Machine learning-based solution temperature prediction method, device, electronic equipment and application

Abstract

The present application belongs to the technical field of machine learning application, and particularly relates to a solution temperature prediction method and device based on machine learning, an electronic device and application. The solution temperature prediction method based on machine learning provided by the present application comprises the following steps: 1) obtaining current and historical heating temperature data (i.e. heat source temperature for heating the solution) of the solution; 2) deriving at least 11-dimensional temperature characteristic values of the current and historical time; constructing a current time input sequence; 3) inputting the current time input sequence into a pre-trained solution temperature prediction model (obtained by training a long short-term memory network model according to a training sample set), and outputting the temperature of the solution at the current time and / or future time. The present application can accurately predict the actual temperature of the solution only by obtaining the heat source temperature, thereby providing a temperature control basis for a DNA nanomachine manufacturing system taking the actual temperature of the solution in the reaction tube as a control target, and being conducive to improving the manufacturing precision and yield of the DNA nanomachine.

Description

Technical Field

[0001] This invention belongs to the field of machine learning application technology, and specifically relates to a solution temperature prediction method, device, electronic device and application based on machine learning. Background Technology

[0002] In fields such as medical testing, biochemical reactions, and materials preparation, precise programmed temperature control of solutions within reaction vessels is often required. Temperature control equipment, such as PCR thermal cyclers, is widely used in temperature-sensitive reaction processes such as DNA nanomachine manufacturing, nucleic acid amplification, and enzymatic reactions. Taking the manufacturing of DNA nanomachines as an example: DNA nanotechnology, through the precise design of DNA strand sequences, utilizes their self-assembly properties to construct DNA nanomachines with specific structures and functions, which have significant application value in fields such as virus detection, drug delivery, DNA computing, and DNA information storage. The DNA strands need to complete the folding from a single strand to a predetermined three-dimensional structure within a specific temperature range according to a set temperature variation (annealing) program. The accuracy of temperature control directly determines the folding efficiency, structural uniformity, and assembly yield of the product.

[0003] However, the temperature control method commonly used in existing temperature control equipment is as follows: attach the temperature sensor to the heating module (such as a metal heating base or heating block), use the measured temperature of the heating module as feedback, and adjust the output power of the heating module through closed-loop control so that the temperature of the heating module tracks the set target temperature. This control method has the following problems: (1) Its control target is actually the temperature of the heating module, rather than the actual temperature of the solution in the reaction vessel. The solution temperature in different areas of the vessel is also inconsistent, and this deviation is more obvious in a large-volume reaction system; for temperature-sensitive reactions such as DNA nanomachine manufacturing, the above deviation will cause the hybridization and folding processes of DNA strands in different areas to be asynchronous, thereby affecting the structural uniformity, folding accuracy and assembly yield of the product. (2) Although the temperature sensor (such as a thermocouple) can be directly inserted into the solution to measure the actual temperature of the solution, this method will contact and contaminate the reaction system, interfere with the reaction process, and the sensor is difficult to leave in the reaction process for a long time, and cannot measure the temperature of each batch of reaction vessels online, so it cannot be used as an online means of obtaining the solution temperature. (3) Some temperature control devices provide a "reaction liquid temperature mode", but in essence, it is a fixed empirical compensation of the control parameters based on the liquid volume. The compensation amount is not adjusted with the real-time changes of the heat transfer state during the reaction. It cannot reflect the dynamic changes of heat transfer lag between the heat source and the solution during the temperature change process, and the compensation accuracy is limited.

[0004] Therefore, existing temperature control equipment cannot accurately and in real time obtain the actual temperature of the solution in the reaction vessel in a non-contact manner, and cannot use the actual temperature of the solution as the control target for temperature regulation, resulting in low control precision and unstable product quality in temperature-sensitive reaction processes such as DNA nanomachine manufacturing. Summary of the Invention

[0005] The purpose of this invention is to provide a solution temperature prediction method, device, electronic device and application based on machine learning. It can predict the actual temperature of the solution in the reaction vessel in a non-contact, accurate and real-time manner by only obtaining the temperature of the heat source, and solve the problem that existing temperature control devices cannot use the actual temperature of the solution as the control target for temperature regulation.

[0006] The technical solution of this invention is summarized as follows: The heating temperature (heat source temperature) of the solution at the current and historical times is obtained; at least 11-dimensional temperature feature values, including base temperature, temperature difference, temperature rolling statistics, hysteresis temperature, and temperature thermal inertia state, are derived from the heating temperature to construct the current input sequence; the current input sequence is input into a pre-trained solution temperature prediction model, which outputs the predicted temperature of the solution at the current and / or future times. The prediction model is a time-series prediction model based on a Long Short-Term Memory (LSTM) network. During its training phase, the measured temperature of a thermocouple inserted into the solution is used as a supervision signal; the prediction and control process after model deployment does not require any temperature sensors in the solution and is completely contactless. Based on this, heat source temperature control information can be generated according to the difference between the predicted temperature and the target temperature, forming a closed-loop control. This achieves precise temperature control with the actual solution temperature as the direct control target, providing a basis for DNA nanomachine manufacturing systems that use the actual temperature of the solution in the reaction tube as the control target, thereby improving the manufacturing accuracy and yield of DNA nanomachines.

[0007] The first aspect of this invention provides a solution temperature prediction method based on machine learning, comprising the following steps: 1) Obtain the current heating temperature and historical heating temperature data of the solution; the heating temperature is the temperature of the heat source used to heat the solution; 2) Based on the heating temperature at the current time and the heating temperature at the historical time described in step 1), derive at least 11-dimensional temperature feature values ​​for the current time; derive at least 11-dimensional temperature feature values ​​for the historical time using the same method; then construct the current time input sequence including the 11-dimensional temperature feature values ​​of the current time and the historical time. 3) Input the current time input sequence obtained in step 2) into the pre-trained solution temperature prediction model, and the pre-trained solution temperature prediction model outputs the predicted temperature of the solution at the current time and / or future time. The pre-trained solution temperature prediction model is obtained by training a long short-term memory network model based on a training sample set. Each training sample in the training sample set includes an input sequence sample at the current time and a target value sample at the same time. The input sequence sample at the current time includes at least 11-dimensional temperature feature values ​​of the solution at the current time and at least 11-dimensional temperature feature values ​​at historical times. The target value sample at the same time includes the solution temperature at the current time and / or the measured solution temperature at future times.

[0008] In some optional implementations, the current time input sequence in step 2) includes an 11-dimensional temperature feature value column vector at the current time and an 11-dimensional temperature feature value column vector at historical time. In some optional implementations, the current input sequence sample in step 3) includes an 11-dimensional temperature feature value column vector at the current time and an 11-dimensional temperature feature value column vector at historical times; In some optional implementations, the 11-dimensional temperature feature values ​​include 1-dimensional basic temperature feature values, 2-dimensional temperature difference feature values, 3-dimensional temperature rolling statistical feature values, 3-dimensional hysteresis temperature feature values, and 2-dimensional temperature thermal inertia state feature values.

[0009] In some alternative implementations, the baseline temperature characteristic value includes , As shown in formula (1-1): (1-1), in, Indicates the current moment. Indicates the first t The baseline temperature characteristic value at time t. Indicates the first t The solution heating temperature is measured at all times, and the heating temperature is obtained by a temperature sensor set in the heat source. The temperature difference characteristic values ​​include first-order temperature difference characteristic values. Second-order temperature difference eigenvalues , As shown in formula (1-2), As shown in formula (1-3): (1-2) (1-3) in, Indicates the first t The first-order temperature difference eigenvalue at time t. Indicates the first t The solution heating temperature at time -1 Indicates the first t The second-order temperature difference eigenvalue at time t. Indicates the first t The first-order temperature change at time -1; The temperature rolling statistical feature value includes the average temperature feature value at 10 sampling times. 60 sampling time point temperature average characteristic value Characteristic values ​​of temperature standard deviation at 10 sampling times , As shown in formula (1-4), As shown in formula (1-5), As shown in formula (1-6): (1-4) (1-5) (1-6) in, Indicates the historical moment up to the 1st t The characteristic value of the average heating temperature of the solution at 10 sampling times. Indicates the historical moment up to the 1st t The characteristic value of the average heating temperature of the solution at 60 sampling times. Indicates a historical period that goes back in time. Indicates the historical moment up to the 1st t The standard deviation eigenvalues ​​of the solution heating temperature at 10 sampling times. Indicates the first t The solution heating temperature at time -i; The hysteresis temperature characteristic value includes 10 steps of hysteresis temperature characteristic value. 30-step hysteresis temperature characteristic value 50-step hysteresis temperature characteristic value , As shown in formula (1-7), As shown in formula (1-8), As shown in formula (1-9): (1-7) (1-8) (1-9) in, This indicates the solution heating temperature at the 10th sampling time prior to the current time. This indicates the solution heating temperature at the 30th sampling time prior to the current time. This represents the solution heating temperature at the 50th sampling time prior to the current time, and its characteristic value. Indicates the first t10-step lag temperature at time Indicates the first t 30-step lag temperature at any given time Indicates the first t Temperature at 50-step lag time; The temperature thermal inertia state characteristic values ​​include the fast exponential moving average temperature characteristic values. Slow exponential moving average temperature characteristic value , As shown in formula (1-10), As shown in formula (1-11): (1-10) (1-11) in, This represents the characteristic value of the rapid exponential moving average temperature. Indicates the fast smoothing coefficient. express t The rapid exponential moving average temperature at time -1 This represents the characteristic value of the slow exponential moving average temperature. Indicates the slow smoothing coefficient. > , express t Slow exponential moving average temperature at time -1.

[0010] In some optional implementations, the current input sequence is determined by... The temperature feature value vectors corresponding to each consecutive sampling time are arranged in chronological order, and the temperature feature value vector corresponding to each sampling time is a column vector formed by the 11-dimensional temperature feature values. In some optional implementations, the current time input sequence in step 2) includes ; in, t Indicates the current moment. express t Temperature characteristic value at time, Indicates the time step. This represents the column vector formed by the 11-dimensional temperature eigenvalues. express The column vector formed by the 11-dimensional temperature eigenvalues ​​at time t, express The column vector formed by the 11-dimensional temperature eigenvalues ​​at time t, express The column vector formed by the 11-dimensional temperature eigenvalues ​​at time t, express The column vector formed by the 11-dimensional temperature eigenvalues ​​at time t; In some alternative implementations, the target value includes ; in, To predict the step size, This indicates the current temperature and / or future temperature of the solution.

