Wireless temperature recorder correction system and method

Abstract

The invention relates to the field of temperature recorder correction, in particular to a wireless temperature recorder correction system and method, and the system comprises a calibration terminal module, a reference temperature source module, a self-adaptive model matching module and a calibration compensation module. According to the invention, the reference temperature source module provides stable reference temperature by using a PID control algorithm; the adaptive model matching module determines an optimal model type and hyper-parameters by calculating skewness, kurtosis, curvature extreme point number and residual sum of squares of error distribution in combination with a random forest classifier and a Bayesian optimization algorithm; and the calibration compensation module carries out initialization and training according to the selected model and the hyper-parameter, and outputs a compensation coefficient to calibrate the recorder. According to the method, the temperature measurement precision can be effectively improved, collaborative optimization of sensor calibration and wireless transmission parameters is realized, the equipment management and calibration efficiency is enhanced, and powerful support is provided for accurate measurement and reliable operation of the wireless temperature recorder.

Description

Technical Field

[0001] The invention relates to the field of wireless temperature recorder calibration, in particular to a wireless temperature recorder calibration system and method. Background Art

[0002] Wireless temperature recorders are widely used in industrial monitoring, cold chain logistics, medical equipment and other fields. Their measurement accuracy directly affects product quality and process control reliability. The existing technology has the following significant problems: traditional temperature sensors are affected by ambient temperature changes during long-term use, which will produce nonlinear drift errors. The existing fixed compensation model is difficult to adapt to dynamic drift characteristics; the existing calibration method relies on manual operation or offline laboratory calibration, and cannot respond to environmental changes in real time. For example, industrial sites need to manually dismantle equipment for calibration every quarter, which is time-consuming and labor-intensive and interrupts monitoring, and cannot meet continuous production needs. Traditional compensation models cannot cope with complex error patterns; The present invention aims to solve the above problems by introducing a high-precision reference temperature source module, an adaptive model matching algorithm and Bayesian optimization technology to achieve dynamic compensation of temperature measurement errors, multi-device collaborative calibration and energy consumption optimization, thereby significantly improving the long-term stability and real-time performance of the wireless temperature recorder. Summary of the invention

[0003] In order to solve the technical problems raised by the above background technology, the present invention provides a wireless temperature recorder calibration system and method.

[0004] The purpose of the present invention can be achieved through the following technical solutions: The invention provides a wireless temperature recorder calibration system, comprising a calibration terminal module, a reference temperature source module, an adaptive model matching module, a calibration compensation module and a calibration database.

[0005] The calibration terminal module establishes a two-way data channel between the calibration terminal and multiple wireless temperature recorders, supports high-concurrency device access to establish a device management list, generates instructions and sends them to the wireless temperature recorder for reset, and collects ambient temperature data in real time through the reset recorder. The specific process is as follows: The calibration terminal module is provided with a communication control unit, a recorder acquisition unit and an error calculation unit; The communication control unit sets the calibration cycle to 24 hours, marks the wireless temperature recorder as a recorder to be calibrated, and the calibration terminal identifies all recorders to be calibrated and arranges them from strong to weak according to signal strength to establish a device management list. The identification methods include Bluetooth signal, NFC touch and QR code scanning. The device management list displays the online status of the recorder in real time, including registration status, ready status, calibrating and completion status; The calibration terminal generates a reset command and sends it to each recorder to be calibrated in turn. Each recorder to be calibrated eliminates the historical calibration deviation and is initialized to the factory parameters and updated to the ready state. If the recorder to be calibrated does not respond within 3 seconds, the retry mechanism is triggered and reset 3 times. If there is no response, the device is marked as a faulty recorder and sent to the corresponding operator; The recorder acquisition unit synchronizes the timestamps of each recorder to be calibrated and the calibration terminal, collects and reports the ambient temperature value and timestamp at a fixed period of 10 seconds, and sends them to the error calculation unit and the calibration database in sequence; The error calculation unit is used to receive the reference temperature value sent by the reference temperature source module and the ambient temperature value sent by the recorder acquisition unit, match the reference temperature value and the ambient temperature value of the same timestamp, subtract the matched reference temperature value and the ambient temperature value to obtain a measurement error value, and send it to the calibration database.

