Temperature fitting model construction method, smart panel and storage medium

Abstract

The invention discloses a temperature fitting model construction method, an intelligent panel and a storage medium, and relates to the field of computer systems based on specific calculation models.The method comprises the steps that equipment is placed in a real environment and an incubator environment, and data in the real environment and data in the incubator environment are collected respectively; various data in a real environment are preprocessed, main influence factors are synthesized, and simulation heat of equipment is designed and calculated; the value range of the simulated heat is adjusted, so that when the value is the maximum value, the device is in thermal stability, and when the value is the minimum value, the device is in a low-load or standby state; dividing a plurality of intervals on the basis of the startup time and the simulated heat, and constructing a segmented multi-parameter linear model A; training and optimizing the model A; preprocessing various data in an incubator environment, adjusting temperature deviation caused by the environment and setting simulation heat; the structure and parameters of the model A are fixed, a linear layer is added behind the model A to serve as a linear optimization model B, and the model B is trained and optimized.

Description

Technical Field

[0001] The present invention relates to the field of computer systems based on specific computing models, and particularly to a method for constructing a temperature fitting model. Background Art

[0002] With the rapid development of the Internet of Things (IoT) and smart home technologies, as the core interactive device of smart homes, smart panels have gradually become an important carrier for home environment monitoring and control. Among them, smart panels need to achieve real-time perception of environmental changes, display relevant data, and perform intelligent linkage with other home devices. Therefore, the real-time prediction and display of environmental temperature are particularly practical and important. Currently, the environmental monitoring function of smart panels is usually implemented by the following several solutions.

[0003] (1) Embedding a temperature sensor in the smart panel to directly collect and display environmental data. However, the embedded temperature sensor is affected by the heat generated by the device itself, especially under high load or long-term operation. Since the temperature sensor can usually only monitor the environment around the sensor, the temperature inside the device may be different from the external environment, and the temperature sensor is prone to deviation, resulting in inaccurate environmental data and being difficult to reflect the actual environmental data.

[0004] (2) Connecting to external sensors through wireless or wired protocols (such as Zigbee, Wi-Fi, RS485) to achieve flexible deployment and precise monitoring. Since external sensors rely on networks or other communication means to transmit data, once the network fails, data transmission will be interrupted, affecting the stability of the system. In addition, external sensors and network devices need to be purchased and maintained additionally, increasing the overall cost.

[0005] (3) Obtaining temperature data through a smart home system or meteorological service and displaying it on the smart panel. Since obtaining cloud data requires a stable network connection, if the network is disconnected, it may not be possible to obtain temperature data in real time, affecting the monitoring effect. Moreover, cloud meteorological data usually represents the average temperature of the surrounding area, rather than the real-time temperature accurate to a specific location. For the temperature control requirements in special scenarios or areas with large local environmental changes, cloud data may not provide sufficiently accurate temperature information.

[0006] (4) An integrated temperature sensor is used, and additional processing is performed on the collected ambient temperature, such as the temperature compensation method, device, equipment, medium, and intelligent panel disclosed in Publication No. CN115727962A. Its disadvantages are as follows: ① The environmental adaptability is weak, and the processing ability for complex environments and various factors is limited. It only relies on the temperature sensor data and the CPU temperature sensor data as the input variables of the model. When the environment is complex, the temperature fluctuates greatly, or the device load changes, the model may not be able to adapt to these changes well. ② The real-time response to dynamic changes is weak. It relies on the weighted average of the temperature prediction values at the previous moment and the current moment for smooth adjustment, but when facing sudden load changes or drastic fluctuations in the ambient temperature, the response speed is slow, and the temperature cannot be adjusted in time. Summary of the Invention

[0007] The purpose of the present invention is to provide a method for constructing a temperature fitting model, aiming to overcome the above problems existing in the prior art.

[0008] To achieve the purpose, the present invention provides the following technical solutions: A method for constructing a temperature fitting model includes the following steps: Step S1: Place the device in a real environment, control the main influencing factors affecting the change of the temperature sensor, and collect various data in the real environment; place the device in an incubator environment, control the temperature of the incubator according to a preset temperature range and temperature interval, and collect various data in the incubator environment.

