection pumps were turned on at the same time by the central control and data acquisition device to perform dynamic titration in a continuous feeding manner. Table 8 is a typical flow rate setting scheme in this embodiment. Each flow combination in the test continued to run for 10 minutes to ensure that the flow path was stable and sufficient thermal signals were collected.

[0268] Table 8

[0269]

[0270]

[0271] According to the set pump parameters, the software control device continuously mixes the trypsin solution and BAEE thoroughly, and the channel design of the microfluidic chip allows the enzyme and the substrate to undergo an enzymatic reaction at a constant temperature of 25°C with an exothermic effect. As in Example 1, the heat Q (mJ) generated or absorbed during the reaction is calculated, and the calculation results are shown in FIG. Figure 12 shown.

[0272] Heat release Q obs The initial concentration [S] is linear, assuming that the linear equation is satisfied:

[0273] Q obs (i) = xΔH rxn [S] i V

[0274] Where x is the achieved “co-conversion ratio”, ΔH rxn is the heat released when a unit mole of substrate is completely hydrolyzed (J / mol), [S] i is the initial concentration of the ith experiment, and V is the reaction volume. The slope ΔH is obtained by least squares fitting.rxn [S] i V. Finally, the calculated ΔH rxn is -14.98 kJ / mol.

[0275] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A continuous flow dynamic titration calorimeter combining microfluidics and machine learning correction, characterized in that: include: A heat preservation device, used to place the calorimetric structure, keep the calorimetric structure warm, collect the ambient temperature of the calorimetric structure and send it to the control and signal processing system; A calorimetric structure, comprising a material preheating aluminum block (461) and a constant temperature aluminum block (462) arranged on the left and right, wherein a material inlet and outlet interface block A (451) is arranged in the front groove of the material preheating aluminum block (461), and a material inlet and outlet interface block B (452) is arranged in the rear groove of the material preheating aluminum block (461), a thermoelectric power generation sheet A (441) is arranged in the front part of the constant temperature aluminum block (462), and a thermoelectric power generation sheet B (442) is arranged in the rear part of the constant temperature aluminum block (462), a thin film heating sheet (43) is arranged above the thermoelectric power generation sheet A (441), a microfluidic chip A (421) is arranged above the thin film heating sheet (43), and a microfluidic chip B (422) is arranged above the thermoelectric power generation sheet B (442); The rear portion of the thermostatic aluminum block (462) is connected to a thermostatic oil inlet (463), the front portion of the thermostatic aluminum block (462) is connected to a thermostatic oil outlet (464), the lower portion of the material inlet and outlet interface block A (451) is respectively connected to a material inlet A1 (471), a material inlet A2 (472) and a material outlet A (473), and the lower portion of the material inlet and outlet interface block B (452) is respectively connected to a material inlet B1 (474), a material inlet B2 (475) and a material outlet B (476); The first injection pump A (481), the second injection pump A (482) and the waste liquid tank (49) are connected to the microfluidic chip A (421) after passing through the material preheating aluminum block (461) and the material inlet and outlet interface block A (451); the first injection pump B (483), the second injection pump B (484) and the waste liquid tank (49) are connected to the microfluidic chip B (422) after passing through the material preheating aluminum block (461) and the material inlet and outlet interface block B (452); The control and signal processing system is used to control the editable DC power supply (55) and the injection pump controller (56); and is used to collect the signal generated by the thermoelectric generator and output the calorimetric result.

2. The continuous flow dynamic titration calorimeter combined with microfluidics and machine learning correction according to claim 1, characterized in that The heat preservation device comprises a calorimeter housing upper cover (21), a calorimeter housing base (22) and a thermocouple (3); The calorimeter housing upper cover (21) and the calorimeter housing base (22) are arranged outside the calorimetric structure, the thermal insulation felt upper cover (11) is arranged outside the calorimeter housing upper cover (21), the thermal insulation felt base (12) is arranged outside the calorimeter housing base (22), and the thermocouple (3) is arranged inside the calorimeter housing upper cover (21).

3. The continuous flow dynamic titration calorimeter combined with microfluidics and machine learning correction according to claim 1, characterized in that The material inlet and outlet interface block A (451) and the material inlet and outlet interface block B (452) are provided with a silicone sealing ring (453).

