Microwave component heat source distribution optimization method based on improved DNN algorithm

Abstract

According to the microwave assembly heat source distribution optimization method based on the improved DNN algorithm, the improved DNN algorithm is applied to microwave assembly heat source distribution optimization, and a new optimization method is provided for microwave assembly heat source distribution optimization. According to the method, an improved DNN prediction model framework is constructed, residual connection, an attention mechanism and parallel CNN branches are fused in the framework, multiple groups of data are substituted into the framework, a hierarchical learning rate strategy is adopted for training, and model parameters are optimized by using a temperature constraint weighted loss function; according to the method, the structure of a model output layer is optimized, a dual-output form of'temperature mean value + prediction variance 'is adopted, and the uncertainty of prediction is quantified while the highest temperature prediction value is output; meanwhile, a composite loss function fusing a mean square error (MSE) and a temperature threshold penalty term is constructed, and for a prediction error exceeding a set safe temperature threshold, 2-3 times of weight is given to the loss function, so that the model preferentially guarantees the prediction precision of a high-temperature scene.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing technology, and specifically to a method for optimizing the heat source distribution of microwave components based on an improved DNN algorithm. Background Technology

[0002] As microwave components continue to evolve towards miniaturization and high integration, they exhibit characteristics such as high integration and high power density, with their power density approaching 1000 W / cm². 3 Currently, heat dissipation technology has become one of the key bottlenecks restricting its application. Traditional simulation iteration methods mainly rely on the designer's experience for iterative optimization, which has disadvantages such as dependence on personal experience, low computational efficiency, and difficulty in finding the optimal solution, and cannot quickly obtain the best heat source layout scheme.

[0003] Therefore, a method is needed to quickly and accurately obtain the heat source layout scheme of microwave components. Summary of the Invention

[0004] This invention aims to solve the problem of heat source layout in microwave components by providing a method for optimizing heat source distribution in microwave components based on an improved DNN algorithm. This method can quickly and efficiently find the optimal heat source distribution in microwave components using the DNN algorithm, thereby significantly improving the thermal design capability of microwave components.

[0005] This invention provides a method for optimizing the heat source distribution of microwave components based on an improved DNN algorithm, comprising the following steps: S1. Establish the center coordinates of the heat source chips and form a variable matrix based on the number and type of heat source chips in the microwave component to be optimized; S2. Based on the size envelope of the microwave component to be optimized and the size of the heat source chip, set the range of values ​​for the center coordinates of the heat source chip. Then, based on the positional relationship between each heat source chip, obtain the heat source chip center coordinate relationship formula that prevents interference between each heat source chip. Then, based on the heat source chip center coordinate relationship formula, use Matlab software to randomly generate input data containing a variable matrix. S3. Establish a parametric simulation model in thermal simulation software; S4. Complete the setting of simulation boundary conditions; S5. Import some of the input data from step S2 into the parametric simulation model and simulate it to obtain the simulation results. The simulation results are the highest temperature values ​​of the microwave components to be optimized. S6. Select multiple sets of heat source chip center coordinates as input data for the improved DNN prediction model framework, and use the simulation results corresponding to the input data as output data to obtain data sets. S7. Establish an improved DNN prediction model; The framework of the improved DNN prediction model includes an input layer, a deep neural network containing 3 to 5 hidden layers, and an output layer connected in sequence. Residual connections are used between hidden layers to alleviate gradient vanishing in deep networks through shortcuts; the number of neurons in the hidden layers is dynamically adjusted, with the number of neurons in the first hidden layer set to 3 to 4 times the dimension of the input data, and the number of neurons in subsequent layers decreasing by 20%; and batch normalization layers are inserted between each layer to accelerate convergence. A channel attention mechanism is embedded after the hidden layer to automatically identify and enhance the feature weights of key heat source locations. At the same time, a lightweight CNN branch with a 3×1 convolutional kernel is added in parallel to extract local correlation features of the coordinate sequence. The output of the CNN branch is concatenated with the output of the DNN fully connected layer as the final feature. The improved DNN prediction model was trained based on at least 500 sets of data. S8. Analyze the prediction results of the improved DNN prediction model, and obtain the trained improved DNN prediction model after achieving the goal.