[0009] Step S2: Preprocess the various data in the real environment, and record the startup time when the temperature value of the temperature sensor is stable from the startup to the screen-off of the device ; Considering the main influencing factors comprehensively, design and calculate the simulated heat of the device; adjust the value range of the simulated heat so that it satisfies: when the value is the maximum, it is when the device is thermally stable, and when the value is the minimum, it is when the device is in a low-load or standby state.

[0010] Step S3: Based on the startup time and the simulated heat, divide several intervals and construct a piecewise multi-parameter linear model A.

[0011] Step S4: Train and optimize model A, and output model A when the deviation value between the predicted temperature output by model A and the actual temperature meets the requirements.

[0012] Step S5: Preprocess the various data in the incubator environment, correct the temperature deviation caused by environmental factors, and set the simulated heat and startup time.

[0013] Step S6: Fix the structure and parameters of model A, and add a linear layer after model A as a linear optimization model B.

[0014] Step S7: Train and optimize Model B, and output Model B when the deviation between the predicted temperature output by Model B and the actual temperature meets the requirements.

[0015] Step S8: Integrate Model A and Model B into the device for actual testing and actual application.

[0016] Furthermore, the device is a smart panel, and the temperature sensor is an embedded temperature sensor installed on the smart panel. In Step S1, the main influencing factor is the heat generation caused by the screen turning on, including whether the screen is on or off, the screen brightness, and the screen brightness duration; the data under real environmental conditions at least includes the following items: the data of the device from power-on to the sensor temperature being stable; the data of the device from the screen turning on to the sensor temperature being stable when the screen brightness is set to different values, and the data of the sensor from the screen turning off to being stable after the sensor temperature is stable.

[0017] Furthermore, in Step S1, the data under the incubator environment at least includes the current temperature of the incubator and the temperature value of the sensor when the device is stable with the screen off.

[0018] Furthermore, in Step S2, the following formula is used to calculate the simulated heat of the device per minute :

[0019] where represents the simulated heat in the th minute; represents the continuous brightness value in the th minute, which is the current screen brightness when the screen is on and the screen brightness before turning off when the screen is off; represents the screen state in the th minute, represents the screen being on, represents the screen being off; represents the empirical brightness value set based on the maximum brightness.

[0020] Furthermore, in Step S3, Model A is specifically as follows:

[0021] where is the predicted temperature output by the model; is based on the predicted temperature output by the model; is the model input parameter; is the parameter that the model needs to be trained; is the device power-on time; is the empirical value of the power-on startup time determined based on the power-on startup time ; is the lower limit of the simulated heat value; is the upper limit of the simulated heat value; are the values of each interval of the simulated heat set according to empirical values.

[0022] Furthermore, in the step S4, training and optimizing the model A specifically includes: Training the model A using a deep learning tool; During the training process, select the smooth L1 loss function to monitor the change of the loss value and check whether the model converges; the smooth L1 loss function is as follows:

[0023] where, is the true label, is the predicted value of the model, is the number of data points, is the smooth L1 loss at a certain point, is a hyperparameter used to adjust the sensitivity of the loss function to the error; By different hyperparameter selections, segmented designs, and regularization methods, continuously optimize the coefficients of the model to minimize the deviation between the predicted temperature and the actual temperature; The model A outputs the linear regression equation of each segment.

[0024] Furthermore, in the step S7, the model B is specifically as follows:

[0025] where, is the empirical lower limit value set based on the lower limit of the temperature data range collected from the real environment; is the empirical upper limit value set based on the upper limit of the temperature data range collected from the real environment; is based on and the empirical temperature value set at the midpoint of, and are the parameters that need to be trained for this linear optimization model.

[0026] The present invention also discloses an intelligent panel, which is embedded with a temperature fitting model constructed by using the method described in any one of the above, and is used to predict the environmental temperature in real time.

[0027] The present invention also discloses a storage medium, which stores a computer-readable program internally, and this computer-readable program is used to execute the method described in any one of the above.

[0028] The present invention has the following beneficial effects compared with the prior art: The present invention overcomes multiple technical problems in the environmental temperature monitoring of equipment by training and optimizing a segmented multi-parameter temperature fitting model using deep learning tools, realizes accurate, real-time, and network-independent environmental temperature prediction and display, greatly improves the temperature management ability of the equipment, and reduces the hardware cost and external dependence at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a flowchart of the present invention.