[0006] The reference temperature source module calculates the control quantity according to the PID control algorithm, adjusts the working state of the semiconductor refrigeration plate and the heating wire, makes the temperature of the reference temperature source close to the set temperature, and transmits the collected reference temperature value to the calibration terminal module and the calibration database. The specific process is as follows: The reference temperature source module includes several temperature standards as temperature acquisition devices, as well as semiconductor cooling sheets and heating wires as temperature adjustment elements. The actual temperature value is collected in real time through the temperature standards. ; Extract the constant temperature value set in the calibration database , the temperature difference is obtained by subtracting the constant temperature value from the actual temperature value ; Use PID control algorithm to calculate the control quantity u. The PID control algorithm formula is: ,in represents the control quantity at time t, represents the temperature difference at time t, represents the integral coefficient, represents the differential coefficient, It is expressed as the derivative of the error e(t) with respect to time t. It should be noted that by adjusting the two coefficients, the reference temperature source can quickly and stably reach the set temperature and control the temperature fluctuation within ±0.05°C. The working states of the semiconductor refrigeration plate and the heating wire are adjusted according to the control quantity u. If u is greater than 0, a heating instruction is generated and sent to the heating wire. The heating wire heats until a constant temperature value is reached and heating ends. When u is less than 0, a cooling instruction is generated and sent to the semiconductor refrigeration plate. The semiconductor refrigeration plate cools until a constant temperature value is reached and cooling ends. The reference temperature value when the environment reaches the constant temperature value is timestamped and sent to the calibration terminal module and the calibration database. It should be noted that the reference temperature source is a temperature environment generating device, which is used to provide traceable reference temperature values ​​for temperature sensors and recorders under test to ensure that the measurement results of the equipment under test are consistent with the true values.

[0007] The adaptive model matching module calculates the skewness and peak value of the error distribution and the number of extreme points of curvature of the temperature-error curve, and then obtains the residual sum of squares, inputs the real-time data into the random forest distributor, and outputs the optimal model type. The specific process is as follows: Extract the measurement error value of each time stamp in the calibration database and mark it as , calculated using the formula Get the skewness value of the error distribution , where n represents the total number of measurement error values, represents any measurement error value, It is expressed as the sample mean, and S represents the sample standard deviation; Then calculate according to the formula Get the kurtosis value of the error distribution , the meaning of each parameter in this formula is the same as that in the skewness formula. It should be noted that kurtosis is used to measure the peak or flatness of the data distribution; A temperature-error curve is established with the reference temperature value as the horizontal axis and the measured error value as the vertical axis. The reference temperature value is marked as , the corresponding measurement error value is marked as ,in , the first-order derivative is calculated according to the central difference method, specifically: When , use the formula of the forward phase difference method to calculate ,when When , use the central difference method to calculate ,when When , use the post-phase difference method to calculate Get the first-order derivative , also based on the central difference method, the second-order derivative is calculated according to the result of the first-order derivative. hour, ,when hour, ,when hour, This gives the second-order derivative , check the second-order derivative The sign change of determines the curvature extreme point. If the signs are different, it is a curvature extreme point. The number of all points that meet the conditions is counted to obtain the number of curvature extreme points TL; Finally, the residual sum of squares formula is used to calculate The residual sum of squares RSS is obtained, and the meanings of the other parameters are the same as those in the derivative formula. It is expressed as the prediction error value; Furthermore, the calculated skewness value, kurtosis value, number of extreme points of curvature and residual sum of squares are integrated into a feature vector ; Train the random forest classifier and extract the feature vectors of the historical calibration data , Z represents the total number of feature vectors, and obtains the optimal model type for each feature vector allocation , p represents the total number of optimal models, the optimal model type is to automatically select the most matching compensation model for the skewness, kurtosis, curvature extreme point and residual square sum in different temperature ranges and equipment states, and the model types include linear model, BP neural network model, polynomial fitting model and adaptive Kalman filter; match the feature vector and the corresponding optimal model type to obtain a training data set, divide the training data set into a training set and a test set, set the number of decision trees to 100, the maximum depth to 5, and the random number seed to 42 to train the model, so as to complete the training of the random forest allocator; obtain the feature vector of the real-time calibration data Input the trained random forest allocator, the classifier predicts the results of each decision tree, and finally outputs the predicted optimal model type; The optimal model hyperparameters are searched through the YES optimization algorithm, the Gaussian process is selected as the prior model to model the uncertainty of the objective function, and several initial hyperparameter combinations are randomly selected in the hyperparameter search space. , c represents the total number of initial hyperparameter combinations, and calculates the corresponding objective function value , and combine them into the initial data set , Represents any initial data, updates the parameters of the Gaussian process model according to the current data set to obtain the posterior distribution to complete the model update, maximizes the acquisition function in the hyperparameter search space, and obtains the next hyperparameter combination to be evaluated , calculate the objective function in The value at which the new sample is added to the dataset In the new data set ,When the function reaches the maximum number of iterations, the algorithm ends and outputs the currently known optimal hyperparameter combination And the corresponding objective function value ; The compensation calibration module initializes the compensation model according to the selected optimal model type and the hyperparameters obtained by Bayesian optimization, inputs the measured ambient temperature value into the initialized model, and outputs the compensation coefficient. The specific process is as follows: First, initialize the model through the optimal hyperparameter combination. If the optimal model is the BP neural network model, use the historical temperature data, historical reference temperature data and the corresponding measurement error to form a training set. Calculate the output through forward propagation, compare it with the actual error to get the loss, and then use the back propagation algorithm to update the weights and biases. Iterate continuously to minimize the loss function to complete the neural network compensation model training; input the current real-time measured ambient temperature value into the trained neural network compensation model training, and calculate the predicted error through forward propagation. , this prediction error is the compensation coefficient; If the optimal model is a linear model, the optimal slope and intercept are obtained by solving the normal equation, and the residual square sum between the predicted value and the true error is minimized, and the compensation model training has been completed; the prediction error is calculated based on the slope and intercept obtained by training to obtain If the error is greater than the preset error, the compensation coefficient is sent to the corresponding wireless temperature recorder. The wireless temperature recorder receives the compensation coefficient and makes adjustments. It then remeasures using the new parameters to verify whether the error converges. If not, it continues to calibrate, stores all adjustment records, adds timestamps, and sends them to the calibration database.

