Heat flow test temperature control method and device for dynamically setting PID parameters based on machine learning

Abstract

The invention discloses a heat flow test temperature control method and device for dynamically setting PID parameters based on machine learning, and belongs to the technical field of machine learning and heat flow tests.The method comprises the steps that multiple state parameters in the heat flow test process are collected; inputting the collected state parameters into a pre-trained neural network model, and outputting dynamically-set PID parameters by using the pre-trained neural network model according to the input state parameters; the output power of the heating equipment is adjusted in real time according to the dynamically-set PID parameters, and temperature control is achieved; and continuously correcting the neural network model through a temperature feedback signal to realize adaptive optimization of PID parameters. By adopting the temperature control method and device for the heat flow test, the accuracy and efficiency of temperature control in the heat flow test can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical fields of machine learning and heat flux test technology, and particularly to a temperature control method and device for heat flux test based on dynamically tuning PID parameters by machine learning. Background Art

[0002] Heat flux tests are mainly used to simulate the thermal characteristics of materials, components, and products in complex thermal environments, and to evaluate their reliability and stability under high-temperature thermal conditions. Heat flux tests are of great significance in the research and engineering applications of fields such as aerospace and electronic equipment. At different stages of heat flux tests, such as the heating-up, steady-state, and cooling-down processes, environmental disturbances and material characteristic parameters (such as specific heat capacity, thermal conductivity, etc.) will cause fixed PID parameters to be difficult to adapt to dynamic adjustments, thereby affecting the temperature control accuracy. Whether the method of tuning PID parameters can perform differential processing according to the heat flux test stage and realize dynamic adjustment of PID parameters according to the characteristics of test materials to adapt to various working conditions and complete high-precision temperature control determines its application value. However, traditional PID tuning methods have the problem of poor temperature control effect due to the lack of dynamic tuning. Summary of the Invention

[0003] The present invention provides a temperature control method and device for heat flux test based on dynamically tuning PID parameters by machine learning to solve the technical problems that traditional PID tuning methods lack dynamic tuning, and existing methods fail to fully consider differential processing of heat flux test stages and material characteristics of tests, resulting in poor temperature control effect.

[0004] To solve the above technical problems, the present invention provides the following technical solutions:

[0005] On the one hand, the present invention provides a temperature control method for heat flux test based on dynamically tuning PID parameters by machine learning. The temperature control method for heat flux test based on dynamically tuning PID parameters by machine learning includes:

[0006] Collecting a plurality of state parameters during the heat flux test process;

[0007] Inputting the collected state parameters into a pre-trained neural network model, and using the pre-trained neural network model to output dynamically tuned PID parameters according to the input state parameters;

[0008] According to the dynamically tuned PID parameters, adjusting the output power of the heating device in real time to achieve temperature control;

[0009] Continuously correcting the neural network model through the temperature feedback signal to achieve adaptive optimization of PID parameters.

[0010] Further, the state parameters include: the current temperature, target temperature, temperature change gradient, specific heat capacity of the material, thermal conductivity of the material, and the current test stage during the heat flow test; wherein, the test stage is divided into three different stages: the heating stage, the steady-state stage, and the cooling stage, and is represented by one-hot encoding.

[0011] Further, the neural network model is a multi-branch dynamic neural network with a hybrid architecture, which includes: an input layer, a material encoding module, a stage perception module, a heating branch, a steady-state branch, a cooling branch, a feature fusion module, and an output layer;

[0012] The process of obtaining the dynamically tuned PID parameters using the multi-branch dynamic neural network includes:

[0013] The collected state parameters enter the multi-branch dynamic neural network through the input layer;

[0014] The material encoding module processes the specific heat capacity and thermal conductivity of the material through a fully connected layer to generate a material feature vector; the stage perception module concatenates the one-hot encoding of the test stage with the material feature vector and sends it to a gated unit; the gated unit calculates a three-dimensional probability distribution through a preset weight matrix and bias to represent the activation weights of the heating branch, the steady-state branch, and the cooling branch; the branch with the maximum probability is taken as the main path, and the outputs of other branches are set to zero; wherein, the heating branch is used to concatenate the current temperature, target temperature, and temperature change gradient with the material feature vector, send it to the first hidden layer, process it through a fully connected layer in the first hidden layer, and after being activated by LeakyReLU, send it to the second hidden layer, and after passing through a mapping process in the second hidden layer and being activated by LeakyReLU, obtain the output of the heating branch; the steady-state branch is used to concatenate the current temperature, target temperature, and temperature change gradient with the material feature vector and send it to a sparse self-attention layer; use the sparse self-attention layer to calculate Query, Key, and Value, and only retain the top three attention weights during the self-attention calculation process, and set the remaining weights to zero to obtain the attention feature; then, map the attention feature through a fully connected layer and activate it with ReLU to obtain the output of the steady-state branch; the cooling branch is used to concatenate the current temperature, target temperature, and temperature change gradient with the material feature vector, send it to a one-dimensional convolutional layer to extract the temporal feature of the temperature change, and further reduce the dimension through a max pooling layer, and finally process it through a fully connected layer and activate it with Tanh to obtain the output of the cooling branch;

[0015] The feature fusion module fuses the output of the main path with the material feature vector to obtain a fused feature vector; at the output layer, the fused feature vector is mapped to a three-dimensional output through a fully connected layer, and the three-dimensional output directly corresponds to the parameters of the PID controller in the temperature control system, including Kp, Ki, and Kd.