[0006] The present invention discloses a microwave component heat source distribution optimization method based on an improved DNN algorithm. In a preferred embodiment, in step S2, the input data is subjected to data augmentation processing, and random perturbations within a specified range are added to the center coordinates of the heat source chip to simulate the installation deviation in the actual process. At the same time, the thermal conductivity parameter of the heat source chip is randomly fluctuated within a specified range to generate an expanded sample. In step S5, the expanded sample is imported into the parameterized simulation model as input data and simulated.

[0007] The microwave component heat source distribution optimization method based on the improved DNN algorithm described in this invention, as a preferred embodiment, includes the following steps in step S3: S3a. Simplify the initial design of the microwave component to be optimized by removing features that have little impact on heat source distribution optimization, including: threaded holes, chamfers and lettering; S3b. In the DesignModeler software that comes with the simulation software ANSYS Workbench, a parametric simulation model is established based on the center coordinates of the heat source chip and the size of the microwave component to be optimized. S3c: Parameterize the center coordinates of the heat source chip.

[0008] The microwave component heat source distribution optimization method based on the improved DNN algorithm described in this invention, in a preferred embodiment, includes the following steps in step S4: S4a. Assign corresponding material properties to each part, including printed circuit board material, chip material, interface connection material and solder; S4b, Define the contact relationship between each part, including contact gap tolerance, contact pair and interface thermal resistance; S4c: Perform mesh generation and refine the mesh of the interface material of the heat source chip; S4d: Set the thermal boundary, and set the heat dissipation, convection coefficient, radiation coefficient and working time of each heat source chip; S4e, The overall temperature field distribution of the microwave component to be optimized is taken as the solution objective.

[0009] The microwave component heat source distribution optimization method based on the improved DNN algorithm described in this invention, as a preferred embodiment, includes the following steps in step S5: S5a. After setting the boundary conditions, the highest temperature value is used as the simulation result and set as a parameter. S5b: Open the parametric interface and import some of the input data obtained in step 2 into the parametric simulation model. S5c, Click the Update All Design Points button to start parametric calculations; After S5d and parametric calculations are completed, export the simulation results.

[0010] The present invention discloses a microwave component heat source distribution optimization method based on an improved DNN algorithm. In a preferred embodiment, in step S7, the output layer of the improved DNN prediction model uses a dual output form of temperature mean + prediction variance, which quantifies the uncertainty of prediction while outputting the highest temperature prediction value. At the same time, a composite loss function that integrates mean square error and temperature threshold penalty term is constructed. For prediction errors exceeding the set safe temperature threshold, a weight of 2 to 3 times is assigned to the loss function, so that the improved DNN prediction model prioritizes the prediction accuracy of high temperature scenarios. The composite loss function is: ; in: For the sample size, To simulate the true value, The predicted values ​​of the improved DNN prediction model. These are the weighting coefficients; when hour, The value should be between 2 and 3, otherwise The value is 1. This is the safe temperature threshold.

[0011] The microwave component heat source distribution optimization method based on an improved DNN algorithm described in this invention, as a preferred embodiment, involves model training in step S7 based on a hierarchical learning rate strategy: the improved DNN prediction model is trained based on 800 sets of data; during training, a hierarchical learning rate setting is used: a smaller learning rate is used for the feature extraction layer to retain the original features, and a larger learning rate is used for the temperature prediction layer for fast fitting; the smaller learning rate is 1e. -4 The maximum learning rate is 1e -3 .

[0012] Error information is extracted during the training process of the improved DNN prediction model, including MAE, MSE and RMSE, and error information statistics are performed.

[0013] The microwave component heat source distribution optimization method based on an improved DNN algorithm described in this invention, as a preferred embodiment, uses the following method to calculate the mean absolute error in step S8: ; in: It is 100. To simulate the true value, These are the predicted values ​​from the improved DNN prediction model.