[0030] Figure 2 is a schematic diagram of parameter data in a partial real environment. Among them, the ntc_temp curve is the temperature value of the NTC temperature, the sensor_temp curve is the temperature value of the temperature sensor, and the real_temp curve is the real environmental temperature.

[0031] Figure 3 is a schematic comparison diagram of partial real temperature and the predicted temperature output by Model A.

[0032] Figure 4 is a schematic comparison diagram of partial real temperature and the predicted temperature output by Model B. DETAILED DESCRIPTION OF THE INVENTION

[0033] The following describes the specific embodiments of the present invention with reference to the accompanying drawings. To fully understand the present invention, many details are described below, but for those skilled in the art, the present invention can be implemented without these details.

[0034] As Figure 1 shown, a method for constructing a temperature fitting model includes the following steps: Step S1: Place the device in a real environment, control the main influencing factors affecting the change of the temperature sensor, and collect various data in the real environment. Place the device in an incubator environment, control the temperature of the incubator according to a preset temperature range and temperature interval, and collect various data in the incubator environment.

[0035] In a specific embodiment, the above device is a smart panel, and the temperature sensor is an embedded temperature sensor installed on the smart panel. The main influencing factors affecting the change of the embedded temperature sensor are mainly affected by the heat generated by the smart panel screen being lit, including whether the screen is lit or not, the screen brightness, and the screen brightness duration. Therefore, the main influencing factors specifically include that the various data in the real environment at least include the following items: ① Data of the device from power-on to the temperature of the sensor being stable.

[0036] ② When the screen brightness is set to , data from the screen being lit to the temperature of the sensor being stable, and data from the screen being turned off to the sensor being stable after the temperature of the sensor is stable.

[0037] ③ The data from when the screen brightness is set to until the sensor temperature stabilizes when the screen is lit, and the data from when the screen is turned off until the sensor stabilizes after the sensor temperature has stabilized.

[0038] Of course, the setting of the screen brightness is not limited to and , and the screen brightness can be set to different values as needed, and the data of the device from when the screen is lit until the sensor temperature stabilizes, and the data from when the screen is turned off until the sensor stabilizes after the sensor temperature has stabilized are collected.

[0039] When the acquisition conditions permit, parameter data such as ①②③ should be collected in each temperature segment. For example, parameter data such as ①②③ are collected between an ambient temperature of 15°C and 20°C, and parameter data such as ①②③ are collected between 20°C and 25°C, etc. Figure 2 shows an example of collecting parameter data between 20°C and 25°C. The more diverse the collected data, the more accurate the temperature predicted by the model. The screen brightness value and the number of segments are set according to the actual hardware design and empirical values. The parameter data includes the boot-up time, screen brightness, whether the screen is turned off, the value of the temperature sensor (if there are multiple temperature sensors, they are collected together) at each moment, and the current actual ambient temperature.

[0040] In the incubator environment, the temperature range and temperature interval are preset according to the actual hardware design and empirical values. The data in the incubator environment at least includes the current temperature of the incubator, and the temperature value of the sensor temperature when the device is stable with the screen turned off (if there are multiple sensors, they are collected together). For example, between -20°C and 40°C, a batch of parameter data is collected at intervals of 5°C. The purpose of the collection is to make up for the temperature range that is difficult to simulate with the data collected in the real environment.

[0041] It should be noted that the amount of data collected should be large, diverse enough and representative enough, because the accuracy of the model highly depends on the quality of the input data.

[0042] Step S2, preprocess the data in the real environment, comprehensively consider the main influencing factors, design and calculate the simulated heat of the device; adjust the value range of the simulated heat so that it satisfies: when taking the maximum value, it is when the device is thermally stable, and when taking the minimum value, it is when the device is in a low-load or standby state.