[0016] Furthermore, the training process of the neural network model includes:

[0017] Collect the state parameters during the historical heat flux test, filter out the outliers and smooth the collected state parameters to obtain effective state parameters, and use the effective state parameters to construct a historical test dataset;

[0018] Use the historical test dataset to optimize the network weights of the neural network model through supervised learning, so that the neural network model can dynamically adjust the PID parameters under different heat flux conditions.

[0019] Furthermore, the PID parameter dynamic adjustment strategy is optimized according to the characteristics of the test material. For materials with a thermal conductivity lower than the preset thermal conductivity threshold, increase the integral parameter to enhance the stability of temperature adjustment; for materials with a thermal inertia greater than the preset thermal inertia threshold, adjust the differential parameter to reduce overshoot.

[0020] On the other hand, the present invention also provides a heat flux test temperature control device for dynamically tuning PID parameters based on machine learning. The heat flux test temperature control device for dynamically tuning PID parameters based on machine learning includes:

[0021] A data acquisition module for collecting multiple state parameters during the heat flux test;

[0022] A neural network calculation module for inputting the collected state parameters into a pre-trained neural network model, and using the pre-trained neural network model to output the dynamically tuned PID parameters according to the input state parameters;

[0023] A temperature control execution module for adjusting the output power of the heating device in real time according to the dynamically tuned PID parameters to achieve temperature control;

[0024] A feedback correction module for continuously correcting the neural network model through a temperature feedback signal to achieve adaptive optimization of the PID parameters.

[0025] Furthermore, the state parameters include: the current temperature, target temperature, temperature change gradient, specific heat capacity of the material, thermal conductivity of the material, and the current test stage during the heat flux test; among them, the test stage is divided into three different stages: the heating stage, the steady state stage, and the cooling stage, and one-hot encoding is used to represent it.

[0026] Furthermore, the neural network model is a multi-branch dynamic neural network with a hybrid architecture, which includes: an input layer, a material encoding module, a stage perception module, a heating branch, a steady state branch, a cooling branch, a feature fusion module, and an output layer;

[0027] The process of obtaining the dynamically tuned PID parameters using the multi-branch dynamic neural network includes:

[0028] The collected state parameters enter the multi-branch dynamic neural network through the input layer;

[0029] The material coding module processes the specific heat capacity and thermal conductivity of the material through a fully connected layer to generate a material feature vector; the stage perception module concatenates the one-hot encoding of the test stage with the material feature vector and sends it to the gating unit; the gating unit calculates a three-dimensional probability distribution through a preset weight matrix and bias to represent the activation weights of the heating branch, steady-state branch, and cooling branch; the branch with the maximum probability is taken as the main path, and the outputs of other branches are set to zero; among them, the heating branch is used to concatenate the current temperature, target temperature, and temperature change gradient with the material feature vector, send it to the first hidden layer, be processed by a fully connected layer in the first hidden layer, and after being activated by LeakyReLU, send it to the second hidden layer, and after being mapped in the second hidden layer and activated by LeakyReLU, the output of the heating branch is obtained; the steady-state branch is used to concatenate the current temperature, target temperature, and temperature change gradient with the material feature vector and send it to the sparse self-attention layer; the sparse self-attention layer is used to calculate Query, Key, and Value, and only the top three attention weights are retained during the self-attention calculation, and the remaining weights are set to zero to obtain the attention feature; then, the attention feature is mapped through a fully connected layer and activated by ReLU to obtain the output of the steady-state branch; the cooling branch is used to concatenate the current temperature, target temperature, and temperature change gradient with the material feature vector, send it to a one-dimensional convolutional layer to extract the temporal feature of the temperature change, and further reduce the dimension through a max-pooling layer, and finally, after being processed by a fully connected layer and activated by Tanh, the output of the cooling branch is obtained;

[0030] The feature fusion module fuses the output of the main path with the material feature vector to obtain a fused feature vector; in the output layer, the fused feature vector is mapped to a three-dimensional output through a fully connected layer, and the three-dimensional output directly corresponds to the parameters of the PID controller in the temperature control system, including Kp, Ki, and Kd.

[0031] Further, the training process of the neural network model includes:

[0032] Collect the state parameters during the historical heat flux test, filter out the outliers and smooth the collected state parameters to obtain effective state parameters, and use the effective state parameters to construct a historical test dataset;

[0033] Using the historical test dataset, optimize the network weights of the neural network model through supervised learning, so that the neural network model can dynamically adjust the PID parameters under different heat flux conditions.

[0034] Furthermore, the PID parameter dynamic adjustment strategy is optimized according to the characteristics of the test materials. For materials with a thermal conductivity lower than the preset thermal conductivity threshold, the integral parameter is increased to enhance the stability of temperature adjustment; for materials with a thermal inertia greater than the preset thermal inertia threshold, the differential parameter is adjusted to reduce overshoot.

[0035] On the other hand, the present invention also provides an electronic device, which includes a processor and a memory; wherein, at least one instruction is stored in the memory, and the instruction is loaded and executed by the processor to implement the above method.