[0014] The microwave component heat source distribution optimization method based on the improved DNN algorithm described in this invention, as a preferred embodiment, further includes step S9: analyzing and verifying the optimal layout results obtained by the improved DNN prediction model, thus completing the optimization of the microwave component heat source distribution.

[0015] Step S9 includes the following steps: S9a. Input the optimal layout result obtained by training the improved DNN prediction model into the parameterized simulation model for simulation. Compare the highest temperature value obtained by simulation with the average temperature and confidence interval predicted by the trained improved DNN prediction model to verify whether the simulation result falls within the reasonable fluctuation range predicted by the model. S9b. When the simulation results fall within a reasonable fluctuation range, the optimized results of the microwave component heat source distribution are obtained. Based on the optimized results of the microwave component heat source distribution, the design is carried out, and the microwave component prototype is manufactured and then tested in practice.

[0016] The present invention discloses a microwave component heat source distribution optimization method based on an improved DNN algorithm. In a preferred embodiment, in step S1, there are two types of heat source chips, with four of each type, and the center coordinates of the heat source chips are (…). x 1 ，y 1 )~( x 4 ，y 4 The variable matrix is ​​() x 1 ，y 1 ，x 2 ，y 2 ，x 3 ，y 3 ，x 4 ，y 4 ); In step S2, 1000 sets of input data are randomly generated, with a specified range of ±5%; In step S5, the first 800 sets of the center coordinates of the heat source chip are imported into the parametric simulation model; In step S7, the improved DNN prediction model is trained based on 800 sets of data. In step S8, the prediction results of 100 improved DNN prediction models are analyzed.

[0017] This invention proposes a method for optimizing the heat source distribution of microwave components based on an improved DNN algorithm. This method can quickly complete the optimization design of the heat source distribution of microwave components and has the advantages of high accuracy, low computational load, and fast calculation speed, which can significantly improve the efficiency and reliability of product thermal design.

[0018] The present invention has the following advantages: (1) This invention applies an improved DNN algorithm to the optimization of heat source distribution in microwave components, proposing a new optimization method for heat source distribution optimization in microwave components, and after verification, it has been found to have good feasibility. Compared with the simulation iteration method, which heavily relies on the designer's experience, the optimization result using the DNN algorithm is closer to the optimal solution, has higher computational accuracy and efficiency, and can effectively shorten the design optimization time by more than 50%, thereby significantly improving design efficiency.

[0019] (2) The present invention constructs an improved DNN prediction model framework, which integrates residual connection, attention mechanism and parallel CNN branch. Multiple sets of data are substituted into the framework, and a hierarchical learning rate strategy is used for training. The model parameters are optimized by using temperature-constrained weighted loss function. (3) The present invention optimizes the output layer structure of the model and adopts a dual output form of "temperature mean + prediction variance" to quantify the uncertainty of prediction while outputting the highest temperature prediction value. At the same time, a composite loss function is constructed that integrates mean square error (MSE) and temperature threshold penalty term. For prediction errors that exceed the set safe temperature threshold, a weight of 2 to 3 times is given in the loss function so that the model prioritizes the prediction accuracy of high temperature scenarios.

[0020] (4) The present invention is universal and can be applied to the optimization of heat source distribution in various microwave components and other similar products by making corresponding simple modifications. Attached Figure Description

[0021] Figure 1 This is a flowchart of a microwave component heat source distribution optimization method based on an improved DNN algorithm; Figure 2 This is a schematic diagram of the initial heat source distribution of a microwave component, representing an embodiment of a microwave component heat source distribution optimization method based on an improved DNN algorithm. Figure 3 An embodiment of a microwave component heat source distribution optimization method based on an improved DNN algorithm is presented, along with a prediction model framework based on the DNN algorithm. Figure 4 This document describes the tolerance variation during the training process of a DNN algorithm in an embodiment of a microwave component heat source distribution optimization method based on an improved DNN algorithm. Figure 5 An embodiment of a microwave component heat source distribution optimization method based on an improved DNN algorithm uses the optimal heat source distribution optimized by the DNN algorithm. Figure 6 Simulation results for an embodiment of a microwave component heat source distribution optimization method based on an improved DNN algorithm; Figure 7 This is a thermal imager measurement result of a product prototype of an embodiment of a microwave component heat source distribution optimization method based on an improved DNN algorithm. Detailed Implementation