[0043] In a specific embodiment, the simulated heat of the device per minute is calculated using the following formula :

[0044] Wherein, represents the simulated heat at the minute; Indicates the continuous brightness value at the th minute (i.e., when the device screen is on, it is the current screen brightness; when the device screen is off, it is the screen brightness before the screen goes off); Indicates the th minute of the screen status, Indicates that the screen is on, Indicates that the screen is off; Indicates the empirical brightness value set based on the maximum brightness. The meaning of the above formula is that if the screen is on, the simulated heat increases, and if the screen is off, the simulated heat decreases.

[0045] When preprocessing the data in the real environment, record the startup time when the temperature value of the temperature sensor on the device stabilizes from startup to after the screen goes off , and calculate the simulated heat per minute , and adjust the upper and lower limits of the simulated heat value. The rule is that when the device reaches the thermal stable state (for example, when the temperature value of the temperature sensor rises and then remains stable from the screen on), the value of the simulated heat should reach the maximum; when the device is in a low-load or standby state (for example, when the temperature value of the temperature sensor drops and then remains stable from the screen off), the value of the simulated heat should be the minimum.

[0046] Step S3: Divide several intervals based on the startup time and the simulated heat, and construct a piecewise multi-parameter linear model A.

[0047] In a specific embodiment, the model A is specifically as follows:

[0048] Wherein, is the predicted temperature output by the model; is the predicted temperature based on the output of the model; is the model input parameter; is the parameter that the model needs to be trained; is the device startup time; is based on the startup time determined empirical value of the startup time; is the lower limit of the simulated heat value; is the upper limit of the simulated heat value; is the value of each interval of the simulated heat set according to the empirical value.

[0049] Step S4: Train and optimize the model A, and output the model A when the deviation value between the predicted temperature output by the model A and the actual temperature meets the requirements.

[0050] Specifically, the model A is trained using deep learning tools, and the training data is various data of the real environment. The model inputs include data such as simulated heat, equipment startup time, temperature values of temperature sensors, etc., and the model output is the actual environmental temperature.

[0051] The model is trained using deep learning tools, and the goal of model training is to minimize the error between the predicted temperature and the actual temperature. The training process is an iterative process, which at least includes the following steps: (a1) Forward propagation: Calculate the input data through the model to obtain the prediction result.

[0052] (a2) Calculate the loss: Calculate the loss based on the predicted value and the true value of the model to measure the accuracy of the prediction.

[0053] (a3) Backward propagation: Calculate the gradient of each model parameter (weight) through the backward propagation algorithm.

[0054] (a4) Update the parameters: Use the optimizer to update the parameters of the model to reduce the value of the loss function.

[0055] During the training process, multiple training epochs are usually carried out, and each training epoch will iterate through the training data once. Each training epoch will perform steps such as forward propagation, calculating the loss, backward propagation, and updating the parameters. During the training process, monitor the change of the loss value to check whether the model converges.

[0056] For regression tasks, the SmoothL1Loss() function can be selected. Different from the ordinary L1 loss (absolute error) and L2 loss (squared error), the SmoothL1 loss behaves like the L2 loss when the error is small and like the L1 loss when the error is large, reducing the sensitivity to outliers.

[0057] The SmoothL1 loss function is specifically as follows:

[0058] Among them, is the true label, is the model prediction value, is the number of data points, is the SmoothL1 loss of a certain point, is a hyperparameter used to adjust the sensitivity of the loss function to the error. For small errors, the SmoothL1 loss function is approximately a quadratic loss (L2 loss), and for large errors, the SmoothL1 loss function is approximately a linear loss (L1 loss).

[0059] The role of the optimizer in step (a4) is to update the parameters of model A according to the gradient of the loss function. Adam (Adaptive Moment Estimation) is used here to adaptively adjust the learning rate.

[0060] Tune model A. When the deviation between the predicted temperature output by model A and the actual temperature meets the requirement, output model A.

[0061] Model tuning also involves adjusting hyperparameters. Hyperparameters that may need to be adjusted include: (1) Number of segments: Too many segments may lead to overfitting, while too few segments may fail to capture the details of the data; (2) Regularization term: prevents overfitting and improves the generalization ability of the model.

[0062] (3) Learning rate adjustment: to accelerate convergence and avoid overfitting, etc.

[0063] The performance of the model can be verified through cross-validation. It divides the dataset into multiple subsets, trains and tests the model on different training and validation sets, and thus can more comprehensively evaluate the stability and generalization ability of the model.