[0036] On yet another aspect, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored, and the instruction is loaded and executed by the processor to implement the above method.

[0037] The beneficial effects brought by the technical solution provided by the present invention at least include:

[0038] The present invention adopts a machine learning method to dynamically adjust the PID parameters through a neural network, avoiding manual intervention and improving the temperature control accuracy and stability; through differential processing in the heat flow test stage, the temperature control strategy can be automatically adjusted according to different stages during the test, and the thermal characteristics of the material (such as specific heat capacity) are also used as one of the parameters, making the temperature control more targeted and reducing overshoot and steady-state error; by using temperature error feedback closed-loop optimization, the adaptive learning of the temperature control system is realized, enhancing the ability to adapt to different test conditions; by adopting a software-hardware combined architecture and modular design, it has good scalability and integration, and can adapt to different application requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0040] Figure 1 It is a schematic execution flow diagram of the heat flow test temperature control method for dynamically tuning PID parameters based on machine learning provided by the embodiments of the present invention;

[0041] Figure 2 It is a schematic structural diagram of a multi-branch dynamic neural network with a hybrid architecture provided by the embodiments of the present invention;

[0042] Figure 3 It is a schematic principle diagram of dynamically tuning PID parameters based on machine learning provided by the embodiments of the present invention;

[0043] Figure 4 It is a schematic structural diagram of a temperature control device for a heat flux test that dynamically tunes PID parameters based on machine learning provided by an embodiment of the present invention;

[0044] Figure 5 It is a system block diagram of an electronic device provided by an embodiment of the present invention. Specific embodiments

[0045] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0046] First of all, it should be noted that in the embodiments of the present invention, words such as "exemplarily" and "for example" are used to represent examples, illustrations, or explanations. Any embodiment or design solution described as an "example" in the present invention should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, the use of the word "exemplarily" is intended to present concepts in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one of the two can be selected.

[0047] The first embodiment

[0048] Aiming at the problems that the existing PID parameter dynamic tuning method is difficult to adapt to the differentiated requirements in the heat flux test stage and the influence of different test material characteristics on temperature control stability, this embodiment fully considers key factors such as material thermal characteristics, temperature change trends, and the test stage, and integrates them into machine learning to solve the influence of non-linear factors in the test and improve the temperature control effect. Based on this, this embodiment combines the advantages of machine learning in dynamically tuning PID parameters, the differentiated processing in the heat flux test stage, and the characteristics of test materials, and adopts an architecture that combines software and hardware to provide a heat flux test temperature control method for dynamically tuning PID parameters based on machine learning. This method can be implemented by an electronic device, and its execution process is as Figure 1 shown, including the following steps:

[0049] S1. Collect multiple state parameters during the heat flux test;

[0050] Among them, it should be noted that this embodiment uses a temperature controller device and a thermocouple to collect multiple state parameters during the heat flux test, including: the current temperature, target temperature, temperature change gradient, specific heat capacity of the material, thermal conductivity of the material, and the current test stage information during the heat flux test.

[0051] S2. Input the collected state parameters into a pre-trained neural network model, and use the pre-trained neural network model to output the dynamically tuned PID parameters according to the input state parameters;

[0052] Among them, the neural network adopted in this embodiment is a multi-branch dynamic neural network with a hybrid architecture. That is, this embodiment uses a multi-branch dynamic neural network with a hybrid architecture to construct a PID dynamic tuning model, and trains the neural network through historical test data. The trained neural network can perform online calculation during the test process to output PID parameters for PID parameter prediction, ensuring that the temperature control system has good adaptability to external disturbances. During the neural network training stage, multiple sets of heat flux test data are used. The input parameters include factors such as the current test temperature, the target test temperature, the temperature change gradient, the specific heat capacity of the test material, and the current test stage. The network weights are optimized through supervised learning to enable it to dynamically adjust PID parameters under different heat flux conditions. The dynamically tuned PID parameters are parameters that can meet the preset temperature control target to adapt to the current test conditions.

[0053] This multi-branch dynamic neural network with a hybrid architecture is used for the dynamic tuning of PID parameters in the heat flux test temperature control system. It can adaptively select different sub-network branches according to the temperature changes, material characteristics, and test stage information during the heat flux test process to achieve the purpose of fast response, high computational efficiency, and anti-interference control. It should be noted that the parameters and modules in this embodiment can be adjusted according to specific application requirements.

[0054] Specifically, in this embodiment, as Figure 2 shown, this multi-branch dynamic neural network with a hybrid architecture includes: an input layer, a material encoding module, a stage perception module, a heating branch, a steady-state branch, a cooling branch, a feature fusion module, and an output layer; the process of obtaining the dynamically tuned PID parameters using this network is as follows:

[0055] The system first collects temperature-related parameters including the current temperature, the target temperature, and the temperature change gradient from the heat flux test. At the same time, it collects the specific heat capacity and thermal conductivity that describe the material characteristics, and uses one-hot encoding representation according to the test stage (for example, [1, 0, 0] represents the heating stage, [0, 1, 0] represents the steady state, and [0, 0, 1] represents the cooling stage) to form a total of six-dimensional input parameters. These six-dimensional inputs serve as the initial inputs of the entire network and enter the subsequent processing modules through the input layer.