[0011] In some optional implementations, the input sequence sample at the current time in step 3) includes ; in, t Indicates the current moment. express t Temperature characteristic value at time, Indicates the time step. This represents the column vector formed by the 11-dimensional temperature eigenvalues. express The column vector formed by the 11-dimensional temperature eigenvalues ​​at time t, express The column vector formed by the 11-dimensional temperature eigenvalues ​​at time t, express The column vector formed by the 11-dimensional temperature eigenvalues ​​at time t, express The column vector formed by the 11-dimensional temperature eigenvalues ​​at time t; The target value sample was obtained by actual measurement using a thermocouple temperature sensor inserted into the solution.

[0012] It should be noted that the thermocouple temperature sensor inserted into the solution is only used to collect target value samples during the pre-training stage; after the pre-trained solution temperature prediction model is deployed, no temperature sensor needs to be placed in the solution during the actual prediction and temperature control process, and the prediction process does not come into contact with the solution at all.

[0013] In some optional implementations, the long short-term memory network model described in step 3) includes two layers, with the first hidden layer having dimensions set to 16-128 and the second hidden layer having dimensions set to 16-128.

[0014] A second aspect of the present invention also provides a solution temperature control method, the solution temperature control method comprising the following steps: S1. Based on the above solution temperature prediction method, predict the current and / or future temperature of the solution, and obtain the target temperature of the solution. S2. Based on the predicted temperature and the target temperature, generate heat source temperature control information for regulating the solution temperature; the heat source is used to heat the solution. In some optional embodiments, generating heat source temperature control information for regulating solution temperature includes: When the difference between the target temperature and the predicted temperature is greater than 0, temperature control information to increase the temperature of the heat source is generated. When the difference between the target temperature and the predicted temperature is less than 0, temperature control information to reduce the temperature of the heat source is generated. When the difference between the target temperature and the predicted temperature is equal to 0, temperature control information for maintaining the temperature of the heat source is generated.

[0015] A third aspect of the present invention provides a solution temperature prediction device, the solution temperature prediction device comprising: The data acquisition module is used to acquire the current heating temperature and historical heating temperature data of the solution; the heating temperature is the temperature of the heat source used to heat the solution. The feature construction module is used to construct at least 11-dimensional temperature feature values ​​for the current time based on the heating temperature data at the current time and the heating temperature data at the historical time obtained by the data acquisition module, and to construct at least 11-dimensional temperature feature values ​​for the historical time in the same way, and to construct the current time input sequence including at least 11-dimensional temperature feature values ​​for the current time and the historical time. The prediction module includes a pre-trained solution temperature prediction model, which is used to input the current time input sequence constructed by the feature construction module, and predicts and outputs the predicted temperature of the solution at the current time and / or future time based on the current time input sequence.

[0016] In some optional implementations, the data acquisition module is further configured to acquire the measured solution temperature at the current time and / or future time. The solution temperature prediction device further includes: The training sample set acquisition module is used to construct at least 11-dimensional temperature feature values ​​for the current time and at least 11-dimensional temperature feature values ​​for the historical time from the heating temperature data at the current time and the heating temperature data at the historical time acquired by the data acquisition module, and to construct the current time input sequence including the at least 11-dimensional temperature feature values ​​at the current time and the historical time. The current time input sequence and the measured solution temperature at the same time and / or future time are used to construct a training sample, and multiple training samples at different times are obtained to obtain the training sample set. The model pre-training module is used to train the long short-term memory network model using the training sample set to obtain a pre-trained solution temperature prediction model.

[0017] A fourth aspect of the present invention provides a solution temperature prediction electronic device, the solution temperature prediction electronic device including a memory, a processor and a computer program stored in the memory, wherein the processor implements the above-described method when executing the computer program.

[0018] The fifth aspect of the present invention provides an application of the above-described machine learning-based solution temperature prediction method, the above-described solution temperature prediction device, or the above-described solution temperature prediction electronic device in a DNA manufacturing system.

[0019] In some alternative embodiments, the DNA manufacturing system includes a DNA nanomachine manufacturing system and a PCR amplification instrument; In some alternative embodiments, the DNA manufacturing system further includes a solution temperature control component; The solution temperature control component is configured to: acquire the target temperature of the solution to be controlled, acquire the predicted temperature of the solution to be controlled based on the above-mentioned machine learning-based solution temperature prediction method, the above-mentioned solution temperature prediction device, or the above-mentioned solution temperature prediction electronic device, and generate heat source temperature control information for controlling the temperature of the solution to be controlled based on the predicted temperature and the target temperature; the heat source is used to heat the solution to be controlled. The heat source temperature control information generated for regulating the temperature of the solution to be controlled includes: When the difference between the target temperature and the predicted temperature is greater than 0, temperature control information to increase the temperature of the heat source is generated. When the difference between the target temperature and the predicted temperature is less than 0, temperature control information to reduce the temperature of the heat source is generated. When the difference between the target temperature and the predicted temperature is equal to 0, temperature control information for maintaining the temperature of the heat source is generated.

[0020] Compared with the prior art, the technical solution of the present invention has the following advantages: 1. The solution temperature prediction method based on machine learning provided by this invention derives at least 11-dimensional temperature feature values ​​from the current and historical heating temperatures of the solution, comprehensively characterizing the heating trend, temperature fluctuations, and thermal inertia characteristics during the heat transfer process. It also utilizes a pre-trained long short-term memory network model to learn the nonlinear heat conduction relationship between the heat source and the solution. It only needs to measure the heat source temperature, which is already available in the temperature control equipment and is easily obtainable online, to accurately and in real time predict the actual temperature of the solution in the reaction vessel, and can output the predicted temperature for future times as needed. The prediction process does not involve contact with the solution, does not contaminate the reaction system, does not interfere with the reaction process, can output data online in real time and continuously, and can be implemented simultaneously for batch reaction vessels.

[0021] 2. This invention trains the model by collecting training samples under target operating conditions, which can be adapted to different container specifications, solution volumes and temperature programs. It can achieve high prediction accuracy under various target operating conditions and has strong adaptability to operating conditions.

[0022] 3. The solution temperature prediction method, device, or electronic device based on machine learning provided by this invention can be applied to DNA nanomachine manufacturing systems. The manufacturing system accurately predicts the real-time temperature of the reaction solution by configuring a pre-trained solution temperature prediction model, providing a basis for DNA nanomachine manufacturing systems that use the actual temperature of the solution in the reaction tube as the control target, thereby improving the manufacturing accuracy and yield of the DNA nanomachine manufacturing system.

[0023] 4. Based on the above applications, the solution temperature control component in the DNA manufacturing system adaptively generates heat source temperature control information according to the difference between the predicted temperature and the target temperature. It works in conjunction with the solution temperature prediction model to form a closed-loop control, realizing precise temperature control with the actual solution temperature as the direct control target. This improves the temperature control accuracy, product structure uniformity, and manufacturing yield of the DNA nanomachine manufacturing system.

[0024] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0025] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0026] Figure 1 This is a schematic diagram of the online real-time prediction and temperature feedback control process of a solution temperature prediction method based on machine learning according to a certain embodiment of the present invention. Figure 2 This is a schematic diagram of the pre-training and deployment method of a pre-trained solution temperature prediction model in a solution temperature prediction method based on machine learning according to a certain embodiment of the present invention. Figure 3 This is a schematic diagram of a DNA nanomachine manufacturing system according to one embodiment of the present invention; Figure 4 This is a schematic diagram of a DNA nanomachine manufacturing system according to one embodiment of the present invention; Figure 5 This is a schematic diagram of a DNA nanomachine manufacturing system according to one embodiment of the present invention; Figure 6 This is a schematic diagram of a DNA nanomachine manufacturing system according to one embodiment of the present invention; Figure 7 This is a comparison chart of the current temperature of the solution in the reaction tube of the DNA nanomachine manufacturing system predicted in Example 3 of the present invention and the actual measured temperature. Detailed Implementation

[0027] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0028] It should be understood that the terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in a particular order described or illustrated unless the order of performance is explicitly indicated. It should also be understood that additional or alternative steps may be used.

[0029] In the description of the embodiments of this invention, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this invention, "multiple" means two or more, unless otherwise explicitly defined.

[0030] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0031] In the description of the embodiments of this invention, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0032] In the description of the embodiments of the present invention, the term "multiple" refers to two or more (including two), similarly, "multiple groups" refers to two or more (including two groups), and "multiple pieces" refers to two or more (including two pieces).

[0033] In the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in the embodiments of the present invention according to the specific circumstances.

[0034] The first aspect of this invention provides a solution temperature prediction method based on machine learning, comprising the following steps: 1) Obtain the current heating temperature and historical heating temperature data of the solution; the heating temperature is the temperature of the heat source used to heat the solution; 2) Based on the heating temperature at the current time and the heating temperature at the historical time described in step 1), derive at least 11-dimensional temperature feature values ​​for the current time; derive at least 11-dimensional temperature feature values ​​for the historical time using the same method; then construct the current time input sequence including at least 11-dimensional temperature feature values ​​for the current time and the historical time. 3) Input the current time input sequence obtained in step 2) into the pre-trained solution temperature prediction model, and the pre-trained solution temperature prediction model outputs the predicted temperature of the solution at the current time and / or future time. The pre-trained solution temperature prediction model is obtained by training a Long Short-Term Memory (LSTM) network model based on a training sample set. Each training sample in the training sample set includes an input sequence sample at the current time and a target value sample at the same time. The input sequence sample at the current time includes at least 11-dimensional temperature feature values ​​of the solution at the current time and at least 11-dimensional temperature feature values ​​at historical times. The target value sample at the same time includes the solution temperature at the current time and / or the measured solution temperature at future times.

[0035] In some optional implementations, the online real-time prediction process and temperature feedback control diagram of the machine learning-based solution temperature prediction method are shown below. Figure 1 As shown, the prediction process does not require contact with the solution.

[0036] like Figure 1As shown, the method includes step 1) acquiring heating temperature data, step 2) constructing the current input sequence, and step 3) model prediction; in some optional embodiments, it also includes step 4) temperature feedback control: comparing the predicted temperature of the solution with the target temperature, generating heat source temperature regulation information, regulating the heat source temperature, and continuously predicting the solution temperature based on a machine learning-based solution temperature prediction method after adjusting the heat source, thus forming a closed-loop control.

[0037] In some optional implementations, the current time input sequence in step 2) includes an 11-dimensional temperature feature value column vector at the current time and an 11-dimensional temperature feature value column vector at historical time. In some optional implementations, the current input sequence sample in step 3) includes an 11-dimensional temperature feature value column vector at the current time and an 11-dimensional temperature feature value column vector at historical times; In some optional implementations, the 11-dimensional temperature feature values ​​include 1-dimensional basic temperature feature values, 2-dimensional temperature difference feature values, 3-dimensional temperature rolling statistical feature values, 3-dimensional hysteresis temperature feature values, and 2-dimensional temperature thermal inertia state feature values.