[0064] Through different hyperparameter selection, segmentation design, regularization methods, etc., the coefficients of the model are continuously optimized to minimize its prediction error. Finally, the model outputs the linear regression equation of each segment and its error evaluation index value.

[0065] For example, if the ambient temperature displayed by the smart panel is required to be within ±1.5°C of the actual ambient temperature, the model can be continuously tuned and a curve of the model-fitted temperature and the actual temperature can be output, and the deviation can be calculated to determine whether the model meets the requirement. Figure 3 As shown in the figure, the blue curve is the fitting temperature and the orange curve is the actual temperature. The maximum deviation between the two is 1.034℃. The model can be output if it meets the requirements.

[0066] Step S5, pre-processing various data under the temperature box environment, correcting the temperature deviation caused by environmental factors, and setting the simulated heat and start-up time.

[0067] Specifically, first, pre-process the incubator environment data, extract the parameter data of the equipment under low load or standby stable state at different temperatures, including the values of all temperature sensors and the incubator environment temperature; secondly, correct the deviation of the incubator environment temperature caused by environmental factors and the real environment temperature under the same conditions, and eliminate the temperature difference between the incubator environment and the real environment as much as possible; then, based on the rule of setting the simulated heat value in step S2, set the simulated heat to the minimum value within the range ; Finally, set the boot time to be greater than the value of which is the boot time empirical value determined by the boot start time in Model A .

[0068] An incubator (also called an environmental chamber or temperature control chamber) is usually used to simulate different temperature environments (high temperature or low temperature). The temperature inside it is regulated by a control system (such as a heater or a refrigeration system), which belongs to the prior art and will not be elaborated here. It should be noted that due to multiple factors such as the heating and cooling mechanisms, hermeticity, and equipment operating status inside the incubator, the temperature of the incubator environment measured by the temperature sensor based on the screen-off stable state is often different from the real environmental temperature.

[0069] For example, under the same standby stable conditions, the sensor temperature is 20 degrees, but the incubator environment temperature may be 25 degrees, while the real environmental temperature may be 23°. There is a difference between them. At this time, the recorded incubator environment temperature needs to be artificially corrected to 23° to simulate the real environmental temperature.

[0070] Therefore, it is necessary to perform de-difference processing on the temperature data of the incubator to correct the "temperature deviation" caused by the incubator environmental factors. The processing method can be based on calculating the difference between the real environmental temperature and the incubator environment temperature under the same input conditions, and adding a temperature compensation value to all incubator environment temperatures based on this difference.

[0071] Under low load or standby conditions, the power consumption of the device is relatively low, and the heat generation is also relatively small. Among the data in the incubator environment, select the data of the device under low load or standby conditions as sample data. For example, for a smart panel, the change of its embedded temperature sensor is mainly affected by the heat generated by the screen turning on. The data is selected as the data when the screen is in the stable state after turning off the screen after setting the temperature of the incubator according to the temperature range and temperature interval. At this time, the simulated heat value of the data is the minimum value within the range , the boot start time , which is the boot time empirical value determined by the boot start time in Model A . It should be noted that the above temperature range and temperature interval are set according to the actual hardware design and empirical values. For example, between -20°C and 40°C, a batch of parameter data is recorded at intervals of 5°C, aiming to make up for the temperature range that is difficult to simulate by the data collected from the real environment.

[0072] Step S6: Fix the structure and parameters of Model A, and add a linear layer after Model A as the linear optimization model B.

[0073] Step S7: Train and optimize Model B, and output Model B when the deviation value between the predicted temperature output by Model B and the actual temperature meets the requirements.

[0074] Specifically, use deep learning tools to train and optimize Model B. The training data is various data in the incubator environment. The model input and output are the same as those of Model A, that is, the model input includes data such as simulated heat, equipment startup time, and the values of temperature sensors, and the model output is the actual environmental temperature.

[0075] In a specific embodiment, Model B is as follows:

[0076] Wherein, is the predicted temperature output by Model A, is the empirical lower limit value set based on the lower limit of the temperature data range collected from the real environment; is the empirical upper limit value set based on the upper limit of the temperature data range collected from the real environment; is based on and the empirical temperature value set at the midpoint of, and are the parameters that need to be trained for this linear optimization model.