[0008] The second aspect of the present invention provides a wireless temperature recorder calibration method, the specific steps are as follows: Step 1, calibrate the terminal: establish a two-way data channel with multiple wireless temperature recorders, support high-concurrency device access, establish a device management list, generate instructions and send them to the wireless temperature recorder for reset, and collect the ambient temperature in real time through the reset recorder; Step 2, reference temperature source: calculate the control quantity according to the PID control algorithm, adjust the working state of the semiconductor cooling plate and the heating wire, make the temperature of the reference temperature source close to the set temperature, and transmit the collected reference temperature value to the calibration terminal module and the calibration database; Step 3: Adaptive model matching: According to the skewness and peak value of the calculated error distribution and the number of extreme points of curvature of the temperature-error curve, the residual sum of squares is obtained, the real-time data is input into the random forest distributor, and the optimal model type is output. Then, the dynamic sensor calibration and wireless transmission parameters are adjusted through the Yes optimization algorithm. Step 4: Compensation calibration: Initialize the model based on the selected optimal model type and the hyperparameters obtained by Bayesian optimization, input the measured ambient temperature value into the initialized model to output the compensation coefficient, and calibrate the wireless temperature recorder through the compensation coefficient Compared with the prior art, the present invention has the following beneficial effects: the compensation calibration module uses the hyperparameters obtained by Bayesian optimization to initialize and train the model, and performs error compensation according to the characteristics of different models, such as BP neural network model and linear model, thereby significantly reducing the measurement error and making the measurement result closer to the true value; The adaptive model matching module can effectively identify different error modes through in-depth analysis of the error distribution. For example, the asymmetry of the error distribution can be determined by skewness, the peak or flatness of the data distribution can be measured by kurtosis, the complexity of the relationship between temperature and error can be determined by the number of extreme points of curvature, and the long-term drift characteristics and periodic error patterns can be reflected by the residual sum of squares. Based on these analysis results, the system can select the most appropriate model from a variety of models such as linear models, BP neural network models, polynomial fitting models, and adaptive Kalman filters for error compensation, thereby improving the ability to adapt to complex errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. The following drawings are not intentionally scaled to the actual size, and the focus is on illustrating the main purpose of the present invention.