[0056] Specifically, in this embodiment, a material property encoding module is designed and adopted to map the original two-dimensional material parameters of specific heat capacity and thermal conductivity to a 32-dimensional high-dimensional feature space through a fully connected layer. The fully connected layer uses a ReLU activation function to enhance the sensitivity of the model to material properties, thereby generating a 32-dimensional material feature vector H_material. At the same time, the system is provided with a stage perception module, which concatenates the one-hot encoding of the test stage with the aforementioned 32-dimensional material feature to form a 35-dimensional input vector and sends it to a gating unit. The gating unit calculates a 3-dimensional probability distribution through a preset weight matrix and bias to represent the activation weights of the three sub-network branches of heating, steady state, and cooling. During this process, by comparing the magnitudes of the probabilities of each branch, the branch with the largest probability is taken as the main path, and the outputs of other branches are set to zero to reduce redundant calculations and ensure the on-demand allocation of computing resources.

[0057] In addition, this embodiment designs three sub-network branches for different test stages. For the heating stage, the present invention provides a high-speed response branch, whose main goal is to rapidly increase the temperature and preferentially optimize the proportional term (Kp) and the derivative term (Kd) to accelerate the response. This branch concatenates the temperature-related parameters (3-dimensional) with the 32-dimensional material feature to form a 35-dimensional vector. After passing through the first hidden layer using a fully connected layer (mapping from 35 dimensions to 64 dimensions) and activating with LeakyReLU, it further passes through the second hidden layer (mapping from 64 dimensions to 32 dimensions), also using the LeakyReLU activation function, and finally outputs a 32-dimensional feature vector H_rise as the output of this branch. For the steady state stage, the present invention sets up a sparse attention branch, whose purpose is to focus only on the key temperature node information while maintaining a low computational cost. This branch also concatenates the temperature parameters with the material feature to form a 35-dimensional vector, and then uses a sparse self-attention layer to calculate Query, Key, and Value (each dimension is set to 16). During the self-attention calculation process, only the top-3 attention weights are retained, and the remaining weights are set to zero, thereby significantly reducing the computational amount. Then, through a fully connected layer, the 16-dimensional attention feature is mapped to 32 dimensions, and the ReLU activation function is used to output a 32-dimensional attention feature H_steady. For the cooling stage, to solve the problems of temperature fluctuation and hysteresis effect, the present invention provides an anti-interference branch. This branch also concatenates the temperature parameters with the material feature to form a 35-dimensional vector, and then extracts the temporal features of temperature changes through a one-dimensional convolutional layer (convolution window size is 5, output channel number is set to 16), and further reduces the dimension through a max-pooling layer (pooling size is 2). Finally, through a fully connected layer (mapping 16 dimensions to 32 dimensions) using the Tanh activation function, a 32-dimensional feature vector H_cool is output as the result of this branch.

[0058] Further, after each branch is independently processed, the system fuses the output of the activated branch with the original 32-dimensional material feature vector and concatenates them into a 64-dimensional fused feature vector H_fused. Finally, the fused feature is mapped to a 3-dimensional output through a fully connected layer, and this 3-dimensional output directly corresponds to the parameters [Kp, Ki, Kd] of the PID controller in the temperature control system.

[0059] Further, this embodiment provides a loss function Loss applicable to this neural network to optimize the adaptive adjustment of the PID parameters during the temperature control process, improve the temperature tracking accuracy, and ensure the stability of the adjustment process. The loss function consists of a temperature error loss term L temperature , a material adaptability loss term L material , and a PID parameter stability loss term L stab and a trial phase factor and is specifically defined as follows:

[0060]

[0061] Among them, the temperature error loss term L temperature uses the mean squared error (MSE) to measure the deviation between the current temperature Tcurrent and the target temperature Ttarget:

[0062]

[0063] where N represents the total number of time steps; T target (i) represents the target temperature at the i-th time step; T current (i) represents the current actual temperature at the i-th time step.

[0064] This loss term ensures that the neural network can accurately track the set temperature, reduce the steady-state error, and improve the temperature control accuracy.

[0065] The material adaptability loss term L material sets an adaptive adjustment mechanism based on the threshold of the material thermal conductivity λ to ensure that the temperature change rate adapts to different material characteristics to optimize the heating and cooling rates. It is defined as follows:

[0066]

[0067] where α and β are material characteristic adjustment coefficients (to be calibrated through experiments). λ threshold represents the preset thermal conductivity threshold used to distinguish low-thermal-conductivity and high-thermal-conductivity materials; T represents the actual temperature at the current time step; t represents the current time step; dT / dt represents the actual temperature change rate at the current time step t.

[0068] The introduction of this loss term can effectively adapt to different material characteristics.

[0069] PID parameter stability loss term L stab By constraining the time variation of PID parameters, the influence of drastic parameter fluctuations on temperature control stability is avoided:

[0070] L stab =(K P (t)-K P (t - 1)) 2 +(K I (t)-K I (t - 1)) 2 +(K D (t)-K D (t - 1)) 2

[0071] Among them, K P (t) represents the proportional term of the PID control parameter at the current time step t; K P (t - 1) represents the proportional term of the PID control parameter at the previous time step t - 1; K I (t) represents the integral term of the PID control parameter at the current time step t; K I (t - 1) represents the integral term of the PID control parameter at the previous time step t - 1; K D (t) represents the derivative term of the PID control parameter at the current time step t; K D (t - 1) represents the derivative term of the PID control parameter at the previous time step t - 1.