[0022] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Example 1

[0023] like Figure 1 As shown, a method for optimizing the heat source distribution of microwave components based on an improved DNN algorithm includes the following steps: S1. Based on the initial product design, establish a variable matrix for the center coordinates of the heat source chip; In step S1, the initial product design scheme is as follows: Figure 2 As shown, the variable matrix for establishing the center coordinates of the heat source chip is established, including the following steps. Step S1a: Based on the initial design scheme of the product, establish... 8 The center coordinates of the heat source chip (x) 1 ,y 1 )~(x 4 y 4 ); Step S1b: Based on the center coordinates of the heat source chip, form a variable matrix (x1, y1, x...). 2 ,y 2 ,x3,y3,x 4 ,y 4 ).

[0024] S2. Determine the coordinate boundaries of the heat source chip and randomly generate 1000 sets of input data containing the center coordinates of the heat source chip; Step S2 includes two types of heat source chips. The heat dissipation and size information of the heat source chips are shown in Table 1. Table 1. Parameter Table of Heat Source Chip

[0025] S2a. Determine the range of values ​​for the center coordinates of the heat source chip based on the product size envelope and chip size; S2b. Based on the positional relationship between each heat source chip, determine the relationship between the center coordinates of each heat source chip to ensure that the heat source chips do not interfere with each other. S2c: Based on the boundary conditions of the heat source chip, 1000 sets of input data containing the center coordinates of the heat source chip were randomly generated using Matlab software. S2d performs data augmentation on the generated input data: adds a random perturbation within ±5% of the center coordinates of the heat source chip to simulate the installation deviation in the actual process; at the same time, randomly fluctuates the thermal conductivity parameter of the chip within a certain range to generate expanded samples and improve the robustness of the model to process fluctuations.

[0026] S3. Establish a parameterized simulation model in thermal simulation software; In step S3, a parameterized simulation model of the microwave component heat source distribution is established, including the following steps: S3a. Simplify the initial design of the product by removing features that have little impact on the optimization of heat source distribution, such as threaded holes, chamfers, and lettering, thereby simplifying the model. S3b. Create a simulation model in the DesignModeler software that comes with the simulation software ANSYS Workbench. When creating the simulation model, model it based on the center point of the heat source chip and the chip size. S3c: Parameterize the center point of the heat source chip.

[0027] S4. Complete the setting of simulation boundary conditions; Further, in step S4, the simulation boundary conditions are set, including the following steps: S4a. Assign corresponding material properties to each part, including printed circuit board material (Si), chip material (GsAs), interface bonding material (conductive adhesive), and solder (Sn63Pb37), etc. S4b: Define the contact relationships between various parts, including contact gap tolerance, contact pairs, and interface thermal resistance, etc. S4c performs mesh generation and refines the mesh of the interface material of the heat source chip.

[0028] S4d: Set the thermal boundary. The heat dissipation of each heat source chip is shown in Table 1. The convection coefficient is set to 10W / m. 2 ·℃, radiation coefficient set to 0.2, working time 1800s; S4e, take the overall temperature field distribution of the product as the solution objective.

[0029] Step S5: Perform parametric simulation and obtain the highest temperature value as the simulation result; Further, in step S5, parametric simulation is performed, including the following steps: S5a. After setting the boundary conditions, the highest temperature value is used as the simulation result and set as a parameter. S5b: Open the parameterization interface and import the first 800 sets of heat source chip center coordinates obtained in step 2c. S5c, Click the "Update All Design Points" button to start parametric calculations; After S5d and parametric calculations are completed, export the calculated data.