[0010] Figure 1 It is a block diagram of the system module connection of the present invention.

[0011] Figure 2 It is a diagram of the method steps of the present invention. DETAILED DESCRIPTION

[0012] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present invention.

[0013] Please refer to Figure 1 As shown, the present invention is a wireless temperature recorder calibration system, including a calibration terminal module, a reference temperature source module, an adaptive model matching module, a calibration compensation module and a calibration database.

[0014] It is understandable that the execution subject of the present application may be a wireless temperature recorder calibration system, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a terminal as the execution subject as an example.

[0015] The calibration terminal module establishes a two-way data channel between the calibration terminal and multiple wireless temperature recorders, supports high-concurrency device access to establish a device management list, generates instructions and sends them to the wireless temperature recorder for reset, and collects ambient temperature data in real time through the reset recorder. The specific process is as follows: The calibration terminal module is provided with a communication control unit, a recorder acquisition unit and an error calculation unit; The communication control unit sets the calibration cycle to 24 hours, marks the wireless temperature recorder as a recorder to be calibrated, and the calibration terminal identifies all recorders to be calibrated and arranges them from strong to weak according to signal strength to establish a device management list. The identification methods include Bluetooth signal, NFC touch and QR code scanning. The device management list displays the online status of the recorder in real time, including registration status, ready status, calibrating and completion status; The calibration terminal generates a reset command and sends it to each recorder to be calibrated in turn. Each recorder to be calibrated eliminates the historical calibration deviation and is initialized to the factory parameters and updated to the ready state. If the recorder to be calibrated does not respond within 3 seconds, the retry mechanism is triggered and reset 3 times. If there is no response, the device is marked as a faulty recorder and sent to the corresponding operator; The recorder acquisition unit synchronizes the timestamps of each recorder to be calibrated and the calibration terminal, collects and reports the ambient temperature value and timestamp at a fixed period of 10 seconds, and sends them to the error calculation unit and the calibration database in sequence; The error calculation unit is used to receive the reference temperature value sent by the reference temperature source module and the ambient temperature value sent by the recorder acquisition unit, match the reference temperature value and the ambient temperature value of the same timestamp, subtract the matched reference temperature value and the ambient temperature value to obtain a measurement error value, and send it to the calibration database.

[0016] The reference temperature source module calculates the control quantity according to the PID control algorithm, adjusts the working state of the semiconductor refrigeration plate and the heating wire, makes the temperature of the reference temperature source close to the set temperature, and transmits the collected reference temperature value to the calibration terminal module and the calibration database. The specific process is as follows: The reference temperature source module includes several temperature standards as temperature acquisition devices, as well as semiconductor cooling sheets and heating wires as temperature adjustment elements. The actual temperature value is collected in real time through the temperature standards. ; Extract the constant temperature value set in the calibration database , the temperature difference is obtained by subtracting the constant temperature value from the actual temperature value ; Use PID control algorithm to calculate the control quantity u. The PID control algorithm formula is: ,in represents the control quantity at time t, represents the temperature difference at time t, represents the integral coefficient, represents the differential coefficient, It is expressed as the derivative of the error e(t) with respect to time t. It should be noted that by adjusting the two coefficients, the reference temperature source can quickly and stably reach the set temperature and control the temperature fluctuation within ±0.05°C. The working states of the semiconductor refrigeration plate and the heating wire are adjusted according to the control quantity u. If u is greater than 0, a heating instruction is generated and sent to the heating wire. The heating wire heats until a constant temperature value is reached and heating ends. When u is less than 0, a cooling instruction is generated and sent to the semiconductor refrigeration plate. The semiconductor refrigeration plate cools until a constant temperature value is reached and cooling ends. The reference temperature value when the environment reaches the constant temperature value is timestamped and sent to the calibration terminal module and the calibration database. It should be noted that the reference temperature source is a temperature environment generating device, which is used to provide traceable reference temperature values ​​for temperature sensors and recorders under test to ensure that the measurement results of the equipment under test are consistent with the true values.