[0072] Meanwhile, in order to adapt to different control stages, the loss function introduces a stage factor to dynamically adjust the stability weight:

[0073]

[0074] This experimental stage factor can be adaptively adjusted according to preset rules or historical data to ensure the optimal temperature control effect in different experimental stages.

[0075] The above loss function enables the neural network to converge accurately through a multi - objective optimization strategy, facilitating the optimization of PID parameters under different material conditions and control stages, improving temperature control accuracy, reducing overshoot, and enhancing system stability, and is applicable to intelligent regulation tasks in complex heat flow environments.

[0076] S3. According to the dynamically tuned PID parameters, the output power of the heating device is adjusted in real time to achieve temperature control;

[0077] S4. Through the temperature feedback signal, the neural network model is continuously corrected to achieve the adaptive optimization of PID parameters.

[0078] It should be noted that after obtaining the dynamically tuned PID parameters, in this embodiment, the PID controller is set according to the dynamically tuned PID parameters to control the output percentage of the temperature controller device for heating, and at the same time, the temperature error is fed back to correct the neural network to complete the closed-loop control of the temperature.

[0079] Specifically, as Figure 3 shown, in this embodiment, a temperature controller device and a thermocouple are used to obtain the real-time temperature data of the controlled object during the heat flux test, and a temperature control target is established; the data preprocessing module is used to filter out outliers and smooth the data to obtain an effective temperature data set; the designed neural network-based PID parameter tuning model is used to input the effective temperature data and the system historical response data, where the system historical response data mainly refers to various data reflecting the dynamic characteristics of the heat flux test temperature control process, such as temperature error, error change rate, error integral, overshoot, and test stage switching timestamp (such as the critical point of heating up to the steady state). The PID control parameters are adjusted in real time to obtain the dynamically optimal PID parameters; the multi-stage control strategy module is used to adjust the parameters according to different stages of the heat flux test (such as heating up, steady state, cooling down, etc.), and combined with the material characteristics, a targeted PID parameter optimization rule is set; a system architecture combining software and hardware is adopted, and the temperature controller device is used to execute real-time PID temperature control, and combined with the host computer software for data monitoring and remote adjustment; according to the simulation training, the strategy is adjusted through the learning algorithm to optimize the weights of the neural network model, so that the system can maintain a high-precision temperature control effect under different test conditions. And further through the test feedback correction mechanism, the weights of the neural network model are adaptively adjusted, so that the temperature control effect is continuously optimized with the progress of the test, and the temperature control accuracy and system stability are improved.

[0080] Furthermore, for the heat flux test temperature control method of this embodiment, combined with the characteristics of the test material, the neural network training data set is optimized, a unique identifier (such as a hash code) is generated for each test material, and it is stored together with the temperature data and PID parameters to form a structured data set, enabling the PID parameters to be adjusted adaptively, thereby improving the temperature control stability and accuracy.

[0081] Furthermore, in this embodiment, the temperature control adopts a closed-loop control mode, and different PID parameter adjustment strategies are set according to different test stages (heating up, steady state, cooling down). In addition, it should be noted that in this embodiment, the PID parameter dynamic adjustment strategy is optimized according to the characteristics of the test material. For example, for materials with a low thermal conductivity, the integral parameter is appropriately increased to enhance the smoothness of temperature adjustment; for materials with a large thermal inertia, the differential parameter is appropriately adjusted to reduce the overshoot phenomenon.

[0082] In summary, this embodiment provides a temperature control method for a heat flux test that dynamically tunes PID parameters based on machine learning. By using machine learning methods, the dynamic adjustment of PID parameters is achieved through a neural network, avoiding manual intervention and improving the temperature control accuracy and stability. Through differential processing in the heat flux test stage, the temperature control strategy can be automatically adjusted according to different stages during the test, and the thermal properties of materials (such as specific heat capacity) are also used as one of the parameters to make the temperature control more targeted and reduce overshoot and steady-state error. Temperature error feedback closed-loop optimization is adopted to achieve the adaptive learning of the temperature control system and enhance the ability to adapt to different test conditions. The hardware-software combined architecture and modular design are used, which have good scalability and integration and can adapt to different application requirements.

[0083] Second Embodiment

[0084] This embodiment provides a temperature control device for a heat flux test that dynamically tunes PID parameters based on machine learning. The system structure of the temperature control device for the heat flux test is as Figure 4 shown, and it consists of a data acquisition module, a neural network calculation module, a temperature control execution module, and a feedback correction module. Through the information processing capabilities of different modules, high-precision control of the temperature in the heat flux test is achieved. Among them, the functions of each module are as follows:

[0085] The data acquisition module is used to collect multiple state parameters during the heat flux test;

[0086] Specifically, in this embodiment, the data acquisition module includes a temperature controller device, a thermocouple, and a communication interface, which can realize real-time monitoring and recording of temperature and heat flux test conditions.