[0077] Use deep learning tools to train Model B. The training process, loss function selection, and optimizer selection refer to Step S4 and will not be elaborated here.

[0078] Tune Model B, and output Model B when the deviation value between the predicted temperature output by Model B and the actual temperature meets the requirements.

[0079] The tuning of Model B also involves adjusting hyperparameters. Referring to Step (4), the tuning can include adjusting , and . During the training and optimization process, evaluate the prediction effect of the model and adjust the model according to the target requirements until the deviation value between the predicted temperature and the actual temperature meets the requirements.

[0080] Step S8: Integrate Model A and Model B into the device for actual testing.

[0081] Adjust the optimized Model A and Model B into a format suitable for the production environment and integrate them into the embedded system of the device for actual testing. Of course, after the model is deployed to the hardware, it needs to be integrated with temperature sensors, display devices, etc. to ensure the acquisition of real-time data streams, predictive processing, and display output. Those skilled in the art can implement it in combination with the existing technology as needed, which will not be elaborated here. If, during the actual test, the prediction error of the model exceeds the expected threshold, the model can be further optimized based on the actual feedback.

[0082] For example, for the environmental temperature required to be displayed on the smart panel (i.e., the predicted temperature output by the model), the error from the real environmental temperature is within ±1.5°C. The model output is as Figure 4 shown. The blue curve in the figure is the fitted temperature, and the orange-yellow curve is the real temperature. The maximum deviation between the two is 0.46°C, meeting the requirements and the model can be output. Step S9: Deploy the tested models (including Model A and Model B) to the actual application to achieve real-time adaptive prediction and display of the environmental temperature.

[0083] It is necessary to ensure that the model has passed all optimizations and tests, and the error range (i.e., the deviation value) meets the requirements. Of course, after the system runs for a period of time, more data can be collected for incremental training. Through online learning or regular batch training, new environmental data is introduced into the model for updating to further improve the accuracy and stability of the model.

[0084] The present invention proposes a method for constructing a temperature fitting model for real-time monitoring of the environmental temperature of a smart panel. The main improvements are as follows: (1) Real-time calibration of temperature deviation: Compared with the "integrated temperature sensor display" of the traditional solution (1), the present invention uses a deep learning model to predict and calibrate in real time the temperature deviation caused by device heating and compensate for the difference between the temperature value of the temperature sensor and the external environment.

[0085] (2) Reduction of hardware costs: Compared with the "external sensor data access" of the traditional solution (2), the present invention does not rely on external sensors, eliminating the procurement and maintenance costs of additional devices and avoiding communication dependence on external devices.

[0086] (3) Avoidance of network dependence and cloud data deviation: Compared with the "cloud data acquisition" of the traditional solution (3), the present invention realizes local temperature prediction without relying on cloud data or network connection, avoiding temperature prediction errors caused by network latency or inaccurate cloud data.

[0087] (4) Strong real-time adaptability and better ability to handle complex environments: The mechanism of adaptively adjusting model parameters based on simulated heat changes can provide more accurate and stable temperature compensation in variable and complex environments.

[0088] In addition, the present invention also discloses a smart panel which is embedded with a temperature fitting model constructed by the above method for real-time prediction of the ambient temperature.

[0089] Specifically, the temperature fitting model is integrated (i.e., embedded) into the software or hardware system of the smart panel and becomes a part of its functions. The smart panel can be a smart screen or other terminals that need to display the ambient temperature.

[0090] The present invention also discloses a storage medium which stores a computer-readable program therein, and the computer-readable program is used to execute the above method.

[0091] The above are only specific embodiments of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantive modification made to the present invention using this concept shall fall within the scope of infringement of the protection scope of the present invention.