[0017] The adaptive model matching module calculates the skewness and peak value of the error distribution and the number of extreme points of curvature of the temperature-error curve, and then obtains the residual sum of squares, inputs the real-time data into the random forest distributor, and outputs the optimal model type. The specific process is as follows: Extract the measurement error value of each time stamp in the calibration database and mark it as , calculated using the formula Get the skewness value of the error distribution , where n represents the total number of measurement error values, represents any measurement error value, It is expressed as the sample mean, S represents the sample standard deviation. It should be noted that by analyzing the skewness of the error value, the asymmetry of the error distribution can be determined; Then calculate according to the formula Get the kurtosis value of the error distribution , the meaning of each parameter in this formula is the same as that in the skewness formula. It should be noted that kurtosis is used to measure the peak or flatness of the data distribution; A temperature-error curve is established with the reference temperature value as the horizontal axis and the measured error value as the vertical axis. The reference temperature value is marked as , the corresponding measurement error value is marked as ,in , the first-order derivative is calculated according to the central difference method, specifically: When , use the formula of the forward phase difference method to calculate ,when When , use the central difference method to calculate ,when When , use the post-phase difference method to calculate Get the first-order derivative , also based on the central difference method, the second-order derivative is calculated according to the result of the first-order derivative. hour, ,when hour, ,when hour, This gives the second-order derivative , check the second-order derivative The sign change of determines the curvature extreme point. Different sign, then is the curvature extreme point, and the number of all points that meet the conditions is counted to obtain the number of curvature extreme points TL; it should be noted that in the temperature-error curve, if the curvature on the left side of a point is positive and the curvature on the right side is negative, then the point is called a curvature point of opposite sign. The more the number of curvature extreme points is, the more complex the relationship between temperature and error is, and there may be multiple nonlinear relationships; Finally, the residual sum of squares formula is used to calculate The residual sum of squares RSS is obtained, and the meanings of the other parameters are the same as those in the derivative formula. It is expressed as the prediction error value; it should be noted that the calculation of the residual sum of squares and the temperature measurement error has a cumulative effect. RSS can reflect the long-term drift characteristics and combine with the kurtosis index to identify the periodic error pattern; Furthermore, the calculated skewness value, kurtosis value, number of extreme points of curvature and residual sum of squares are integrated into a feature vector ; Train the random forest classifier and extract the feature vectors of the historical calibration data , Z represents the total number of feature vectors, and obtains the optimal model type for each feature vector allocation , p represents the total number of optimal models, the optimal model type is to automatically select the most matching compensation model for the skewness, kurtosis, curvature extreme point and residual square sum in different temperature ranges and equipment states, and the model types include linear model, BP neural network model, polynomial fitting model and adaptive Kalman filter; match the feature vector and the corresponding optimal model type to obtain a training data set, divide the training data set into a training set and a test set, set the number of decision trees to 100, the maximum depth to 5, and the random number seed to 42 to train the model, so as to complete the training of the random forest allocator; obtain the feature vector of the real-time calibration data Input the trained random forest allocator, the classifier predicts the results of each decision tree, and finally outputs the predicted optimal model type; The optimal model hyperparameters are searched through the YES optimization algorithm, the Gaussian process is selected as the prior model to model the uncertainty of the objective function, and several initial hyperparameter combinations are randomly selected in the hyperparameter search space. , c represents the total number of initial hyperparameter combinations, and calculates the corresponding objective function value , and combine them into the initial data set , Represents any initial data, updates the parameters of the Gaussian process model according to the current data set to obtain the posterior distribution to complete the model update, maximizes the acquisition function in the hyperparameter search space, and obtains the next hyperparameter combination to be evaluated , calculate the objective function in The value at which the new sample is added to the dataset In the new data set When the function reaches the maximum number of iterations, the algorithm ends and outputs the currently known optimal hyperparameter combination And the corresponding objective function value ; It should be noted that one of the core challenges of wireless temperature recorders is the impact of ambient temperature drift on sensor accuracy. For example, in an industrial high-temperature environment, the sensitivity and offset of the sensor will change with temperature. The Yes optimization algorithm dynamically optimizes sensor calibration and wireless transmission parameters to achieve coordinated optimization of high-precision measurement and low-power transmission. The compensation calibration module initializes the compensation model according to the selected optimal model type and the hyperparameters obtained by Bayesian optimization, inputs the measured ambient temperature value into the initialized model, and outputs the compensation coefficient. The specific process is as follows: First, initialize the model through the optimal hyperparameter combination. If the optimal model is the BP neural network model, use the historical temperature data, historical reference temperature data and the corresponding measurement error to form a training set. Calculate the output through forward propagation, compare it with the actual error to get the loss, and then use the back propagation algorithm to update the weights and biases. Iterate continuously to minimize the loss function to complete the neural network compensation model training; input the current real-time measured ambient temperature value into the trained neural network compensation model training, and calculate the predicted error through forward propagation. , this prediction error is the compensation coefficient; If the optimal model is a linear model, the optimal slope and intercept are obtained by solving the normal equation, and the residual square sum between the predicted value and the true error is minimized, and the compensation model training has been completed; the prediction error is calculated based on the slope and intercept obtained by training to obtain If the error is greater than the preset error, the compensation coefficient is sent to the corresponding wireless temperature recorder. The wireless temperature recorder receives the compensation coefficient and makes adjustments. It then remeasures using the new parameters to verify whether the error converges. If not, it continues to calibrate. All adjustment records are stored and timestamped and sent to the calibration database. The status is updated to completion in the device management list.