[0087] The neural network calculation module is used to input the collected state parameters into a pre-trained neural network model, and use the pre-trained neural network model to output the dynamically tuned PID parameters according to the input state parameters;

[0088] The temperature control execution module is used to adjust the temperature controller device in real time according to the dynamically tuned PID parameters to drive the temperature controller device to control the heating device and achieve precise temperature control;

[0089] Specifically, in this embodiment, the temperature control execution module includes a PID controller, a temperature controller device, and a power adjustment unit, which are used to adjust the power output of the heating device to achieve temperature adjustment.

[0090] The feedback correction module is used to receive the real-time temperature feedback signal, and based on the temperature feedback signal, continuously correct the neural network model based on parameters such as temperature error, adjustment time, and overshoot amount to achieve the adaptive optimization of PID parameters, make the test environment reach the expected temperature, and improve the temperature control accuracy and system stability.

[0091] In addition, it should be noted that for the sake of convenience of description, Figure 4 only the main components of the device are shown. Moreover, the temperature control device for the heat flux experiment based on machine learning for dynamically tuning PID parameters in this embodiment corresponds to the method for the heat flux experiment based on machine learning for dynamically tuning PID parameters in the above-mentioned first embodiment; among them, the functions realized by each functional module in the temperature control device for the heat flux experiment based on machine learning for dynamically tuning PID parameters in this embodiment correspond one by one to each process step in the method for the heat flux experiment based on machine learning for dynamically tuning PID parameters in the above-mentioned first embodiment; therefore, it will not be elaborated here.

[0092] In summary, this embodiment provides a temperature control device for a heat flux experiment based on machine learning for dynamically tuning PID parameters. The temperature control device for the heat flux experiment based on machine learning for dynamically tuning PID parameters adopts a neural network model, and adjusts the PID parameters in real time according to the current temperature, target temperature, PID parameters, temperature change gradient, specific heat capacity of the heat-conducting material, and current test stage information to achieve dynamic optimal control; combined with a multi-stage control strategy, differential PID parameter adjustment is carried out in different test stages such as heating, steady state, and cooling to improve the temperature control accuracy; through a system architecture combining software and hardware, a temperature control device composed of an algorithm, software, and temperature controller hardware uses the temperature controller to perform PID temperature control, and combines the upper computer software for data monitoring and remote control; the weight of the neural network model is optimized by using a feedback correction mechanism, effectively improving the temperature control stability and adaptive ability.

[0093] Third Embodiment

[0094] This embodiment provides an electronic device, as Figure 5 shown, the electronic device includes: a processor and a memory; wherein, the processor and the memory can be connected through a communication bus; at least one instruction is stored in the memory, and the instruction is loaded and executed by the processor to implement the method in the above-mentioned first embodiment. In addition, the electronic device may further include a transceiver, and the processor and the transceiver can be connected through a communication bus, and the transceiver is used for communicating with other devices.

[0095] Next, in combination with Figure 5 specific introductions will be made to each component of the electronic device:

[0096] Among them, the processor is the control center of the electronic device. The electronic device may include multiple processors, and each of these processors may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The processor here may be a single processor or a collective term for multiple processing elements. For example, the processor may be one or more central processing units (CPUs), or other general-purpose processors, application specific integrated circuits (ASICs), or one or more integrated circuits configured to implement the embodiments of the present invention. For example: one or more digital signal processors (DSPs), or one or more field programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor can execute various functions of the electronic device by running or executing software programs stored in the memory and calling data stored in the memory.

[0097] In a specific implementation, as an embodiment, the processor may include one or more CPUs. For example Figure 5 CPU0 and CPU1 shown in [figure], of course, this is only an exemplary illustration.

[0098] The memory is used to store the software program for implementing the solution of the present invention and is controlled by the processor for execution. The specific implementation method can refer to the above method embodiments and will not be elaborated here.

[0099] Optionally, the memory may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, or may be an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory may be integrated with the processor or may exist independently and be coupled to the processor through the interface circuit of the electronic device ( Figure 5 not shown in the figure), and the embodiments of the present invention do not make specific limitations in this regard.

[0100] The transceiver may include a receiver and a transmitter ( Figure 5 not separately shown in the figure). Among them, the receiver is used to implement the receiving function, and the transmitter is used to implement the sending function. The transceiver may be integrated with the processor or may exist independently and be coupled to the processor through the interface circuit of the electronic device ( Figure 5 not shown in the figure), and the embodiments of the present invention do not make specific limitations in this regard.

[0101] In addition, it should be noted that Figure 5 the structure of the electronic device shown in the figure does not constitute a limitation on the device. The actual device may include more or fewer components than shown in the figure, or combine some components, or have a different component arrangement. In addition, the technical effects achieved by the electronic device when executing the method of the first embodiment above may refer to the technical effects described in the first embodiment above, so they will not be elaborated here.

[0102] Fourth Embodiment

[0103] This embodiment provides a computer-readable storage medium, in which at least one instruction is stored, and the instruction is loaded and executed by the processor to implement the method of the first embodiment above. Among them, the computer-readable storage medium may be ROM, random access memory, CD-ROM, magnetic tape, floppy disk, and optical data storage devices, etc. The instructions stored therein can be loaded and executed by the processor in the terminal to execute the above method.