[0038] In some alternative implementations, the baseline temperature characteristic value includes , As shown in formula (1-1): (1-1), in, Indicates the current moment. Indicates the first t The baseline temperature characteristic value at time t. Indicates the first t The constant solution heating temperature is obtained by measuring a temperature sensor located on the heat source. The temperature difference characteristic values ​​include first-order temperature difference characteristic values. Second-order temperature difference eigenvalues , As shown in formula (1-2), As shown in formula (1-3): (1-2) (1-3) in, Indicates the first t The first-order temperature difference eigenvalue at time t. Indicates the first t The solution heating temperature at time -1 Indicates the first t The second-order temperature difference eigenvalue at time t. Indicates the firstt The first-order temperature change at time -1; The temperature rolling statistical feature value includes the average temperature feature value at 10 sampling times. 60 sampling time point temperature average characteristic value Characteristic values ​​of temperature standard deviation at 10 sampling times , As shown in formula (1-4), As shown in formula (1-5), As shown in formula (1-6): (1-4) (1-5) (1-6) in, Indicates the historical moment up to the 1st t The characteristic value of the average heating temperature of the solution at 10 sampling times. Indicates the historical moment up to the 1st t The characteristic value of the average heating temperature of the solution at 60 sampling times. Indicates a historical period that goes back in time. Indicates the historical moment up to the 1st t The standard deviation eigenvalues ​​of the solution heating temperature at 10 sampling times. Indicates the first t -i is the solution heating temperature at time i; in some optional implementations, when t When the value is less than 60, the system uses boundary padding (such as forward padding with the first available value) to ensure online availability.

[0039] The hysteresis temperature characteristic value includes 10 steps of hysteresis temperature characteristic value. 30-step hysteresis temperature characteristic value 50-step hysteresis temperature characteristic value , As shown in formula (1-7), As shown in formula (1-8), As shown in formula (1-9): (1-7) (1-8) (1-9) in, This indicates the solution heating temperature at the 10th sampling time prior to the current time. This indicates the solution heating temperature at the 30th sampling time prior to the current time. This represents the solution heating temperature at the 50th sampling time prior to the current time, and its characteristic value. Indicates the first t 10-step lag temperature at time Indicates the first t 30-step lag temperature at any given time Indicates the first t The temperature is 50 steps behind the time; in some alternative implementations, when (t < 10 / 30 / 50), the system uses boundary padding (e.g., forward padding with the first available value) to ensure online availability.

[0040] The temperature thermal inertia state characteristic values ​​include the fast exponential moving average temperature characteristic values. Slow exponential moving average temperature characteristic value , As shown in formula (1-10), As shown in formula (1-11): (1-10) (1-11) in, This represents the characteristic value of the rapid exponential moving average temperature. Indicates the fast smoothing coefficient. express t The rapid exponential moving average temperature at time -1 This represents the characteristic value of the slow exponential moving average temperature. Indicates the slow smoothing coefficient. > , express t Slow exponential moving average temperature at time -1.

[0041] In some alternative implementations, The value ranges from 0.1 to 0.5. The value ranges from 0.01 to 0.05. In some alternative implementations, this coefficient ( , It can be used as a hyperparameter to be adaptively adjusted for different working conditions through optimization algorithms (such as grid search).

[0042] In some optional implementations, the current input sequence is determined by... The temperature feature value vectors corresponding to each consecutive sampling time are arranged in chronological order, and the temperature feature value vector corresponding to each sampling time is a column vector formed by the 11-dimensional temperature feature values. In some optional implementations, the current time input sequence in step 2) includes ; in, t Indicates the current moment. express t Temperature characteristic value at time, Indicates the time step. This represents the column vector formed by the 11-dimensional temperature feature values; for example, express The column vector formed by the 11-dimensional temperature feature values ​​at time t is specifically as follows: , express The column vector formed by the 11-dimensional temperature feature values ​​at time t is specifically as follows: , express The column vector formed by the 11-dimensional temperature feature values ​​at time t is specifically as follows: , express The column vector formed by the 11-dimensional temperature feature values ​​at time t is specifically as follows: ; Current input sequence The expanded form is: .

[0043] In some alternative implementations, the target value includes ; in, To predict the step size, This indicates the current temperature and / or future temperature of the solution.

[0044] In some optional implementations, the input sequence sample at the current time in step 3) includes ; in, t Indicates the current moment. express t Temperature characteristic value at time, Indicates the time step. This represents the column vector formed by the 11-dimensional temperature eigenvalues. express The column vector formed by the 11-dimensional temperature feature values ​​at time t is specifically as follows: , express The column vector formed by the 11-dimensional temperature feature values ​​at time t is specifically as follows: , express The column vector formed by the 11-dimensional temperature feature values ​​at time t is specifically as follows: , express The column vector formed by the 11-dimensional temperature feature values ​​at time t is specifically as follows: ; Current input sequence sample The expanded form is: .

[0045] The target value sample is obtained by actually measuring the current and / or future temperature of the solution using a thermocouple temperature sensor inserted into the solution.

[0046] In some optional implementations, the long short-term memory network model described in step 3) includes two layers, with the first hidden layer having dimensions set to 16-128 and the second hidden layer having dimensions set to 16-128.

[0047] In some alternative implementations, the pre-trained solution temperature prediction model includes a two-layer long short-term memory network model trained with training samples; the first hidden layer dimension of the two-layer long short-term memory network model is set to 16-128, and the second hidden layer dimension is set to 16-128.

[0048] In some alternative implementations, the first hidden layer dimension of the two-layer long short-term memory network model is set to 64-128, and the second hidden layer dimension is set to 32-64. This results in a stronger model memory capacity, suitable for complex long-sequence dependencies, but if the hidden layer dimension is too high (greater than 128), it is prone to overfitting (especially when training data is limited). Setting the first hidden layer dimension of the two-layer long short-term memory network model to 32-64 and the second hidden layer dimension to 16-32 results in a lighter model and faster training, but if the hidden layer dimension is too low (less than 16), it may underfit (failing to remember complex temperature patterns). Preferably, the first hidden layer dimension of the two-layer long short-term memory network model is set to 64, and the second hidden layer dimension is set to 32.

[0049] In some alternative implementations, the pre-training of the pre-trained solution temperature prediction model includes: (I) Construction of Temperature Prediction Model This invention constructs a deep neural network model based on Long Short-Term Memory (LSTM) as a temperature prediction model. The temperature prediction model employs a two-layer LSTM network (in this embodiment, the hidden layer dimensions are 64 and 32, respectively) plus a fully connected output layer. This model learns the nonlinear heat conduction characteristics between the heat source and the solution to predict the estimated temperature of the current solution based on the current and historical heating module temperatures.

[0050] (II) Temperature Prediction Model Training S1: Data Acquisition and Preprocessing. Simultaneously acquire two types of temperature data under different operating conditions (e.g., different solution volumes, different temperature control programs): one is the temperature of the heat source (reaction tube heating element)... T block The temperature is measured by a first temperature sensor (such as a PT1000 temperature sensor) installed on the heat source (reaction tube heat sink); the second is the solution temperature in the reaction tube assembly. T sol The data was measured by a thermocouple inserted into the solution inside the reaction tube. Both types of data were acquired synchronously at the same sampling frequency (2Hz). Each data record includes time t and the corresponding time. T block ( t )and T sol ( t The collected data undergoes format standardization and cleaning, including: numerical type conversion, handling of missing / outlier values, and removal of duplicate rows. After format standardization and cleaning, the data needs to be filtered in step S2.

[0051] S2: Screening of operating condition similarity data. The dynamic relationship between the heating module temperature and the solution temperature differs under different operating conditions. This is achieved through calculation... T block and T sol The lagged correlation between them can quantify their dynamic relationship. Similarity filtering is performed on each training file to be tested within a preset search window. Internal calculation:

[0052] in, denoted as Pearson correlation coefficient, which measures the strength of the linear correlation between two sequences in terms of their trend of change, and its value ranges from [-1, 1]. It's about the lag step size. k The function reflects the relevance score at a specific delay; The first subset of the heating module sequence, i.e. , The latter subset of the solution sequence is used to ensure that the two sequences have the same length through this "sliding window alignment"; L represents the total length of the original temperature sampling sequence; k represents the hysteresis search step size, i.e., the solution temperature sequence Move the sampling points backward by k to match the temperature of the heating module. Alignment is used to simulate the physical time required for heat transfer; This represents the maximum lag search step size, used to limit the search range for lag correlation calculation. In the example, it can be 10, 20, or 30. This indicates that the Pearson correlation coefficient is... Lag step size for achieving the maximum value ; The main lag step, i.e., the delay corresponding to the Pearson correlation coefficient reaching its maximum value, reflects the time delay of heat transfer from the heating block to the solution under this operating condition. This represents the maximum correlation coefficient value.

[0053] when (Maximum allowable hysteresis threshold) or When the minimum relevance threshold is reached, it will not participate in the construction of subsequent training samples; in a specific instance, it is set... and The values ​​are 100 and 0.85 respectively.

[0054] By filtering based on operating condition similarity, training samples with excessive latency or mismatched dynamic characteristics are removed, allowing the model to focus on learning the temperature dynamic characteristics under the target operating condition, thereby improving the prediction accuracy under that condition. After filtering, the final data constructed by multi-dimensional feature engineering is obtained, and the final data is randomly divided into training set, validation set, and test set according to a certain ratio (e.g., 8:1:1).

[0055] S3: Multidimensional Feature Engineering Construction. For the temperature data after S2 filtering, an 11-dimensional feature column vector is constructed for the temperature data at each time point to represent the temperature rise trend, fluctuation, and thermal inertia state; each feature uses only the temperature data at the current time and historical time points.