[0018] like Figure 2 As shown, the second aspect of the present invention provides a wireless temperature recorder calibration method, the specific steps are as follows: Step 1, calibrate the terminal: establish a two-way data channel with multiple wireless temperature recorders, support high-concurrency device access, establish a device management list, generate instructions and send them to the wireless temperature recorder for reset, and collect the ambient temperature in real time through the reset recorder; Step 2, reference temperature source: calculate the control quantity according to the PID control algorithm, adjust the working state of the semiconductor cooling plate and the heating wire, make the temperature of the reference temperature source close to the set temperature, and transmit the collected reference temperature value to the calibration terminal module and the calibration database; Step 3: Adaptive model matching: According to the skewness and peak value of the calculated error distribution and the number of extreme points of curvature of the temperature-error curve, the residual sum of squares is obtained, the real-time data is input into the random forest distributor, and the optimal model type is output. Then, the dynamic sensor calibration and wireless transmission parameters are adjusted through the Yes optimization algorithm. Step 4: Compensation calibration: Initialize the model based on the selected optimal model type and the hyperparameters obtained by Bayesian optimization, input the measured ambient temperature value into the initialized model to output the compensation coefficient, and calibrate the wireless temperature recorder through the compensation coefficient.

[0019] The above is an explanation of the present invention and should not be considered as a limitation thereof. Although several exemplary embodiments of the present invention have been described, it will be readily appreciated by those skilled in the art that many modifications may be made to the exemplary embodiments without departing from the novel teachings and advantages of the present invention. Therefore, all such modifications are intended to be included within the scope of the present invention as defined in the claims. It should be understood that the above is an explanation of the present invention and should not be considered as being limited to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The present invention is defined by the claims and their equivalents.

Claims

1. A wireless temperature recorder calibration system, comprising a calibration terminal module, a reference temperature source module, an adaptive model matching module, a calibration compensation module and a calibration database, characterized in that: The calibration terminal module is provided with a communication control unit, a recorder acquisition unit and an error calculation unit; The adaptive model matching module calculates the skewness and peak value of the error distribution and the number of extreme points of curvature of the temperature-error curve, and then obtains the residual sum of squares, inputs the real-time data into the random forest distributor, and outputs the optimal model type; The compensation calibration module initializes the compensation model according to the selected optimal model type and the optimal hyperparameters obtained by Bayesian optimization, and inputs the measured ambient temperature value into the initialized model to output the compensation coefficient. Specifically, the model is initialized by the optimal hyperparameter combination. If the optimal model is a BP neural network model, the historical temperature data, the historical reference temperature data and the corresponding measurement error are used to form a training set. The output is calculated by forward propagation, and the loss is obtained by comparing it with the actual error. The weights and biases are updated by the back propagation algorithm, and the training of the neural network compensation model is completed by continuous iteration to minimize the loss function. The current real-time measured ambient temperature value is input into the trained neural network compensation model for training, and the compensation coefficient is obtained through forward propagation calculation; If the optimal model is a linear model, the optimal slope and intercept are obtained by solving the normal equations, and the residual sum of squares between the predicted value and the true error is minimized, and the compensation model training is completed; the prediction error is calculated based on the slope and intercept obtained from the training. If the error is greater than the preset error, the compensation coefficient is sent to the corresponding wireless temperature recorder. The wireless temperature recorder receives the compensation coefficient and adjusts it, and then re-measures with the new parameters to verify whether the error converges. If not, continue to calibrate.