[0104] In addition, it should be noted that the present invention can be provided as a method, an apparatus, or a computer program product. Therefore, the embodiments of the present invention can take the form of all or part of a hardware embodiment, all or part of a software embodiment, or an embodiment combining software and hardware aspects. Moreover, when implemented using software, the embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or a data center containing one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0105] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of a general-purpose computer, an embedded processor, or other programmable data processing terminal devices to generate a machine, such that the instructions executed by the processors of the computer or other programmable data processing terminal devices generate means for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0106] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1The functions specified in one or more boxes. These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, so that a series of operation steps are executed on the computer or other programmable terminal device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable terminal device provide for implementing in the process Figure 1 One process or multiple processes and / or boxes Figure 1 Steps for the functions specified in one box or multiple boxes.

[0107] It should also be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or terminal device comprising a series of elements includes not only those elements but also other elements not expressly listed, or also elements inherent to such process, method, article or terminal device. Without further limitation, an element defined by the statement "comprising one..." does not exclude the existence of additional identical elements in the process, method, article or terminal device comprising the said element. In addition, the term "and / or" is merely a description of the associated relationship of the associated objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. Here, A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship, which can be understood specifically with reference to the context. "At least one" means one or more, and "multiple" means two or more. "At least one of the following (items)" or similar expressions refer to any combination of these items, including any combination of single (item) or plural (items). For example, at least one of a, b or c can mean: a, b, c, a - b, a - c, b - c, or a - b - c, where a, b, c can be single or multiple.

[0108] In addition, it can be understood that in various embodiments of the present invention, the magnitudes of the sequence numbers of the above processes do not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0109] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0110] In several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of functional modules / units is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of devices or units can be in electrical, mechanical, or other forms. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in each embodiment of the present invention, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0111] If the method is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present invention. The foregoing storage medium includes various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0112] Finally, it should be noted that the above description is only the preferred embodiment of the present invention. It should be pointed out that although the preferred embodiments of the present invention have been described, for those of ordinary skill in the art, once they know the basic creative concept of the present invention, several improvements and refinements can still be made without departing from the principle described in the present invention. These improvements and refinements should also be regarded as the protection scope of the present invention. Therefore, the appended claims are intended to be construed as including the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A temperature control method for a heat flux experiment based on dynamically tuning PID parameters by machine learning, characterized in that The temperature control method for heat flux experiments based on machine learning dynamic tuning of PID parameters includes: Collecting multiple state parameters during the heat flux experiment process; Inputting the collected state parameters into a pre-trained neural network model, and using the pre-trained neural network model to output the dynamically tuned PID parameters according to the input state parameters; According to the dynamically tuned PID parameters, the output power of the heating device is adjusted in real time to achieve temperature control; Through the temperature feedback signal, the neural network model is continuously corrected to achieve the adaptive optimization of the PID parameters.

2. The temperature control method for the heat flux experiment based on dynamically tuning PID parameters by machine learning according to claim 1, characterized in that, The state parameters include: the current temperature, target temperature, temperature change gradient, specific heat capacity of the material, thermal conductivity of the material, and the current test stage during the heat flux experiment process; among them, the test stage is divided into three different stages: the heating stage, the steady state stage, and the cooling stage, and is represented by one-hot encoding.

3. The temperature control method for the heat flux test with PID parameters dynamically tuned based on machine learning according to claim 2, characterized in that The neural network model is a multi-branch dynamic neural network with a hybrid architecture, which includes: an input layer, a material encoding module, a stage perception module, a heating branch, a steady state branch, a cooling branch, a feature fusion module, and an output layer; The process of obtaining the dynamically tuned PID parameters using the multi-branch dynamic neural network includes: The collected state parameters enter the multi-branch dynamic neural network through the input layer; The material encoding module processes the specific heat capacity of the material and the thermal conductivity of the material through a fully connected layer to generate a material feature vector; the stage perception module splices the one-hot encoding of the test stage with the material feature vector and sends it to a gated unit; the gated unit calculates a three-dimensional probability distribution through a preset weight matrix and bias to represent the activation weights of the heating branch, the steady state branch, and the cooling branch; the branch with the largest probability is taken as the main path, and the outputs of other branches are set to zero; among them, the heating branch is used to splice the current temperature, target temperature, and temperature change gradient with the material feature vector, send it to the first hidden layer, be processed by a fully connected layer in the first hidden layer, and after being activated by LeakyReLU, send it to the second hidden layer, and after passing through a mapping process in the second hidden layer and being activated by LeakyReLU, the output of the heating branch is obtained; the steady state branch is used to splice the current temperature, target temperature, and temperature change gradient with the material feature vector and send it to a sparse self-attention layer; use the sparse self-attention layer to calculate Query, Key, and Value, and only retain the top three attention weights during the self-attention calculation process, and set the remaining weights to zero to obtain the attention feature; then, the attention feature is mapped through a fully connected layer and activated by ReLU to obtain the output of the steady state branch; the cooling branch is used to splice the current temperature, target temperature, and temperature change gradient with the material feature vector, send it to a one-dimensional convolutional layer to extract the temporal feature of the temperature change, and further reduce the dimension through a max pooling layer, and finally be processed by a fully connected layer and activated by Tanh to obtain the output of the cooling branch. The feature fusion module fuses the output of the main path with the material feature vector to obtain a fused feature vector; at the output layer, the fused feature vector is mapped to a three-dimensional output through a fully connected layer, and the three-dimensional output directly corresponds to the parameters of the PID controller in the temperature control system, including Kp, Ki, and Kd.