[0056] The 11-dimensional temperature characteristics, as shown above, include 1-dimensional basic temperature characteristics, 2-dimensional temperature difference characteristics, 3-dimensional temperature rolling statistics characteristics, 3-dimensional hysteresis temperature characteristics, and 2-dimensional temperature thermal inertia state characteristics. The basic temperature characteristic values ​​corresponding to the basic temperature characteristics include , As shown in formula (1-1): (1-1), in, This represents the baseline temperature characteristic value at time t. The temperature monitored by the first temperature sensor at time t is obtained by measuring the temperature of the temperature sensor located on the heat source. The temperature difference eigenvalues ​​corresponding to the temperature difference features include first-order temperature difference eigenvalues. Second-order temperature difference eigenvalues , As shown in formula (1-2), As shown in formula (1-3): (1-2) (1-3) in, Indicates the first t The first-order temperature difference eigenvalue at time t. Indicates the first t The temperature monitored by the first temperature sensor at time -1. Indicates the first t The second-order temperature difference eigenvalue at time t. Indicates the first t The first-order temperature change at time -1; The temperature rolling statistical feature includes the average temperature value at 10 sampling times. 60 sampling time point temperature average characteristic value Characteristic values ​​of temperature standard deviation at 10 sampling times , As shown in formula (1-4), As shown in formula (1-5), As shown in formula (1-6): (1-4) (1-5) (1-6) in, Indicates the historical moment up to the 1st t The characteristic value of the average temperature monitored by the first temperature sensor at 10 sampling times. Indicates the historical moment up to the 1st t The characteristic value of the average temperature monitored by the first temperature sensor at 60 sampling times. Indicates a historical period that goes back in time. Indicates the historical moment up to the 1st t The standard deviation characteristic values ​​of the monitored temperature of the first temperature sensor at 10 sampling times. Indicates the first t - i The temperature monitored by the first temperature sensor at any given time; when ( t When the value is less than 60, the system uses boundary padding (such as forward padding with the first available value) to ensure online availability.

[0057] The hysteresis temperature characteristic corresponding to the hysteresis temperature characteristic includes 10-step hysteresis temperature characteristic values. 30-step hysteresis temperature characteristic value 50-step hysteresis temperature characteristic value , As shown in formula (1-7), As shown in formula (1-8), As shown in formula (1-9): (1-7) (1-8) (1-9) in, This indicates the temperature monitored by the first temperature sensor at the 10th sampling time prior to the current time. This indicates the temperature monitored by the first temperature sensor at the 30th sampling time prior to the current time. This represents the monitored temperature of the first temperature sensor at the 50th sampling time prior to the current time, and its characteristic value. Indicates the first t 10-step lag temperature at time Indicates the first t 30-step lag temperature at any given time Indicates the first t The temperature with a 50-step lag at time t; when (t < 10 / 30 / 50), the system uses boundary padding (e.g., forward padding with the first available value) to ensure online availability.

[0058] The temperature thermal inertia state characteristic values ​​corresponding to the temperature thermal inertia state characteristics include the fast exponential moving average temperature characteristic values. Slow exponential moving average temperature characteristic value , As shown in formula (1-10), As shown in formula (1-11): (1-10) (1-11) in, This represents the characteristic value of the rapid exponential moving average temperature. Indicates the fast smoothing coefficient. This represents the rapid exponential moving average temperature at time t-1. This represents the characteristic value of the slow exponential moving average temperature. Indicates the slow smoothing coefficient. > , This represents the slow exponential moving average temperature at time t-1. , This coefficient (represents short-term and long-term thermal dynamics respectively) , This parameter can be used as a hyperparameter and adaptively adjusted for different working conditions through optimization algorithms (such as grid search). The value ranges from 0.1 to 0.5. The value ranges from 0.01 to 0.05.

[0059] Column vector composed of 11-dimensional features This data was used to construct subsequent sliding window samples.

[0060] S4: Sliding Window Sample Construction. The feature sequence generated in step S3 is transformed into the input of the subsequent temperature prediction model. The window length is set to W (160 in this example), and the prediction step size is H (0 in a certain example, i.e., predicting the solution temperature at the same moment). The input of the model is constructed by sliding each feature vector along the time axis. and output sample: ; ; The above samples are packaged into binary array files, and meta-information containing normalization parameters is output. These binary files are the direct input source for the next step of the training process.

[0061] S5: Model Training and Consolidation: The temperature prediction model constructed in this invention has the following architecture: Input Layer: Receives a feature time-series window of dimension (W, 11) generated by step S4, where W is the time step and 11 is the feature dimension extracted at each time step. Hidden Layer: Consists of a two-layer LSTM network. The first LSTM hidden layer has a dimension of 64, used to extract high-dimensional dynamic thermal characteristics from the original features; the second LSTM hidden layer has a dimension of 32, used to further compress the time-series features and enhance the extraction of long-range dependencies. Output Layer: Maps the terminal hidden states of the LSTM to scalar space through a fully connected layer, outputting the predicted value of the solution temperature at the current time step.

[0062] The model training objective is to minimize the average loss of the training set samples. During model training, dropout and gradient pruning techniques are used to optimize the model's generalization ability under different operating conditions. After training, the model parameters (resulting in a fixed model) and standardized parameters are saved to obtain a pre-trained temperature prediction model, which is then used for subsequent test set evaluation and deployment for inference.

[0063] In some optional implementations, a schematic diagram of the pre-training and deployment method of the pre-trained solution temperature prediction model in the machine learning-based solution temperature prediction method is shown below. Figure 2 As shown, the process includes: S1 data acquisition and preprocessing, S2 data screening and sample set partitioning, S3 feature construction, S4 training sample construction, S5 model training, and S6 model deployment; wherein the measured solution temperature is only collected by thermocouples inserted into the solution during the training phase, and the prediction process after model deployment does not require contact with the solution.

[0064] A second aspect of the present invention also provides a solution temperature control method, the solution temperature control method comprising the following steps: A1. Based on the above solution temperature prediction method, predict the current and / or future temperature of the solution, and obtain the target temperature of the solution. A2. Based on the predicted temperature and the target temperature, generate heat source temperature control information for regulating the solution temperature; the heat source is used to heat the solution. In some optional embodiments, generating heat source temperature control information for regulating solution temperature includes: When the difference between the target temperature and the predicted temperature is greater than 0, temperature control information to increase the temperature of the heat source is generated. When the difference between the target temperature and the predicted temperature is less than 0, temperature control information to reduce the temperature of the heat source is generated. When the difference between the target temperature and the predicted temperature is equal to 0, temperature control information for maintaining the temperature of the heat source is generated.

[0065] In some optional embodiments, the heat source is a semiconductor cooling chip, and the temperature control information for increasing the temperature of the heat source includes driving the heat source to heat and adaptively increasing the output power of the heat source; in some optional embodiments, the temperature control information for decreasing the temperature of the heat source includes driving the heat source to cool and adaptively increasing the output power of the heat source; in some optional embodiments, the temperature control information for maintaining the temperature of the heat source includes adaptively decreasing the output power of the heat source and decreasing the duty cycle.

[0066] A third aspect of the present invention provides a solution temperature prediction device, the solution temperature prediction device comprising: The data acquisition module is used to acquire the current heating temperature and historical heating temperature data of the solution; the heating temperature is the temperature of the heat source used to heat the solution. The feature construction module is used to construct at least 11-dimensional temperature feature values ​​for the current time based on the heating temperature data at the current time and the heating temperature data at the historical time obtained by the data acquisition module, and to construct at least 11-dimensional temperature feature values ​​for the historical time in the same way, and to construct the current time input sequence including at least 11-dimensional temperature feature values ​​for the current time and the historical time. The prediction module includes a pre-trained solution temperature prediction model, which is used to input the current time input sequence constructed by the feature construction module, and predicts and outputs the predicted temperature of the solution at the current time and / or future time based on the current time input sequence.

[0067] In some optional implementations, the data acquisition module is further configured to acquire the measured solution temperature at the current time and / or future time. The solution temperature prediction device further includes: The training sample set acquisition module is used to construct at least 11-dimensional temperature feature values ​​for the current time and at least 11-dimensional temperature feature values ​​for the historical time from the heating temperature data at the current time and the heating temperature data at the historical time acquired by the data acquisition module, and to construct the current time input sequence including the at least 11-dimensional temperature feature values ​​at the current time and the historical time. The current time input sequence and the measured solution temperature at the same time and / or future time are used to construct a training sample, and multiple training samples at different times are obtained to obtain the training sample set. The model pre-training module is used to train the long short-term memory network model using the training sample set to obtain a pre-trained solution temperature prediction model.

[0068] A fourth aspect of the present invention provides a solution temperature prediction electronic device, the solution temperature prediction electronic device including a memory, a processor and a computer program stored in the memory, wherein the processor implements the above-described method when executing the computer program.

[0069] The fifth aspect of the present invention provides an application of the above-described machine learning-based solution temperature prediction method, the above-described solution temperature prediction device, or the above-described solution temperature prediction electronic device in a DNA manufacturing system.

[0070] In some optional embodiments, the DNA manufacturing system includes a DNA nanomachine manufacturing system and a PCR amplification instrument; in some optional embodiments, the DNA manufacturing system further includes a solution temperature control component. The solution temperature control component is configured to: acquire the target temperature of the solution to be controlled, acquire the predicted temperature of the solution to be controlled based on the above-mentioned machine learning-based solution temperature prediction method, the above-mentioned solution temperature prediction device, or the above-mentioned solution temperature prediction electronic device, and generate heat source temperature control information for controlling the temperature of the solution to be controlled based on the predicted temperature and the target temperature; the heat source is used to heat the solution to be controlled. The heat source temperature control information generated for regulating the temperature of the solution to be controlled includes: When the difference between the target temperature and the predicted temperature is greater than 0, temperature control information to increase the temperature of the heat source is generated. When the difference between the target temperature and the predicted temperature is less than 0, temperature control information to reduce the temperature of the heat source is generated. When the difference between the target temperature and the predicted temperature is equal to 0, temperature control information for maintaining the temperature of the heat source is generated.

[0071] In some optional embodiments, the heat source is a semiconductor cooling chip, and the temperature control information for increasing the temperature of the heat source includes driving the heat source to heat and adaptively increasing the output power of the heat source; in some optional embodiments, the temperature control information for decreasing the temperature of the heat source includes driving the heat source to cool and adaptively increasing the output power of the heat source; in some optional embodiments, the temperature control information for maintaining the temperature of the heat source includes adaptively decreasing the output power of the heat source and decreasing the duty cycle.