2. A wireless temperature recorder calibration system according to claim 1, characterized in that: The adaptive model matching module calculates the skewness and peak value of the error distribution and the number of extreme points of curvature of the temperature-error curve, and then obtains the residual sum of squares. The specific process is as follows: Extract the measurement error value of each time stamp in the calibration database and mark it as , calculated using the formula Get the skewness value of the error distribution , where n represents the total number of measurement error values, represents any measurement error value, It is expressed as the sample mean, and S represents the sample standard deviation; According to the formula Calculate the kurtosis value of the error distribution ; A temperature-error curve is established with the reference temperature value as the horizontal axis and the measured error value as the vertical axis. The reference temperature value is marked as , the corresponding measurement error value is marked as ,in , the first-order derivative is calculated according to the central difference method, specifically: When , use the formula of the forward phase difference method to calculate ,when When , use the central difference method to calculate ,when When , use the post-phase difference method to calculate Get the first-order derivative , also based on the central difference method, the second-order derivative is calculated according to the result of the first-order derivative. hour, ,when hour, ,when hour, This gives the second-order derivative , check the second-order derivative The sign change of determines the curvature extreme point. Different sign, then is the curvature extreme point, and the number of all points that meet the conditions is counted to obtain the number of curvature extreme points TL; Finally, the residual sum of squares formula is used to calculate Get the residual sum of squares RSS, where It is expressed as the prediction error value.

3. A wireless temperature recorder calibration system according to claim 2, characterized in that: The adaptive model matching module trains the random forest allocator, inputs real-time data into the random forest allocator, and outputs the optimal model type, specifically: The calculated skewness value, kurtosis value, number of extreme curvature points and residual sum of squares are integrated into a feature vector; The random forest classifier is trained to extract the feature vectors of the historical calibration data in the calibration database, and the optimal model types for each feature vector assignment are obtained. The model types include linear model, BP neural network model, polynomial fitting model and adaptive Kalman filter. Match the feature vectors with the corresponding optimal model type to obtain the training data set, divide the training data set into a training set and a test set, set the number of decision trees to 100, the maximum depth to 5, and the random number seed to 42 to train the model, thereby completing the training of the random forest allocator; Get the feature vector of real-time calibration data The trained random forest allocator is input, and the classifier predicts the results of each decision tree and finally outputs the predicted optimal model type.

4. A wireless temperature recorder calibration system according to claim 3, characterized in that: The adaptive model matching module searches for the hyperparameters of the optimal model through the Yes optimization algorithm, specifically: Select Gaussian process as the prior model to model the uncertainty of the objective function, and randomly select several initial hyperparameter combinations in the hyperparameter search space. , c represents the total number of initial hyperparameter combinations, and calculates the corresponding objective function value , and combine them into the initial data set , Represents any initial data, updates the parameters of the Gaussian process model according to the current data set to obtain the posterior distribution to complete the model update, maximizes the acquisition function in the hyperparameter search space, and obtains the next hyperparameter combination to be evaluated , calculate the objective function in The value at which the new sample is added to the dataset In the new data set ,When the function reaches the maximum number of iterations, the algorithm ends and outputs the currently known optimal hyperparameter combination And the corresponding objective function value .