4. The continuous flow dynamic titration calorimeter combined with microfluidics and machine learning correction according to claim 1, characterized in that A material inlet and outlet visual window (411) is provided above the material inlet and outlet interface block A (451) and the material inlet and outlet interface block B (452), and a microfluidic visual window (412) is provided above the microfluidic chip A (421) and the microfluidic chip B (422).

5. The continuous flow dynamic titration calorimeter combined with microfluidics and machine learning correction according to claim 1, characterized in that The microfluidic chip A (421) and the microfluidic chip B (422) have the same structure, and the thermoelectric power generation chip A (441) and the thermoelectric power generation chip B (442) are of the same model.

6. The continuous flow dynamic titration calorimeter combined with microfluidics and machine learning correction according to claim 1, characterized in that The control and signal processing system comprises a central control and data acquisition system (51), a signal amplifier A (521), a signal amplifier B (522), a low-pass filter A (531), a low-pass filter B (532), an analog-to-digital converter (54), an editable DC power supply (55), and an injection pump controller (56); The positive and negative electrodes of the editable DC power supply (55) are connected to the thin film heating plate (43) through electric wires; the signal generated by the thermoelectric power generation plate A (441) is connected to the signal amplifier A (521), the low-pass filter A (531) and the analog-to-digital converter (54) in sequence through the signal transmission line; the signal generated by the thermoelectric power generation plate B (442) is connected to the signal amplifier B (522), the low-pass filter B (532) and the analog-to-digital converter (54) in sequence through the signal transmission line; the analog-to-digital converter (54) transmits the signal to the central control and data acquisition system (51) through the signal transmission line; the central control and data acquisition system (51) controls the editable DC power supply (55) and the injection pump controller (56) through the communication line.

7. A method for using a continuous flow dynamic titration calorimeter combined with microfluidics and machine learning correction, based on the continuous flow dynamic titration calorimeter combined with microfluidics and machine learning correction according to any one of claims 1 to 6, characterized in that: The steps include: S1. Add materials continuously to microfluidic chip A and microfluidic chip B at the same flow rate at the same time; the material continuously introduced into microchannel reference chip A by the first syringe pump A in the test part is the titrant of the test titration process, and the material continuously introduced into microchannel reference chip A by the second syringe pump A is the titrant of the test titration process; the materials continuously introduced into microfluidic chip B by the first syringe pump B and the second syringe pump B in the reference part are both titrants; the flow rates of the first syringe pump A and the first syringe pump B are the same, and the flow rates of the second syringe pump A and the second syringe pump B are the same; S2. The target concentration of the solution in the microfluidic chip A is controlled by controlling the flow rate of each injection pump; the titration process is simulated by adjusting the injection pump flow rate ratio, and constant temperature oil is introduced into the material preheating aluminum block and the constant temperature aluminum block through the constant temperature oil inlet, and the temperature of the constant temperature oil is the set test temperature; the signal values ​​generated by the thermoelectric power generation sheet A and the thermoelectric power generation sheet B are respectively amplified by 100 times through the signal amplifier A and the signal amplifier B, and then the high-frequency noise is filtered out through the low-pass filter A and the low-pass filter B respectively; S3, collecting the signal values ​​of the two channels through the analog-to-digital converter and transmitting them to the central control and data acquisition system; S4, the central control and data acquisition system collects the temperature signal generated by the thermocouple, cleans the data of the signal value, and obtains a valid data set; S41. The temperature signal generated by the thermocouple is continuously collected under the same test conditions through the central control and data acquisition system. The collection time is 10 minutes, and the collected temperature signal is stored in the database, which is recorded as Wherein, T(k) represents the temperature signal value at the kth sampling moment, and N is the total number of sampling points; S42, performing fast Fourier transform on the collected temperature signal data to obtain the energy distribution of the signal in the frequency domain, and transforming the time domain signal {T(k)} into the frequency domain can be expressed as: Where ω is the frequency sampling point index (0≤ω <N); Remove the DC component: set the component corresponding to ω=0 (DC quantity) to 0; Find the maximum peak: Find the frequency component ω with the largest amplitude in the range of ω>0 max , and convert to the corresponding frequency where f s Indicates the signal sampling frequency; set the cutoff frequency: 1.5×f max As the cutoff frequency of the Butterworth low-pass filter, denoted as f cut =1.5·f max ; S43, using a Butterworth low-pass filter to filter the temperature signal {T(k)} to remove high-frequency noise. The transfer function of the filter can be expressed as: Among them, ω c =2πf cut is the angular cutoff frequency of the filter. The transfer function is mapped to the z-domain by discretization to achieve digital filtering. The filtered data is recorded as {T flt1 (k)}, and obtain the preliminary filtering results; S44, based on the preliminary filtering result {T flt1 (k)}, and further use two-dimensional Kalman filtering to estimate and correct the measurement signal; the system state model is: x k =Ax k-1 +Bu k +w k Among them, x k represents the state vector at the kth moment, u k is an optional control quantity vector, w k is process noise, A and B are the system state transfer matrix and control matrix respectively; the observation model is: z k =Hx k +v k Among them, z k is the observed value, v k is the observation noise, H is the observation matrix; in the Kalman filter recursive equation, the prediction stage is: P k|k-1 =AP k|k-1 From T +Q Among them, P k|k-1 is the state covariance prediction, Q is the process noise covariance matrix, and the update phase is: K k =P k|k-1 H T (HP k|k-1 H T +R) -1 P k|k =(I-K k H)P k|k-1 Among them, K k is the Kalman gain matrix, R is the observation noise covariance matrix, and finally the temperature signal {T flt2 (k)}; S45, record the heat release power data of the thin film heating plate as And expand it into a one-dimensional data sequence in time order to maintain the comparability of heat release power at different time points; P(k) is related to the sampling time T of the temperature signal flt2 (k) one-to-one correspondence; S46, using a window function to perform segmented analysis on {P(k)}; setting the window width W = 60, the window moving step s = 2; performing statistics on the power data in the mth window, the data in the window is {P(k)|k∈[k start ,k end ]}, calculate the first-order quartile Q1 and the third-order quartile Q3 in the window, where Q1 represents the value at the 25% position of the data after sorting in ascending order, and Q3 represents the value at the 75% position of the data after sorting in ascending order; S47. Perform the following judgment on the data {P(k)} in each window, and calculate the interquartile range IQR in the window: IQR=Q3-Q1 Determine whether the data is an outlier. If P(k) exceeds the interval [Q1-1.5×IQR, Q3+1.5×IQR], it is considered an outlier. All heat release power values ​​marked as abnormal and the corresponding measurement signal data {T flt2 (k)} are removed together, and only normal data and its temperature signal pairs are retained; through the above window function traversal and abnormal marking, the extreme noise points that appear in the observation or acquisition process are removed, so as to obtain a valid data set {(T flt2 (k),P(k))}; S5. Use machine learning-based methods to build a meta-model based on valid data sets; S6. Use metamodel to predict reaction heat.