[0072] Combination Figure 3-6 As described above, the sixth aspect of this invention also provides a DNA nanomachine manufacturing system, such as... Figure 3 As shown, the DNA nanomachine manufacturing system includes a temperature control device 1 and a reaction tube assembly 2. The temperature control device 1 includes a main heating module group 12 and a control module group. The main heating module group 12 includes multiple independent temperature control zones, each of which includes a reaction tube heat sink, a first temperature sensor, and a first semiconductor refrigeration chip. The reaction tube heat sink is used to place the reaction tube assembly. The first temperature sensor is disposed on the side wall of the reaction tube heat sink and is used to monitor the temperature of the reaction tube heat sink. The first semiconductor refrigeration chip is disposed below the reaction tube heat sink and is used to regulate the temperature of the reaction tube heat sink. The control module group is used to obtain the monitored temperature of the first temperature sensor (i.e., the temperature of the reaction tube heat sink) from the main heating module group and to obtain the target temperature of the solution contained in the reaction tube assembly from the user end. The control module group is configured to include the above-mentioned solution temperature prediction device and solution temperature regulation component. The solution temperature prediction device uses the monitored temperature of the first temperature sensor (i.e., the temperature of the reaction tube heat sink) as the heating temperature to implement the above-mentioned solution temperature prediction method. The solution temperature control component is configured to: acquire the target temperature of the solution contained in the reaction tube assembly, acquire the predicted solution temperature predicted by the solution temperature prediction device, and generate reaction tube heat sink temperature control information for controlling the temperature of the solution contained in the reaction tube assembly based on the predicted temperature and the target temperature. The generation of reaction tube hot seat temperature control information for regulating the temperature of the solution contained in the reaction tube assembly includes: When the difference between the target temperature and the predicted temperature is greater than 0, temperature control information for increasing the temperature of the reaction tube hot seat is generated; in some optional embodiments, the temperature control information for increasing the temperature of the reaction tube hot seat includes adaptively increasing the output power of the first semiconductor refrigeration chip and driving the first semiconductor refrigeration chip to heat up. When the difference between the target temperature and the predicted temperature is less than 0, temperature control information for reducing the temperature of the reaction tube hot seat is generated; in some optional embodiments, the temperature control information for reducing the temperature of the reaction tube hot seat includes adaptively increasing the output power of the first semiconductor refrigeration chip and driving the first semiconductor refrigeration chip to cool. When the difference between the target temperature and the predicted temperature is equal to 0, temperature control information for maintaining the temperature of the reaction tube hot seat is generated. In some optional embodiments, the temperature control information for maintaining the temperature of the reaction tube hot seat includes adaptively reducing the output power of the first semiconductor cooling chip and reducing the duty cycle; In some optional implementations, the control module group is also configured to obtain the volume of the solution contained in the reaction tube assembly from the user terminal, and to adaptively increase or decrease the output power of the first semiconductor refrigeration chip in the temperature control information of the reaction tube hot seat, which is determined based on the magnitude of the difference between the obtained volume of the solution contained in the reaction tube assembly and / or the predicted temperature of the solution contained in the reaction tube assembly at the current temperature.

[0073] In some optional implementations, the control module group is further configured to obtain the volume of the solution contained in the reaction tube assembly from the user terminal, wherein the difference in the adaptive increase of the output power of the first semiconductor cooler in the temperature control information of the reaction tube hot seat is positively correlated with the volume of the solution contained in the reaction tube assembly and / or the difference in the predicted temperature of the solution contained in the reaction tube assembly at the current temperature is positively correlated with the volume of the solution contained in the reaction tube assembly in the temperature control information of the reaction tube hot seat; wherein the difference in the adaptive decrease of the output power of the first semiconductor cooler in the temperature control information of the reaction tube hot seat is positively correlated with the volume of the solution contained in the reaction tube assembly and / or the difference in the predicted temperature of the solution contained in the reaction tube assembly at the current temperature is positively correlated with the volume of the solution contained in the reaction tube assembly.

[0074] like Figure 4 As shown, the temperature control device 1 includes a housing 11, a main heating module group 12 and a heating cover module group 13. The housing 11 includes an upper housing 111, a lower housing 112, an input command window 113 (in some optional embodiments, the input command window 113 includes a touch screen) and a connector 114 (in some optional embodiments, the connector includes a hinge). An input command window 113 is embedded in the upper outer shell 111 and is used to input the target temperature of the solution contained in the reaction tube assembly. Specifically, it is used to input key parameters such as the set target temperature, the target temperature holding time, and the temperature change rate. In some optional embodiments, the input command window 113 is also used to input the solution volume in the reaction tube assembly 2. The upper outer shell 111 and the lower outer shell 112 are connected by a flip-top connector 114 (in some optional embodiments, the upper outer shell 111 and the lower outer shell 112 are connected by a hinge flip-top), which facilitates the placement and removal of the reaction tube and provides a stable and closed reaction environment for the internal main heating module group. In some optional embodiments, the lower outer shell 112 is rigidly fixed to the main heating module group 12 by hexagonal studs to ensure the stability of the whole structure.

[0075] like Figure 5As shown, the temperature control device includes a main heating module group 12 and a control module group (not shown in the figure). The main heating module group includes multiple independent temperature control zones, each of which includes a reaction tube heat sink, a first temperature sensor 128, and a first semiconductor cooling chip 123. In some optional embodiments, the reaction tube heat sink is made of a high thermal conductivity material, including an integral reaction tube heat sink 122 and / or a nine-unit reaction tube heat sink 127. In some optional embodiments, an independent temperature control zone may include one or more reaction tube heat sinks. Each reaction tube heat sink is isolated by an air gap and the upper gap is sealed with a heat insulation component made of a heat insulation material such as closed-cell silicone to insulate against heat and prevent foreign objects. The heat insulation component 121 is disposed on the upper edge of the reaction tube heat sink. In some optional embodiments, one reaction tube heat sink can insert one or more reaction tube assemblies 2, for example, a nine-unit reaction tube heat sink 127 can insert three nine-unit reaction tubes 21, and an integral reaction tube heat sink 122 can insert three integral reaction tubes 22.

[0076] like Figure 5 As shown, in some optional embodiments, the reaction tube heat sink is used to place the reaction tube assembly 2; the first temperature sensor 128 is disposed on the side wall of the reaction tube heat sink for monitoring the temperature of the reaction tube heat sink; the first semiconductor refrigeration chip 123 is disposed below the reaction tube heat sink for regulating the temperature of the reaction tube heat sink. In some optional embodiments, one or two first temperature sensors (such as PT100, PT1000, etc.) 128 are arranged on the side of each reaction tube heat sink to monitor the temperature in real time; in some optional embodiments, the main heating module group further includes: a heat insulation component 121 to seal the gap above the reaction tube heat sink; a heat sink 124 disposed below the first semiconductor refrigeration chip 123 for dissipating the heat generated by the first semiconductor refrigeration chip 123; in some optional embodiments, the nine-unit reaction tube heat sink 127 and / or the integrated reaction tube heat sink 122 are both connected to the heat sink 124 by heat insulation bolts. In some optional embodiments, the main heating module assembly 12 further includes a fixing member 125 disposed below the heat sink 124; in some optional embodiments, the main heating module assembly 12 further includes a fan 126 disposed below the fixing member 125. The heat sink 124 efficiently conducts heat away from the first thermoelectric cooler (TEC) 123, the fan 126 enhances heat dissipation, and the fixing member 125 is secured to the lower housing 112 with hexagonal studs.

[0077] In some alternative embodiments, the first semiconductor cooling chip 123 is located between the reaction tube heat sink and the copper heat sink 124.

[0078] In some optional embodiments, the main heating module group includes 4, 5, or 6 independent temperature control zones; in some optional embodiments, the main heating module group includes 4 independent temperature control zones, each of which includes an independent reaction tube heat sink, an independent first temperature sensor 128, and an independent first semiconductor cooling chip 123. Each independent temperature control zone can be independently set and precisely controlled in temperature without interference, allowing for the simultaneous parallel fabrication of DNA nanomachines with different temperature requirements.

[0079] like Figure 3 As shown, in some optional embodiments, the reaction tube assembly 2 includes a flat reaction tube; in some optional embodiments, the flat reaction tube includes a nine-tube reaction assembly 21 and / or a single-piece reaction tube 22. In some optional embodiments, the flat reaction tube includes a U-shaped flat reaction tube. Compared to traditional conical PCR tubes, the reaction tubes of the present invention significantly improve the temperature variation efficiency and temperature uniformity of the liquid by optimizing the shape of the tube and the thickness of the liquid layer, ensuring the consistency and stability of DNA nanomachine assembly, while also meeting the requirements for large-volume single-batch manufacturing. Compared with the solution temperature distribution in traditional conical reaction tubes, the solution temperature in the U-shaped flat reaction tubes of the present invention is more uniform, and less heat is lost in the upper solution. In some optional embodiments, the reaction tube assembly 2 is made of biocompatible materials such as polypropylene (PP), providing a closed and uniform reaction environment for the self-assembly of DNA nanomachines.

[0080] In some optional implementations, the control module group is used to obtain the monitored temperature of the first temperature sensor 128 from the main heating module group 12 and the target temperature of the solution contained in the reaction tube assembly 2 from the user terminal. The control module group is configured to input the monitored temperature of the first temperature sensor 128 and the target temperature into a pre-trained solution temperature prediction model, obtain the predicted temperature of the solution contained in the reaction tube assembly 2 output by the solution temperature prediction model, and generate temperature control information for the reaction tube heat sink based on the predicted temperature and the target temperature. At the same time, the control module group controls the temperature of the first semiconductor refrigeration chip 123 of the reaction tube heat sink based on the generated temperature control information of the reaction tube heat sink.

[0081] like Figure 6As shown, in some optional embodiments, the temperature control device 1 further includes a heat cover module group 13; the heat cover module group 13 includes a plurality of heat cover plates 131 corresponding to a plurality of independent temperature control zones, a plurality of second temperature sensors 134 corresponding to a plurality of heat cover plates, a plurality of second thermoelectric coolers 132 corresponding to a plurality of heat cover plates, and a thermoelectric cooler fixing plate 133 for fixing the second thermoelectric coolers 132; in some optional embodiments, the heat cover plates 131 are disposed above the independent temperature control zones of the main heating module group 12; the second temperature sensors 134 are disposed on the heat cover plates 131 for monitoring the temperature of the heat cover plates 131; the second thermoelectric coolers 132 are disposed above the heat cover plates 131 for regulating the temperature of the heat cover plates 131; in some optional embodiments, the second temperature sensors 134 are embedded in the heat cover plates 131 to monitor the temperature of the heat cover plates 131 in real time.

[0082] In some optional embodiments, the control module group (not shown in the figure) further includes acquiring the monitoring temperature of the second temperature sensor 134, feeding back the temperature control information of the hot cover plate 131 to the second semiconductor cooling chip 132 based on the monitoring temperature of the second temperature sensor 134 and the monitoring temperature of the first temperature sensor 128, and controlling the temperature of the hot cover plate 131 to be higher than the temperature of the reaction tube hot seat, so as to prevent the formation of condensate on the inner wall of the cover of the reaction tube and ensure that the concentration of the reaction solution is constant.