5. A wireless temperature recorder calibration system according to claim 1, characterized in that: The reference temperature source module calculates the control quantity according to the PID control algorithm, adjusts the working state of the semiconductor refrigeration plate and the heating wire, makes the temperature of the reference temperature source close to the set temperature, and transmits the collected reference temperature value to the calibration terminal module and the calibration database. The specific process is as follows: The reference temperature source module includes several temperature standards as temperature acquisition devices, as well as semiconductor cooling sheets and heating wires as temperature adjustment elements. The actual temperature value is collected in real time through the temperature standards. ; Extract the constant temperature value set in the calibration database , the temperature difference is obtained by subtracting the constant temperature value from the actual temperature value ; Use PID control algorithm to calculate the control quantity u. The PID control algorithm formula is: ,in represents the control quantity at time t, represents the temperature difference at time t, represents the integral coefficient, represents the differential coefficient, Expressed as the derivative of the error e(t) with respect to time t; The working states of the semiconductor refrigeration plate and the heating wire are adjusted according to the control quantity u. If u is greater than 0, a heating instruction is generated and sent to the heating wire. The heating wire heats until a constant temperature value is reached and heating ends. When u is less than 0, a cooling instruction is generated and sent to the semiconductor refrigeration plate. The semiconductor refrigeration plate cools until a constant temperature value is reached and cooling ends. The reference temperature value when the environment reaches the constant temperature value is timestamped and sent to the calibration terminal module and calibration database.

6. A wireless temperature recorder calibration system according to claim 1, characterized in that: The communication control unit establishes a two-way data channel according to the calibration terminal and multiple wireless temperature recorders, supports high-concurrency device access to establish a device management list, generates instructions and sends them to the wireless temperature recorder for reset. The specific process is as follows: The calibration cycle is set to 24 hours, and the wireless temperature recorder is marked as a recorder to be calibrated. The calibration terminal identifies all recorders to be calibrated and arranges them from strong to weak according to signal strength to establish a device management list. The identification methods include Bluetooth signal, NFC touch and QR code scanning. The device management list displays the online status of the recorder in real time, including registration status, ready status, calibrating and completion status; The calibration terminal generates a reset instruction and sends it to each recorder to be calibrated in turn. Each recorder to be calibrated eliminates the historical calibration deviation and is initialized to the factory parameters, and is updated to the ready state. If the recorder to be calibrated does not respond within 3 seconds, the retry mechanism is triggered and reset 3 times. If there is no response, the device is marked as a faulty recorder and sent to the corresponding operator.

7. A wireless temperature recorder calibration system according to claim 6, characterized in that: The recorder acquisition unit collects ambient temperature data in real time through the reset recorder, specifically: The timestamps of each recorder to be calibrated and the calibration terminal are synchronized, and the ambient temperature value and timestamp are collected and reported at a fixed period of 10 seconds, and sent to the error calculation unit and the calibration database in turn.

8. A wireless temperature recorder calibration system according to claim 7, characterized in that: The error calculation unit is used to receive the reference temperature value sent by the reference temperature source module and the ambient temperature value sent by the recorder acquisition unit, match the reference temperature value and the ambient temperature value of the same timestamp, subtract the matched reference temperature value and the ambient temperature value to obtain a measurement error value, and send it to the calibration database.

9. A wireless temperature recorder calibration method, characterized in that: A wireless temperature recorder calibration system applied to any one of claims 1 to 8, wherein the specific steps are: Step 1: Calibrate the terminal: Establish a two-way data channel with multiple wireless temperature recorders, support high-concurrency device access to establish a device management list, and generate instructions to send to the wireless temperature recorder for reset. The reset recorder collects the ambient temperature in real time; Step 2, reference temperature source: calculate the control quantity according to the PID control algorithm, adjust the working state of the semiconductor cooling plate and the heating wire, make the temperature of the reference temperature source close to the set temperature, and transmit the collected reference temperature value to the calibration terminal module and the calibration database; Step 3: Adaptive model matching: According to the skewness and peak value of the calculated error distribution and the number of extreme points of curvature of the temperature-error curve, the residual sum of squares is obtained, the real-time data is input into the random forest distributor, and the optimal model type is output. Then, the dynamic sensor calibration and wireless transmission parameters are adjusted through the Yes optimization algorithm. Step 4: Compensation calibration: Initialize the model based on the selected optimal model type and the hyperparameters obtained by Bayesian optimization, input the measured ambient temperature value into the initialized model to output the compensation coefficient, and calibrate the wireless temperature recorder through the compensation coefficient.

Images