[0030] Step S6: Select multiple sets of heat source chip center coordinates as input data, and the corresponding simulation results as output data; Step S7: Train the DNN algorithm by substituting 800 sets of data into the DNN algorithm framework, which specifically includes the following steps: S7a. Establish an improved DNN prediction model framework, as shown in the appendix. Figure 3As shown: First, a deep neural network with 3-5 hidden layers is constructed, and a ResNet mechanism is introduced to alleviate the gradient vanishing problem in deep networks through shortcut connections. Second, the number of neurons is dynamically adjusted, with the number of neurons in the first hidden layer set to 3-4 times the input dimension, and the number of neurons in subsequent layers decreasing by 20%. Batch normalization (BN) layers are inserted between each layer to accelerate convergence. Finally, a channel attention mechanism (SE module) is embedded after the hidden layers to automatically identify and strengthen the feature weights of key heat source locations. At the same time, a lightweight CNN branch with a 3×1 convolutional kernel is added in parallel to extract local correlation features of the coordinate sequence. The output of the CNN branch is concatenated with the output of the fully connected layer of the DNN as the final feature.

[0031] S7b, Define Model Output and Loss Function: Optimize the model output layer structure by adopting a dual output form of "temperature mean + prediction variance" to quantify the uncertainty of prediction while outputting the highest temperature prediction value; simultaneously, construct a composite loss function that integrates mean squared error (MSE) and temperature threshold penalty term. For prediction errors exceeding the set safe temperature threshold, assign 2 to 3 times the weight in the loss function to ensure the model prioritizes the prediction accuracy of high-temperature scenarios. The mathematical expression of the composite loss function is formula (1): (1) In the formula: For the sample size, To simulate the true value, These are the model's predicted values. For weighting coefficients, when (When setting a safe temperature threshold) The value should be between 2 and 3, otherwise The value is 1.

[0032] S7c. Model training based on a hierarchical learning rate strategy: The prediction model framework is trained based on 800 sets of input and output data; a hierarchical learning rate setting is adopted during the training process: a smaller learning rate (set to 1e in this embodiment) is used for the feature extraction layer (i.e., the first two layers). -4 To preserve the original features, a larger learning rate (set to 1e in this embodiment) is used for the temperature prediction layer (i.e., the last 2-3 layers). -3 (to achieve rapid fitting)

[0033] S7d. Extract error information during the training process of the DNN algorithm framework, mainly including MAE (Mean Absolute Error), MSE (Mean Squared Error), and RMSE (Root Mean Squared Error). The error changes during the 250 training cycles in this embodiment are shown in the attached figure. Figure 4 As shown; Table 2 shows the error statistics during the training process of the S7e and DNN algorithms. As can be seen from the table, with the increase in the number of network training iterations, the mean absolute error decreased from 0.359 after 20 iterations to 0.092 after 200 iterations, a decrease of 74%; the mean squared error and root mean square error decreased by 66.8% and 71.2%, respectively. With the increase in the number of network training iterations, the error between the DNN model's predicted value and the module's actual temperature becomes smaller and smaller.

[0034] Table 2 Error Statistics during DNN Algorithm Training

[0035] S8. Select 100 sets of data for testing and analysis. If the error meets the requirements, the optimal layout result can be obtained. In step 8, the mean absolute error is calculated using formula (2), as shown in formula (2): (2) In the formula: It is 100. To simulate the true value, The predicted value is the value of the improved DNN prediction model.

[0036] The calculated mean absolute error is 0.12, which meets the requirements, thus yielding the result shown in the attached figure. Figure 5 The optimal heat source distribution shown has a heat source temperature of approximately 89.87℃.

[0037] S9. The optimal layout result is analyzed and verified, and the heat source distribution optimization of the microwave component is completed.

[0038] Furthermore, in step S9, the optimal layout result is analyzed and verified, including the following steps: S9a. Based on the optimal layout, perform simulations and compare the simulated maximum temperature value with the model-predicted average temperature and confidence interval to verify whether the simulation results fall within the reasonable fluctuation range of the model prediction. In this embodiment, the simulation results are attached. Figure 6 As shown, the highest temperature was 89.9℃, with an error of less than 0.1℃ compared to the optimized result.