[0083] In some optional embodiments, the control module group (not shown) further includes acquiring the monitored temperature of the second temperature sensor 134. When the DNA nanomachine manufacturing system is powered on, the control module group feeds back information to the second thermoelectric cooler 132 to maintain a fixed temperature above the reaction temperature of the heated cover plate 131, and adjusts the power of the second thermoelectric cooler 132 in real time based on the monitored temperature of the second temperature sensor 134. The heated cover plate 131 provides uniform heat transfer under relatively high constant temperature conditions (e.g., 100°C), preventing solution evaporation and condensation.

[0084] In some optional embodiments, thermally conductive grease with a thickness of about 0.1 mm is applied to the contact surface of the first thermoelectric cooler 123 with the reaction tube heat sink, the contact surface of the second thermoelectric cooler 132 with the heat cover plate 131, the contact surface of the first temperature sensor 128 with the reaction tube heat sink, and the contact surface of the second temperature sensor 134 with the heat cover plate 131 to prevent poor surface adhesion from reducing heat transfer performance.

[0085] The DNA nanomachine manufacturing system enables precise temperature control of the solution, effectively mitigating temperature distortion caused by differences in tube wall thermal resistance and liquid layer thickness. The method is achieved collaboratively by the control logic of the temperature regulation information from the control module group and a neural network model.

[0086] In some alternative implementations, the real-time solution temperature prediction method is performed during the operation of the DNA nanomachine manufacturing system according to the following steps: (1) Real-time data acquisition: The first temperature sensor acquires the temperature data of the heating module in real time at a specific frequency (e.g., 2Hz); (2) Real-time feature construction: Based on the current and historical heating module temperature data, the 11-dimensional feature vector is calculated in real time according to the S3 step method; (3) Sliding window update: Maintain a feature window of length W. Newly generated... Upon entering the window, the oldest features are removed, forming an inference input block consistent with the S4 step format; (4) Standardization processing: Call the standardized parameters saved in S5 to process the input block; (5) Model inference: Input the standardized features into the solidified model to obtain the normalized prediction output; (6) Inverse standardization: The model output is inversely transformed into the actual temperature unit to obtain the predicted value of the solution temperature.

[0087] This invention compares the predicted solution temperature with the target temperature to calculate the temperature difference. The invention achieves precise closed-loop feedback of solution temperature through PWM regulation. Using the aforementioned machine learning-based real-time dynamic prediction method for solution temperature, the system can accurately predict the actual temperature of the solution inside the reaction tube in real time, taking the heating module temperature as input. This effectively overcomes the temperature distortion problems caused by differences in tube wall thermal resistance and liquid layer thickness in traditional methods, achieving accurate temperature control in the DNA nanomachine manufacturing process.

[0088] Unless otherwise defined, the technical terms used in the following embodiments have the same meaning as commonly understood by those skilled in the art. Unless otherwise specified, the experimental reagents used in the following embodiments are conventional biochemical reagents; the raw materials, instruments, and equipment used in the following embodiments can all be obtained commercially or through existing methods; unless otherwise specified, the amounts of experimental reagents used are the amounts used in conventional experimental operations; unless otherwise specified, the experimental methods are conventional methods. It should be further noted that the following descriptions are merely exemplary and not intended to limit the specific scope of the invention. Moreover, the comparative examples below are selected to compare with the technical solutions of the present invention to demonstrate the advancement of the present invention, and do not necessarily represent prior art in this technical field.

[0089] Example 1 This embodiment provides a pre-training method for a pre-trained solution temperature prediction model, including: (I) Construction of Temperature Prediction Model This invention constructs a deep neural network model based on Long Short-Term Memory (LSTM) as a temperature prediction model. The temperature prediction model employs a two-layer LSTM network (in this embodiment, the hidden layer dimensions are 64 and 32, respectively) plus a fully connected output layer. This model learns the nonlinear heat conduction characteristics between the heat source and the solution to predict the estimated temperature of the current solution based on the current and historical heating module temperatures.

[0090] (II) Temperature Prediction Model Training S1: Data Acquisition and Preprocessing. Simultaneously acquire two types of temperature data under different operating conditions (e.g., different solution volumes, different temperature control programs): one is the temperature of the reaction tube heating seat (heat source temperature, ...). T block The temperature is measured by a first temperature sensor (such as a PT1000 temperature sensor) installed on the reaction tube heating element; the second is the temperature of the solution in the reaction tube assembly. T sol The data was measured by a thermocouple inserted into the solution inside the reaction tube. Both types of data were acquired synchronously at the same sampling frequency (2Hz). Each data record includes time t and the corresponding time. T block ( t )and T sol ( t The collected data undergoes format standardization and cleaning, including: numerical type conversion, handling of missing / outlier values, and removal of duplicate rows. After format standardization and cleaning, the data needs to be filtered in step S2.

[0091] S2: Screening of operating condition similarity data. The dynamic relationship between the heating module temperature and the solution temperature differs under different operating conditions. This is achieved through calculation... T block and T sol The lagged correlation between them can quantify their dynamic relationship. Similarity filtering is performed on each training file to be tested within a preset search window. Internal calculation:

[0092] in, denoted as Pearson correlation coefficient, which measures the strength of the linear correlation between two sequences in terms of their trend of change, and its value ranges from [-1, 1]. It's about the lag step size. k The function reflects the relevance score at a specific delay; The first subset of the heating module sequence, i.e. , The latter subset of the solution sequence is used to ensure that the two sequences have the same length through this "sliding window alignment"; L represents the total length of the original temperature sampling sequence; k represents the hysteresis search step size, i.e., the solution temperature sequence Move the sampling points backward by k to match the temperature of the heating module. Alignment is used to simulate the physical time required for heat transfer; This represents the maximum lag search step size, used to limit the search range for lag correlation calculation. In the example, it can be 10, 20, or 30. This indicates that the Pearson correlation coefficient is... Lag step size for achieving the maximum value ; The main lag step, i.e., the delay corresponding to the Pearson correlation coefficient reaching its maximum value, reflects the time delay of heat transfer from the heating block to the solution under this operating condition. This represents the maximum correlation coefficient value.

[0093] when (Maximum allowable hysteresis threshold) or When the minimum relevance threshold is reached, it will not participate in the construction of subsequent training samples; in a specific instance, it is set... and The values ​​are 100 and 0.85 respectively.

[0094] By filtering based on operating condition similarity, training samples with excessive latency or mismatched dynamic characteristics are removed, allowing the model to focus on learning the temperature dynamic characteristics under the target operating condition, thereby improving the prediction accuracy under that condition. After filtering, the final data constructed by multi-dimensional feature engineering is obtained, and the final data is randomly divided into training set, validation set, and test set according to a certain ratio (e.g., 8:1:1).

[0095] S3: Multidimensional Feature Engineering Construction. For the temperature data after S2 filtering, an 11-dimensional feature column vector is constructed for the temperature data at each time point to represent the temperature rise trend, fluctuation, and thermal inertia state; each feature uses only the temperature data at the current time and historical time points.

[0096] The 11-dimensional temperature characteristics, as shown above, include 1-dimensional basic temperature characteristics, 2-dimensional temperature difference characteristics, 3-dimensional temperature rolling statistics characteristics, 3-dimensional hysteresis temperature characteristics, and 2-dimensional temperature thermal inertia state characteristics. The basic temperature characteristic values ​​corresponding to the basic temperature characteristics include , As shown in formula (1-1): (1-1), in, This represents the baseline temperature characteristic value at time t. This represents the temperature monitored by the first temperature sensor at time t, i.e., the temperature sensor set up on the heat source. The temperature difference eigenvalues ​​corresponding to the temperature difference features include first-order temperature difference eigenvalues. Second-order temperature difference eigenvalues , As shown in formula (1-2), As shown in formula (1-3): (1-2) (1-3) in, Indicates the first t The first-order temperature difference eigenvalue at time t. Indicates the first t The temperature monitored by the first temperature sensor at time -1. Indicates the first t The second-order temperature difference eigenvalue at time t. Indicates the first t The first-order temperature change at time -1; The temperature rolling statistical feature includes the average temperature value at 10 sampling times. 60 sampling time point temperature average characteristic value Characteristic values ​​of temperature standard deviation at 10 sampling times , As shown in formula (1-4), As shown in formula (1-5), As shown in formula (1-6): (1-4) (1-5) (1-6) in, Indicates the historical moment up to the 1st t The characteristic value of the average temperature monitored by the first temperature sensor at 10 sampling times. Indicates the historical moment up to the 1st t The characteristic value of the average temperature monitored by the first temperature sensor at 60 sampling times. Indicates a historical period that goes back in time. Indicates the historical moment up to the 1st t The standard deviation characteristic values ​​of the monitored temperature of the first temperature sensor at 10 sampling times. Indicates the first tiThe temperature monitored by the first temperature sensor at time t; when t < 60, the system uses boundary padding (such as forward padding with the first available value) to ensure online availability.

[0097] The hysteresis temperature characteristic corresponding to the hysteresis temperature characteristic includes 10-step hysteresis temperature characteristic values. 30-step hysteresis temperature characteristic value 50-step hysteresis temperature characteristic value , As shown in formula (1-7), As shown in formula (1-8), As shown in formula (1-9): (1-7) (1-8) (1-9) in, This indicates the temperature monitored by the first temperature sensor at the 10th sampling time prior to the current time. This indicates the temperature monitored by the first temperature sensor at the 30th sampling time prior to the current time. This represents the monitored temperature of the first temperature sensor at the 50th sampling time prior to the current time, and its characteristic value. Indicates the first t 10-step lag temperature at time Indicates the first t 30-step lag temperature at any given time This represents the 50-step lag temperature at time t; when (t < 10 / 30 / 50), the system uses boundary padding (such as forward padding with the first available value) to ensure online availability.

[0098] The temperature thermal inertia state characteristic values ​​corresponding to the temperature thermal inertia state characteristics include the fast exponential moving average temperature characteristic values. Slow exponential moving average temperature characteristic value , As shown in formula (1-10), As shown in formula (1-11): (1-10) (1-11) in, This represents the characteristic value of the rapid exponential moving average temperature. Indicates the fast smoothing coefficient. express t The rapid exponential moving average temperature at time -1 This represents the characteristic value of the slow exponential moving average temperature. Indicates the slow smoothing coefficient. > , express t Slow exponential moving average temperature at time -1. , This coefficient (represents short-term and long-term thermal dynamics respectively) , This parameter can be used as a hyperparameter and adaptively adjusted for different working conditions through optimization algorithms (such as grid search). The value ranges from 0.1 to 0.5. The value ranges from 0.01 to 0.05.