4. The temperature control method for the heat flux test with PID parameters dynamically tuned based on machine learning according to claim 1, wherein, The training process of the neural network model includes: Collecting the state parameters during the historical heat flux experiment, filtering out outliers and smoothing the collected state parameters to obtain effective state parameters, and using the effective state parameters to construct a historical experiment data set; Using the historical experiment data set, optimizing the network weights of the neural network model through supervised learning, so that the neural network model can dynamically adjust the PID parameters under different heat flux conditions.

5. The temperature control method for the heat flow test with PID parameters dynamically tuned based on machine learning according to claim 4, characterized in that, The PID parameter dynamic adjustment strategy is optimized according to the characteristics of the test material. For materials with a thermal conductivity lower than the preset thermal conductivity threshold, the integral parameter is increased to enhance the stability of temperature adjustment; for materials with a thermal inertia greater than the preset thermal inertia threshold, the differential parameter is adjusted to reduce overshoot.

6. A temperature control device for a heat flux test that dynamically tunes PID parameters based on machine learning, characterized in that, The heat flux experiment temperature control device for dynamically tuning PID parameters based on machine learning includes: A data acquisition module for collecting multiple state parameters during the heat flux experiment; A neural network calculation module for inputting the collected state parameters into a pre-trained neural network model, and using the pre-trained neural network model to output the dynamically tuned PID parameters according to the input state parameters; A temperature control execution module for adjusting the output power of the heating device in real time according to the dynamically tuned PID parameters to achieve temperature control; A feedback correction module for continuously correcting the neural network model through a temperature feedback signal to achieve adaptive optimization of the PID parameters.

7. The temperature control device for heat flow test that dynamically tunes PID parameters based on machine learning according to claim 6, characterized in that, The state parameters include: the current temperature, target temperature, temperature change gradient, specific heat capacity of the material, thermal conductivity of the material, and the current test stage during the heat flux experiment; among them, the test stage is divided into three different stages: the heating stage, the steady state stage, and the cooling stage, and one-hot encoding is used to represent them.

8. The temperature control device for heat flow test that dynamically tunes PID parameters based on machine learning as claimed in claim 6, wherein The neural network model is a multi-branch dynamic neural network with a hybrid architecture, which includes: an input layer, a material encoding module, a stage perception module, a heating branch, a steady state branch, a cooling branch, a feature fusion module, and an output layer; The process of obtaining the dynamically tuned PID parameters using the multi-branch dynamic neural network includes: The collected state parameters enter the multi-branch dynamic neural network through the input layer; The material coding module processes the specific heat capacity and thermal conductivity of materials through a fully connected layer to generate a material feature vector; the stage perception module concatenates the one-hot encoding of the test stage with the material feature vector and then sends it to a gating unit; the gating unit calculates a three-dimensional probability distribution through a preset weight matrix and bias to represent the activation weights of the heating branch, steady-state branch, and cooling branch; the branch with the maximum probability is taken as the main path, and the outputs of other branches are set to zero; among them, the heating branch is used to concatenate the current temperature, target temperature, and temperature change gradient with the material feature vector, send it to the first hidden layer, be processed by a fully connected layer in the first hidden layer, and after being activated by LeakyReLU, send it to the second hidden layer, and after being mapped and activated by LeakyReLU in the second hidden layer, the output of the heating branch is obtained; the steady-state branch is used to concatenate the current temperature, target temperature, and temperature change gradient with the material feature vector and send it to a sparse self-attention layer; use the sparse self-attention layer to calculate Query, Key, and Value, and only retain the top three attention weights during the self-attention calculation process, and set the remaining weights to zero to obtain attention features; then, map the attention features through a fully connected layer and activate them with ReLU to obtain the output of the steady-state branch; the cooling branch is used to concatenate the current temperature, target temperature, and temperature change gradient with the material feature vector, send it to a one-dimensional convolutional layer to extract the temporal features of temperature changes, and further reduce the dimension through a max pooling layer, and finally be processed by a fully connected layer and activated by Tanh to obtain the output of the cooling branch. The feature fusion module fuses the output of the main path with the material feature vector to obtain a fused feature vector; at the output layer, the fused feature vector is mapped to a three-dimensional output through a fully connected layer, and the three-dimensional output directly corresponds to the parameters of the PID controller in the temperature control system, including Kp, Ki, and Kd.

9. The temperature control device for the heat flux experiment that dynamically tunes PID parameters based on machine learning as claimed in claim 6, wherein, The training process of the neural network model includes: Collect the state parameters during the historical heat flux test process, filter out outliers and smooth the collected state parameters to obtain effective state parameters, and use the effective state parameters to construct a historical test data set. Use the historical test data set to optimize the network weights of the neural network model through supervised learning, so that the neural network model can dynamically adjust the PID parameters under different heat flux conditions.

10. The temperature control device for heat flow test that dynamically tunes PID parameters based on machine learning as claimed in claim 9, wherein The PID parameter dynamic adjustment strategy is optimized according to the characteristics of the test materials. For materials with a thermal conductivity lower than the preset thermal conductivity threshold, the integral parameter is increased to enhance the stability of temperature adjustment; for materials with a thermal inertia greater than the preset thermal inertia threshold, the differential parameter is adjusted to reduce overshoot.

Images