8. The method for using the continuous flow dynamic titration calorimeter combined with microfluidics and machine learning correction according to claim 7, characterized in that: S5 specifically includes the following steps: S51, respectively use XGBoost, LightGBM and random forest algorithms to model the training data, and optimize the parameters through grid search, random search or Bayesian optimization to obtain relatively stable parameter configurations, and then form three sets of base models; the objective function of XGBoost is: Among them, y i represents the true label, is the cumulative prediction of the first m-1 trees, f m represents the residual function learned by the mth tree, and Ω(·) is the tree complexity regularization term; For the mth tree, the output of the LightGBM model can be approximately expressed as: F m (x)=F m-1 (x)+α m T m (x) Among them, α m is the learning rate of the mth tree, T m (x) represents the estimate of the input x by the mth tree; the random forest constructs K CART decision trees and obtains the final prediction output by voting or averaging. The model can be expressed as: Among them, T k (x) is the output of the kth decision tree; S52. After adjusting the parameters of XGBoost, LightGBM and random forest respectively, the optimal hyperparameter combination finally locked is used for training to obtain three sets of base models, which are recorded as: M XGB ,M LGB ,M RF S53, input the same training set or validation set data into three sets of base models in sequence to obtain three sets of prediction values: Where X represents the input feature matrix, Represent the heat release prediction value vectors output by XGBoost, LightGBM and random forest respectively; compare the above heat release prediction value vectors with the corresponding true value Y to obtain the error and residual information generated by each of the three models; S54, the true value Y is used as the training label of the meta-model, and Concatenate or combine into a new feature vector, denoted as: S55, using support vector machine regression as the meta-model, denoted by M SVM ; Its objective function can be in the form of ∈-insensitive loss: Among them, w and b are the weight vector and bias term of the support vector machine, C is the regularization coefficient, ξ i represents the slack variable beyond the ∈-insensitive interval; the kernel function K(x i ,x j ) You can choose RBF kernel, linear kernel or polynomial kernel form according to actual needs; S56. Use cross-validation, leave-out validation or other validation techniques to adjust and optimize various parameters of SVM, including kernel function type, regularization coefficient C, and kernel function parameter γ; specifically, they can be expressed as: Among them, CVLoss represents the comprehensive loss value obtained by measuring and averaging the prediction errors during the cross-validation process; through this process, the meta-model fully learns the mapping relationship between the base model output and the true label; S57. When the prediction error of the meta-model on the validation set meets the preset accuracy requirement, the meta-model parameter determination work is completed and the final meta-model is obtained.