[0039] S9b. Design was carried out based on the optimization results of the microwave component heat source distribution. After fabricating the product prototype, actual testing was conducted. The test results of the thermal imager are attached. Figure 7 As shown; S9c, Comparison Appendix Figure 6 and attached Figure 7 It can be seen that the prediction results of the DNN algorithm and the actual test results have an error of about 3°C, which is less than 5%. The error mainly comes from factors such as model simplification, simulation boundary conditions and the accuracy of thermal imager testing.

[0040] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for optimizing the heat source distribution of microwave components based on an improved DNN algorithm, characterized in that: Includes the following steps: S1. Establish the center coordinates of the heat source chips and form a variable matrix based on the number and type of heat source chips in the microwave component to be optimized; S2. Based on the size envelope of the microwave component to be optimized and the size of the heat source chip, set the range of values ​​for the center coordinates of the heat source chip. Then, based on the positional relationship between the heat source chips, obtain the heat source chip center coordinate relationship formula that prevents interference between the heat source chips. Then, based on the heat source chip center coordinate relationship formula, use Matlab software to randomly generate input data containing the variable matrix. S3. Establish a parametric simulation model in thermal simulation software; S4. Complete the setting of simulation boundary conditions; S5. Import some of the input data from step S2 into the parameterized simulation model and simulate it to obtain the simulation result. The simulation result is the highest temperature value of the microwave component to be optimized. S6. Select multiple sets of heat source chip center coordinates as input data for the improved DNN prediction model framework, and use the simulation results corresponding to the input data as output data to obtain data sets; S7. Establish an improved DNN prediction model; The framework of the improved DNN prediction model includes an input layer, a deep neural network containing 3 to 5 hidden layers, and an output layer connected in sequence. The hidden layers use residual connections to alleviate gradient vanishing in deep networks through shortcuts; the number of neurons in the hidden layers is dynamically adjusted, with the number of neurons in the first hidden layer set to 3 to 4 times the dimension of the input data, and the number of neurons in subsequent layers decreasing by 20%; and batch normalization layers are inserted between each layer to accelerate convergence. A channel attention mechanism is embedded after the hidden layer to automatically identify and enhance the feature weights of key heat source locations. At the same time, a lightweight CNN branch with a 3×1 convolutional kernel is added in parallel to extract local correlation features of the coordinate sequence. The output of the CNN branch is concatenated with the output of the DNN fully connected layer as the final feature. The improved DNN prediction model is trained based on at least 500 sets of the data. S8. Analyze the prediction results of the improved DNN prediction model, and obtain the trained improved DNN prediction model after the target is achieved.

2. The microwave component heat source distribution optimization method based on an improved DNN algorithm according to claim 1, characterized in that: In step S2, the input data is augmented by adding random perturbations within a specified range to the center coordinates of the heat source chip to simulate installation deviations in the actual process. At the same time, the thermal conductivity parameters of the heat source chip are randomly fluctuated within a specified range to generate an expanded sample. In step S5, the expanded sample is imported into the parameterized simulation model as the input data and simulated.

3. The microwave component heat source distribution optimization method based on an improved DNN algorithm according to claim 1, characterized in that: Step S3 includes the following steps: S3a. Simplify the initial design of the microwave component to be optimized by removing features that have little impact on heat source distribution optimization, including: threaded holes, chamfers, and lettering; S3b. In the DesignModeler software built into the simulation software ANSYS Workbench, the parametric simulation model is established based on the center coordinates of the heat source chip and the size of the microwave component to be optimized. S3c, Parameterize the center coordinates of the heat source chip.

4. The microwave component heat source distribution optimization method based on an improved DNN algorithm according to claim 1, characterized in that: Step S4 includes the following steps: S4a. Assign corresponding material properties to each part, wherein the parts include printed circuit board material, chip material, interface connection material and solder; S4b, Define the contact relationship between each part, the contact relationship including contact gap tolerance, contact pair and interface thermal resistance; S4c: Perform mesh generation and refine the mesh of the interface material of the heat source chip; S4d, Set the thermal boundary, and set the heat dissipation, convection coefficient, radiation coefficient and working time of each heat source chip; S4e, Take the overall temperature field distribution of the microwave component to be optimized as the solution objective.