Claims

1. A method for constructing a temperature fitting model, characterized in that: Including the following steps, Step S1: Place the device in a real environment, control the main influencing factors affecting the change of the temperature sensor, and collect various data in the real environment; place the device in an incubator environment, control the temperature of the incubator according to the preset temperature range and temperature interval, and collect various data in the incubator environment; Step S2: Preprocess various data in the real environment and record the startup time when the temperature value of the temperature sensor stabilizes after the device is powered on until the screen is turned off. ; Considering the main influencing factors, design and calculate the simulated heat of the device; adjust the value range of the simulated heat so that it satisfies: when the value is at the maximum, it is when the device is thermally stable, and when the value is at the minimum, it is when the device is in a low-load or standby state; Step S3: Based on the boot time and the simulated heat, divide into several intervals and construct a piecewise multi-parameter linear model A; Step S4: Train and optimize model A, and output model A when the deviation value between the predicted temperature output by model A and the actual temperature meets the requirements; Step S5: Preprocess the various data in the incubator environment, correct the temperature deviation caused by environmental factors, and set the simulated heat and boot time; Step S6: Fix the structure and parameters of model A, and add a linear layer after model A as a linear optimization model B; Step S7: Train and optimize model B, and output model B when the deviation value between the predicted temperature output by model B and the actual temperature meets the requirements; Step S8: Integrate model A and model B into the device for actual testing and actual application.

2. The method for constructing a temperature fitting model according to claim 1, wherein: The device is an intelligent panel, and the temperature sensor is an embedded temperature sensor installed on the intelligent panel. In step S1, the main influencing factor is the heat generated by the screen turning on, including whether the screen is on or off, the screen brightness, and the screen brightness duration; The various data in the real environment at least include the following items: Data of the device from boot to the sensor temperature stabilizing; Data of the device from the screen turning on to the sensor temperature stabilizing and data of the sensor from the screen turning off to stabilizing after the sensor temperature has stabilized when the screen brightness is set to different values.

3. A method for constructing a temperature fitting model according to claim 1 or 2, characterized in that: In step S1, the various data in the incubator environment at least include the current temperature of the incubator and the temperature value of the sensor when the screen of the device is off and stable.

4. A method for constructing a temperature fitting model according to claim 1, characterized in that: In the step S2, the simulated heat per minute of the device is calculated using the following formula :[[]]END]] Among them, represents the simulated heat at the th minute; represents the continuous brightness value at the th minute, which is the current screen brightness when the screen is on and the screen brightness before the screen is turned off when the screen is off; represents the screen state at the th minute, represents the screen being on, represents the screen being off; represents the empirical brightness value set based on the maximum brightness.

5. A method for constructing a temperature fitting model according to claim 1, characterized in that: In step S3, model A is specifically as follows: Among them, is the predicted temperature output by the model; is the predicted temperature based on the output of the model; are the input parameters of the model; are the parameters that the model needs to be trained; is the device startup time; is based on the startup time determined empirical value of the startup time; is the lower limit of the simulated heat value; is the upper limit of the simulated heat value; are the values of each interval of the simulated heat set according to the empirical value.

6. A method for constructing a temperature fitting model according to claim 1 or 5, characterized in that: In step S4, training and optimizing model A specifically means: Use a deep learning tool to train model A; During the training process, select a smooth L1 loss function to monitor the change of the loss value and check whether the model converges; the smooth L1 loss function is as follows: Among them, is the true label, is the model prediction value, is the number of data points, is the smooth L1 loss of a certain point, is a hyperparameter used to adjust the sensitivity of the loss function to errors; Through different hyperparameter selections, piecewise designs, and regularization methods, continuously optimize the coefficients of the model to minimize the deviation value between the predicted temperature and the actual temperature; Model A outputs the linear regression equation for each segment.

7. A method for constructing a temperature fitting model according to claim 1, characterized in that: In step S7, model B is specifically as follows: Among them, is the predicted temperature output by Model A, is the empirical lower limit value set based on the lower limit of the temperature data range collected from the real environment; is the empirical upper limit value set based on the upper limit of the temperature data range collected from the real environment; is based on and the empirical temperature value set at the midpoint of, and are the parameters that need to be trained for this linear optimization model.

8. An intelligent panel, characterized in that: The intelligent panel is embedded with a temperature fitting model constructed by the method described in any one of claims 1-7 for real-time prediction of the environmental temperature.

9. A storage medium internally stores a computer-readable program, characterized in that: This computer-readable program is used to execute the method described in any one of claims 1-7.

Images