[0099] Column vector composed of 11-dimensional features This data was used to construct subsequent sliding window samples.

[0100] S4: Sliding Window Sample Construction. The feature sequence generated in step S3 is transformed into the input for the subsequent temperature prediction model. The window length is set to W (160 in this example), and the prediction step size is H (0 in one example, i.e., predicting the solution temperature at the same time). The input for the model is constructed by sliding the feature vector along the time axis. and output sample: ; ; The above samples are packaged into binary array files, and meta-information containing normalization parameters is output. These binary files are the direct input source for the next step of the training process.

[0101] S5: Model Training and Consolidation: The temperature prediction model constructed in this invention has the following architecture: Input Layer: Receives a feature time-series window of dimension (W, 11) generated by step S4, where W is the time step and 11 is the feature dimension extracted at each time step. Hidden Layer: Consists of a two-layer LSTM network. The first LSTM hidden layer has a dimension of 64, used to extract high-dimensional dynamic thermal characteristics from the original features; the second LSTM hidden layer has a dimension of 32, used to further compress the time-series features and enhance the extraction of long-range dependencies. Output Layer: Maps the terminal hidden states of the LSTM to scalar space through a fully connected layer, outputting the predicted value of the solution temperature at the current time step.

[0102] The model training objective is to minimize the average loss of the training set samples. During model training, dropout and gradient pruning techniques are used to optimize the model's generalization ability under different operating conditions. After training, the model parameters (resulting in a fixed model) and standardized parameters are saved to obtain a pre-trained temperature prediction model, which is then used for subsequent test set evaluation and deployment for inference.

[0103] Example 2 This embodiment provides a method for predicting solution temperature, including the following steps: 1) Obtain the current heating temperature and historical heating temperature data of the solution; the heating temperature is the temperature of the heat source used to heat the solution; 2) Based on the current heating temperature and the historical heating temperature described in step 1), derive the 11-dimensional temperature feature value for the current moment (same as the 11-dimensional temperature feature value listed in Example 1); derive the 11-dimensional temperature feature value for the historical moment using the same method (same as the 11-dimensional temperature feature value listed in Example 1); then construct the current moment input sequence including the 11-dimensional temperature feature values ​​for the current moment and the historical moment; the current moment input sequence is... ; in, t Indicates the current moment. express t Temperature characteristic value at time, Indicates the time step. This represents a column vector formed by the temperature eigenvalues ​​corresponding to the 11-dimensional temperature features. express The column vector formed by the temperature feature values ​​corresponding to the 11-dimensional temperature features at time t is specifically as follows: , express The column vector formed by the temperature feature values ​​corresponding to the 11-dimensional temperature features at time t is specifically as follows: , express The column vector formed by the temperature feature values ​​corresponding to the 11-dimensional temperature features at time t is specifically as follows: , express The column vector formed by the temperature feature values ​​corresponding to the 11-dimensional temperature features at time t is specifically as follows: ; 3) Input the current time sequence obtained in step 2) into the pre-trained solution temperature prediction model obtained in Example 1, and the pre-trained solution temperature prediction model outputs... ; in, To predict the step size, This indicates the current temperature and / or future temperature of the solution contained in the reaction tube assembly.

[0104] Example 3 This embodiment provides a solution temperature prediction device that can implement the solution temperature prediction method of Embodiment 2, comprising: The data acquisition module is used to acquire the current heating temperature and historical heating temperature data of the solution; the heating temperature is the temperature of the heat source used to heat the solution. The feature construction module is used to construct an 11-dimensional temperature feature value for the current moment based on the heating temperature data at the current moment and the heating temperature data at the historical moment obtained by the data acquisition module, and to construct an 11-dimensional temperature feature value for the historical moment in the same way, and to construct an input sequence for the current moment that includes the 11-dimensional temperature feature values ​​at the current moment and the historical moment. The prediction module includes a pre-trained solution temperature prediction model, which is used to input the current time input sequence constructed by the feature construction module, and predicts and outputs the predicted temperature of the solution at the current time and / or future time based on the current time input sequence.

[0105] The training sample set acquisition module is used to construct 11-dimensional temperature feature values ​​for the current time and the historical time from the heating temperature data acquired by the data acquisition module. It also constructs the current time input sequence including the 11-dimensional temperature feature values ​​of the current time and the historical time. The current time input sequence and the measured solution temperature at the same time and / or future time are used to construct a training sample. Multiple training samples at different times are obtained to obtain the training sample set. The model pre-training module is used to train the long short-term memory network model using the training sample set to obtain a pre-trained solution temperature prediction model. The data acquisition module is also used to acquire the measured solution temperature at the current time and / or future time.

[0106] The comparison between the predicted current temperature of the solution in the reaction tube of the DNA nanomachine manufacturing system in this embodiment and the actual measured temperature is as follows: Figure 7 As shown.

[0107] Example 4 This embodiment provides a DNA nanomachine manufacturing system, which includes a temperature control device 1 and a reaction tube assembly 2. The temperature control device 1 includes a housing 11, a main heating module group 12, and a hot cover module group 13. The housing 11 includes an upper housing 111, a lower housing 112, an input command window 113 (the input command window 113 is a touch screen), and a connector 114 (the connector is a hinge). The input command window 113 is embedded in the upper housing 111 and is used to input the target temperature of the solution contained in the reaction tube assembly. Specifically, it is used to input key parameters such as the set target temperature, the target temperature holding time, and the temperature change rate, and also to input the solution volume in the reaction tube assembly 2. The upper housing 111 and the lower housing 112 are connected by a flip-top via the connector 114 (the upper housing 111 and the lower housing 112 are connected by a hinge), which facilitates the placement and removal of the reaction tube and provides a stable and sealed reaction environment for the internal temperature control module. The lower housing 112 is rigidly fixed to the main heating module group 12 by hexagonal studs to ensure the stability of the entire structure.

[0108] The temperature control device includes a main heating module group 12 and a control module group. The main heating module group includes four independent temperature control zones. Each independent temperature control zone includes a reaction tube heat sink, a first temperature sensor 128, and a first semiconductor cooling chip 123. The reaction tube heat sink is made of a high thermal conductivity material and includes two integral reaction tube heat sinks 122 (corresponding to two independent temperature control zones respectively) and two nine-unit reaction tube heat sinks 127 (corresponding to two independent temperature control zones respectively). Each reaction tube heat sink is isolated by an air gap and the upper gap is sealed with a heat insulation component made of heat insulation material such as closed-cell silicone to insulate against heat and prevent foreign objects. The heat insulation component 121 is located on the upper edge of the reaction tube heat sink. The reaction tube heat sink is used to place the reaction tube assembly 2, and each reaction tube heat sink can accommodate three reaction tube assemblies 2. The first temperature sensor 128 is located on the side wall of the reaction tube heat sink and is used to monitor the temperature of the reaction tube heat sink. The first semiconductor cooling chip 123 is located below the reaction tube heat sink and is used to regulate the temperature of the reaction tube heat sink. Each reaction tube heat sink has one or two first temperature sensors (such as PT100, PT1000, etc.) 128 arranged on its side to monitor the temperature in real time. The main heating module group also includes: a heat insulation component 121 to seal the gap above the reaction tube heat sink; a copper heat sink 124, located below the first thermoelectric cooler 123, to dissipate the heat generated by the first thermoelectric cooler 123; the reaction tube heat sink is connected to the heat sink 124 by heat insulation bolts. The main heating module group 12 also includes a fixing component 125 located below the heat sink 124; the main heating module group 12 also includes a fan 126 located below the fixing component 125. The heat sink 124 efficiently conducts away the heat of the first thermoelectric cooler (TEC) 123, the fan 126 enhances heat dissipation, and the fixing component 125 is secured to the lower housing 112 with hexagonal studs.

[0109] The reaction tube assembly 2 includes a U-shaped flat reaction tube; the U-shaped flat reaction tube includes a U-shaped nine-tube reaction tube and a U-shaped integrated reaction tube.

[0110] The control module group is used to obtain the monitored temperature of the first temperature sensor 128 from the main heating module group 12, and to obtain the volume of the solution contained in the reaction tube assembly 2 and the target temperature from the user end. The control module group is configured to include the temperature prediction device and temperature control component of Embodiment 3. The temperature prediction device inputs the monitored temperature and target temperature of the first temperature sensor 128 into the pre-trained solution temperature prediction model obtained in Example 1, and obtains the predicted temperature of the solution contained in the reaction tube assembly 2 output by the solution temperature prediction model. The temperature control component generates temperature control information for the reaction tube hot seat based on the predicted temperature and the target temperature; at the same time, the control module group regulates the temperature of the first semiconductor refrigeration chip 123 based on the generated temperature control information for the reaction tube hot seat.

[0111] The temperature control information for the reaction tube hot seat includes the following: When the difference between the target temperature and the predicted temperature of the solution contained in the reaction tube assembly at the current temperature is greater than 0, temperature control information for increasing the temperature of the reaction tube heat sink is generated. The temperature control information for increasing the temperature of the reaction tube heat sink includes adaptively increasing the output power of the first semiconductor refrigeration chip and driving the first semiconductor refrigeration chip to heat up based on the difference between the volume of the solution contained in the reaction tube assembly and the predicted temperature of the solution contained in the reaction tube assembly at the current temperature. When the difference between the target temperature and the predicted temperature of the solution contained in the reaction tube assembly at the current temperature is less than 0, temperature control information for reducing the temperature of the reaction tube hot seat is generated; the temperature control information for reducing the temperature of the reaction tube hot seat includes adaptively increasing the output power of the first semiconductor refrigeration chip and driving the first semiconductor refrigeration chip to cool down based on the difference between the volume of the solution contained in the reaction tube assembly and the predicted temperature of the solution contained in the reaction tube assembly at the current temperature. When the difference between the target temperature and the predicted temperature of the solution contained in the reaction tube assembly at the current temperature is equal to 0, temperature control information for maintaining the temperature of the reaction tube hot seat is generated. This temperature control information includes adaptively reducing the output power of the first semiconductor cooling chip and reducing the duty cycle based on the difference between the solution volume and the predicted temperature of the solution contained in the reaction tube assembly at the current temperature. The temperature control device 1 also includes a heat cover module group 13; the heat cover module group 13 includes multiple heat cover plates 131 corresponding to multiple independent temperature control zones, multiple second temperature sensors 134 corresponding to multiple heat cover plates, multiple second semiconductor cooling chips 132 corresponding to multiple heat cover plates, and a semiconductor cooling chip fixing plate 133 for fixing the second semiconductor cooling chips 132; the heat cover plates 131 are disposed above the independent temperature control zones of the main heating module group 12; the second temperature sensors 134 are disposed on the heat cover plates 131 for monitoring the temperature of the heat cover plates 131; the second semiconductor cooling chips 132 are disposed above the heat cover plates 131 for regulating the temperature of the heat cover plates 131; the second temperature sensors 134 are embedded in the heat cover plates 131 to monitor the temperature of the heat cover plates 131 in real time.