9. The method for using the continuous flow dynamic titration calorimeter combined with microfluidics and machine learning correction according to claim 7, characterized in that: When the actual measured ambient temperature exceeds the temperature range of the machine learning training set, the method of improving the robustness of the model output in the extrapolation scenario by linear mapping is used to map the temperature outside the temperature range into the temperature range. The steps are as follows: P1. Based on the corresponding relationship between temperature T and measurement signal M in the existing training data set, the linear fitting model is obtained by using the least squares method or other linear regression methods, which is recorded as: M=k·T+b Where M represents the measurement signal, k and b are the slope and intercept of the linear model respectively, and the formula is as follows: Among them, T i and M i is the temperature and measurement signal of the i-th sample in the training set, and are the average values ​​of temperature and measurement signal, respectively; P2. Set the temperature range of the training data to [T min ,T max ], this interval can be regarded as the core area where the model has high prediction reliability in the temperature dimension; {k,b} and T min ,T max Stored in the parameter library for subsequent calls; P3. For new test data set in is the temperature value of the jth test sample, Represents other feature vectors or measurement signals of the sample, which need to be determined first Whether it falls in [T min ,T max ]; if The temperature is considered normal and no correction is required; if or It is marked as a sample with temperature out of range and recorded in set O; P4. For samples marked as out of range Call the linear relationship and correct the corresponding measurement signal; set the original value of the measurement signal of the out-of-range sample to The mapping correction formula is: in, Represents the corrected measurement signal, with Guarantee new test samples Closer to the training zone [T min ,T max ] corresponding to the feature distribution; P5. Replace the original out-of-range data with all linearly corrected test samples to form a new test data set At this point, the existing meta-model can be used to make predictions on the updated test data.

10. The method for using the continuous flow dynamic titration calorimeter combined with microfluidics and machine learning correction according to claim 7, characterized in that: The specific steps of S6 are as follows: S61, determine whether the process is an endothermic process or an exothermic process. When the process is determined to be an exothermic process, execute S62 to S65; when the process is determined to be an exothermic process, execute S66, S67, S63, S64, and S65 in sequence; S62, input the measurement signal collected in the early stage and pre-processed by filtering and removing abnormal values ​​into the machine learning model, and the meta-model outputs the predicted value of the process heat release power, which is recorded as P 预测 =P; S63. Substitute the heat release power prediction value P output by the machine learning model into the following formula to calculate the process heat release Q: Wherein, V is the liquid volume in the microfluidic chip, which indicates the volume of liquid stored in the chip during the titration process or other test process; q is the flow rate of the material in the microfluidic channel, which is usually set and monitored by a microfluidic pump or other precision flow rate control device; c is the concentration of the target substance in the test material, which is used for the calculation of the subsequent reaction heat ΔH; S64. If the target concentration of the test part material is known to be c, then the total feed amount or reactant content can be characterized under the condition that the volume of the microfluidic chip is V; convert μL to L; S65. Calculate the reaction heat ΔH based on the measured heat release Q and concentration c: S66. If the process is determined to be an endothermic process, the editable DC power supply needs to provide additional heat release power P to the thin film heating element. 补偿 And ensure that the compensation value can cover the power absorbed by the endothermic process, ensuring that the entire system is in the desired temperature or energy balance state; S67, based on the predicted heat absorption power P and compensation power P 补偿 Calculate the actual required power of the heating element: P=P 补偿 -P 预测 Among them, P 预测 is the heat absorption power predicted by the machine learning model; P 补偿 Need to be greater than |P 预测 |, to meet the energy compensation of external heat absorption process.

Images