5. The microwave component heat source distribution optimization method based on an improved DNN algorithm according to claim 1, characterized in that: Step S5 includes the following steps: S5a. After setting the boundary conditions, the highest temperature value is used as the simulation result and set as a parameter. S5b. Open the parameterization interface and import some of the input data obtained in step 2 into the parameterization simulation model; S5c, Click the Update All Design Points button to start parametric calculations; After the S5d and parametric calculations are completed, the simulation results are exported.

6. The microwave component heat source distribution optimization method based on an improved DNN algorithm according to claim 1, characterized in that: In step S7, the output layer of the improved DNN prediction model uses a dual output form of temperature mean + prediction variance, which quantifies the uncertainty of prediction while outputting the highest temperature prediction value; at the same time, a composite loss function that integrates mean square error and temperature threshold penalty term is constructed. For prediction errors that exceed the set safe temperature threshold, a weight of 2 to 3 times is given in the loss function, so that the improved DNN prediction model prioritizes the prediction accuracy of high temperature scenarios. The composite loss function is: ； in: For the sample size, To simulate the true value, The predicted value of the improved DNN prediction model is... These are the weighting coefficients; when hour, The value should be between 2 and 3, otherwise The value is 1. This is the safe temperature threshold.

7. The microwave component heat source distribution optimization method based on an improved DNN algorithm according to claim 1, characterized in that: In step S7, model training is performed based on a hierarchical learning rate strategy: the improved DNN prediction model is trained based on 800 sets of data. During training, a tiered learning rate setting is used: a smaller learning rate is applied to the feature extraction layer to preserve the original features, and a larger learning rate is applied to the temperature prediction layer for fast fitting. The smaller learning rate is 1e. -4 The maximum learning rate is 1e -3 ; Error information is extracted during the training process of the improved DNN prediction model, including MAE, MSE and RMSE, and error information statistics are performed.

8. The microwave component heat source distribution optimization method based on an improved DNN algorithm according to claim 1, characterized in that: The method for calculating the mean absolute error in step S8 is as follows: ； in: It is 100. To simulate the true value, The predicted value is the value of the improved DNN prediction model.

9. The microwave component heat source distribution optimization method based on an improved DNN algorithm according to claim 1, characterized in that: It also includes step S9: analyzing and verifying the optimal layout results obtained by the improved DNN prediction model, and the optimization of the heat source distribution of the microwave component is completed; Step S9 includes the following steps: S9a. The optimal layout result obtained by the trained improved DNN prediction model is input into the parameterized simulation model for simulation. The highest temperature value obtained by simulation is compared with the average temperature and confidence interval predicted by the trained improved DNN prediction model to verify whether the simulation result falls within the reasonable fluctuation range predicted by the model. S9b. When the simulation results fall within a reasonable fluctuation range, the optimized results of the microwave component heat source distribution are obtained. Based on the optimized results of the microwave component heat source distribution, the design is carried out, and the microwave component prototype is manufactured and then tested in practice.

10. The microwave component heat source distribution optimization method based on an improved DNN algorithm according to claim 2, characterized in that: In step S1, there are two types of heat source chips, and four of each type are used. The center coordinates of the heat source chips are ( x 1 ， y 1 )~( x 4 ，y 4 The variable matrix is ​​( x 1 ，y 1 ，x 2 ，y 2 ，x 3 ，y 3 ，x 4 ，y 4 ); In step S2, 1000 sets of the input data are randomly generated, and the specified range is ±5%. In step S5, the first 800 sets of the center coordinates of the heat source chip are imported into the parameterized simulation model; In step S7, the improved DNN prediction model is trained based on the 800 sets of data. In step S8, the prediction results of the 100 sets of the improved DNN prediction models are analyzed.

Images