[0112] The control module group also includes acquiring the monitored temperature of the second temperature sensor 134. When the DNA nanomachine manufacturing system is powered on, the control module group feeds back the information to the second semiconductor cooler 132 to regulate the temperature of the heated cover plate 131 to a fixed 100°C higher than the reaction temperature, and adjusts the second semiconductor cooler 132 in real time according to the monitored temperature of the second temperature sensor 134. The heated cover plate 131 provides uniform heat transfer under a relatively high constant temperature condition (100°C), preventing the solution in the reaction tube assembly from evaporating and condensing.

[0113] Thermal grease with a thickness of about 0.1 mm is applied to the contact surfaces of the reaction tube heat sink and the heat cover plate 131, the contact surface of the first temperature sensor 128 and the reaction tube heat sink, and the contact surface of the second temperature sensor 134 and the heat cover plate 131 to prevent poor surface adhesion from reducing heat transfer performance.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions 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 solution temperature prediction method based on machine learning, characterized in that, Includes the following steps: 1) Obtain the current heating temperature and historical heating temperature data of the solution; the heating temperature is the temperature of the heat source used to heat the solution; 2) Based on the current heating temperature and the historical heating temperature described in step 1), derive at least 11-dimensional temperature feature values ​​for the current time; derive at least 11-dimensional temperature feature values ​​for the historical time using the same method; Then, construct the current time input sequence, which includes at least 11-dimensional temperature feature values ​​of the current time and historical times; 3) Input the current time input sequence obtained in step 2) into the pre-trained solution temperature prediction model, and the pre-trained solution temperature prediction model outputs the predicted temperature of the solution at the current time and / or future time. The pre-trained solution temperature prediction model is obtained by training a long short-term memory network model based on a training sample set. Each training sample in the training sample set includes an input sequence sample at the current time and a target value sample at the same time. The input sequence sample at the current time includes at least 11-dimensional temperature feature values ​​of the solution at the current time and at least 11-dimensional temperature feature values ​​at historical times. The target value sample at the same time includes the solution temperature at the current time and / or the measured solution temperature at future times.

2. The solution temperature prediction method based on machine learning according to claim 1, characterized in that, The input sequence at the current time mentioned in step 2) includes an 11-dimensional temperature feature value column vector at the current time and an 11-dimensional temperature feature value column vector at historical time. And / or, the current input sequence sample mentioned in step 3) includes an 11-dimensional temperature feature value column vector at the current time and an 11-dimensional temperature feature value column vector at historical times; And / or, the 11-dimensional temperature feature values ​​include 1-dimensional basic temperature feature values, 2-dimensional temperature difference feature values, 3-dimensional temperature rolling statistical feature values, 3-dimensional hysteresis temperature feature values, and 2-dimensional temperature thermal inertia state feature values.

3. The solution temperature prediction method based on machine learning according to claim 2, characterized in that, The basic temperature characteristic value includes , As shown in formula (1-1): （1-1） ， in, Indicates the current moment. Indicates the first t The baseline temperature characteristic value at time t. Indicates the first t The solution heating temperature is measured at all times, and the heating temperature is obtained by a temperature sensor set in the heat source. The temperature difference characteristic values ​​include first-order temperature difference characteristic values. Second-order temperature difference eigenvalues , As shown in formula (1-2), As shown in formula (1-3): （1-2）， （1-3）， in, Indicates the first t The first-order temperature difference eigenvalue at time t. Indicates the first t- The solution heating temperature at time 1. Indicates the first t The second-order temperature difference eigenvalue at time t. Indicates the first t- The first-order temperature change at time 1; The temperature rolling statistical feature value includes the average temperature feature value at 10 sampling times. 60 sampling time point temperature average characteristic value Characteristic values ​​of temperature standard deviation at 10 sampling times , As shown in formula (1-4), As shown in formula (1-5), As shown in formula (1-6): （1-4）， （1-5）， （1-6）， in, Indicates the historical moment up to the 1st t The characteristic value of the average heating temperature of the solution at 10 sampling times. Indicates the historical moment up to the 1st t The characteristic value of the average heating temperature of the solution at 60 sampling times. Indicates a historical period that goes back in time. Indicates the historical moment up to the 1st t The standard deviation eigenvalues ​​of the solution heating temperature at 10 sampling times. Indicates the first ti The solution heating temperature at any given time; The hysteresis temperature characteristic value includes 10 steps of hysteresis temperature characteristic value. 30-step hysteresis temperature characteristic value 50-step hysteresis temperature characteristic value , As shown in formula (1-7), As shown in formula (1-8), As shown in formula (1-9): （1-7）， （1-8）， （1-9）， in, This indicates the solution heating temperature at the 10th sampling time prior to the current time. This indicates the solution heating temperature at the 30th sampling time prior to the current time. This represents the solution heating temperature at the 50th sampling time prior to the current time, and its characteristic value. Indicates the first t 10-step lag temperature at time Indicates the first t 30-step lag temperature at any given time Indicates the first t Temperature at 50-step lag time; The temperature thermal inertia state characteristic values ​​include the fast exponential moving average temperature characteristic values. Slow exponential moving average temperature characteristic value , As shown in formula (1-10), As shown in formula (1-11): （1-10）， （1-11）， in, This represents the characteristic value of the rapid exponential moving average temperature. Indicates the fast smoothing coefficient. express t- The rapid exponential moving average temperature at time 1. This represents the characteristic value of the slow exponential moving average temperature. Indicates the slow smoothing coefficient. > , express t- The slow exponential moving average temperature at time 1.

4. The solution temperature prediction method based on machine learning according to claim 2, characterized in that, The input sequence at the current time mentioned in step 2) includes ; in, t Indicates the current moment. express t Temperature characteristic value at time, Indicates the time step. This represents the column vector formed by the 11-dimensional temperature eigenvalues. express The column vector formed by the 11-dimensional temperature eigenvalues ​​at time t, express The column vector formed by the 11-dimensional temperature eigenvalues ​​at time t, express The column vector formed by the 11-dimensional temperature eigenvalues ​​at time t, express The column vector formed by the 11-dimensional temperature eigenvalues ​​at time t; And / or, the target value includes ; in, To predict the step size, This indicates the current temperature and / or future temperature of the solution.

5. The solution temperature prediction method based on machine learning according to claim 2, characterized in that, The input sequence samples at the current time mentioned in step 3) include ; in, t Indicates the current moment. express t Temperature characteristic value at time, Indicates the time step. This represents the column vector formed by the 11-dimensional temperature eigenvalues. express The column vector formed by the 11-dimensional temperature eigenvalues ​​at time t, express The column vector formed by the 11-dimensional temperature eigenvalues ​​at time t, express The column vector formed by the 11-dimensional temperature eigenvalues ​​at time t, express The column vector formed by the 11-dimensional temperature eigenvalues ​​at time t; The target value sample was obtained by actual measurement using a thermocouple temperature sensor inserted into the solution; And / or, the long short-term memory network model described in step 3) includes two layers, with the first hidden layer having dimensions set to 16-128 and the second hidden layer having dimensions set to 16-128.

6. A solution temperature prediction device, characterized in that, The solution temperature prediction device includes: The data acquisition module is used to acquire the current heating temperature and historical heating temperature data of the solution; the heating temperature is the temperature of the heat source used to heat the solution. The feature construction module is used to construct at least 11-dimensional temperature feature values ​​for the current time based on the heating temperature data at the current time and the heating temperature data at the historical time obtained by the data acquisition module, and to construct at least 11-dimensional temperature feature values ​​for the historical time in the same way, and to construct the current time input sequence including at least 11-dimensional temperature feature values ​​for the current time and the historical time. The prediction module includes a pre-trained solution temperature prediction model, which is used to input the current time input sequence constructed by the feature construction module, and predicts and outputs the predicted temperature of the solution at the current time and / or future time based on the current time input sequence.

7. The solution temperature prediction device according to claim 6, characterized in that, The data acquisition module is also used to acquire the measured solution temperature at the current time and / or future time. The solution temperature prediction device further includes: The training sample set acquisition module is used to construct at least 11-dimensional temperature feature values ​​for the current time and at least 11-dimensional temperature feature values ​​for the historical time from the heating temperature data at the current time and the heating temperature data at the historical time acquired by the data acquisition module, and to construct the current time input sequence including the at least 11-dimensional temperature feature values ​​at the current time and the historical time. The current time input sequence and the measured solution temperature at the same time and / or future time are used to construct a training sample, and multiple training samples at different times are obtained to obtain the training sample set. The model pre-training module is used to train the long short-term memory network model using the training sample set to obtain a pre-trained solution temperature prediction model.

8. An electronic device for predicting solution temperature, characterized in that, The solution temperature prediction electronic device includes a memory, a processor, and a computer program stored in the memory, wherein the processor implements the method of any one of claims 1-5 when executing the computer program.

9. The application of the machine learning-based solution temperature prediction method according to any one of claims 1-5, the solution temperature prediction device according to claim 6 or 7, or the solution temperature prediction electronic device according to claim 8 in a DNA manufacturing system.

10. The application according to claim 9, characterized in that, The DNA manufacturing system includes a DNA nanomachine manufacturing system and a PCR amplification instrument; And / or, the DNA manufacturing system further includes a solution temperature control component; The solution temperature control component is configured to: acquire the target temperature of the solution to be controlled, acquire the predicted temperature of the solution to be controlled predicted by the machine learning-based solution temperature prediction method according to any one of claims 1-5, or the solution temperature prediction device according to claim 6 or 7, or the solution temperature prediction electronic device according to claim 8, and generate heat source temperature control information for controlling the temperature of the solution to be controlled based on the predicted temperature and the target temperature. The heat source is used to heat the solution to be controlled; The heat source temperature control information generated for regulating the temperature of the solution to be controlled includes: When the difference between the target temperature and the predicted temperature is greater than 0, temperature control information to increase the temperature of the heat source is generated. When the difference between the target temperature and the predicted temperature is less than 0, temperature control information to reduce the temperature of the heat source is generated. When the difference between the target temperature and the predicted temperature is equal to 0, temperature control information for maintaining the temperature of the heat source is generated.

Images