High-precision temperature measuring equipment correction system

Abstract

The invention discloses a high-precision temperature measuring equipment correction system, and relates to the technical field of temperature measuring equipment correction, and the high-precision temperature measuring equipment correction system mainly comprises a hardware layer, an algorithm layer and an embedded layer which are electrically connected with one another. The hardware layer is used for acquiring temperature data, humidity data and pressure data; the hardware layer comprises a temperature sensor array, a humidity sensor, a pressure sensor and an analog-to-digital converter; the algorithm layer is used for correcting the temperature data according to the temperature data, the humidity data and the pressure data, and the algorithm layer comprises a data preprocessing module, a hybrid correction module and a dynamic compensation module; and the embedded layer is used for task scheduling and temperature data output. By implementing the high-precision temperature measurement equipment correction system provided by the invention, the correction precision, efficiency and reliability of the temperature measurement equipment can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of temperature measuring equipment correction, and more particularly to a high-precision temperature measuring equipment correction system. BACKGROUND

[0002] Temperature measurement technology has developed for many years, and has formed a traditional technical system represented by hardware compensation method, table lookup interpolation method, polynomial fitting method and two-point calibration method. These methods meet the temperature measurement needs of basic industries and consumer electronics in different historical stages, but in the current application scenarios of high precision, high dynamic and complex environment, their inherent defects are increasingly prominent. The existing technology generally faces three major contradictions. The first contradiction is the contradiction between precision and complexity. Hardware compensation and polynomial fitting need to balance between precision improvement and resource consumption, and it is difficult to balance wide temperature range, high precision and low computational load. The second contradiction is the contradiction between static calibration and dynamic environment. Traditional calibration relies on laboratory calibration data and cannot adapt to dynamic changes such as sudden temperature changes and environmental disturbances. The third contradiction is the contradiction between speciality and universality. Existing solutions are mostly designed for specific sensors or scenarios, and lack a unified correction framework across platforms and multiple sensors. These contradictions lead to the bottleneck of traditional technology in emerging fields such as high-end manufacturing, biomedicine and new energy. Hardware compensation method is to physically correct the output signal of temperature sensor through precise resistance network, temperature sensitive element or feedback circuit. Its typical implementations include: bridge compensation circuit: using the Wheatstone bridge balance principle, the non-linear error of the sensor is offset by configuring compensation resistance. For example, in platinum resistance (PT100) temperature measurement, the output curve is adjusted by connecting / parallel precise resistance network; temperature sensitive element compensation: using auxiliary elements (such as thermistor) complementary to the main sensor characteristics to superimpose signals and offset the influence of environmental temperature on the main sensor; feedback compensation: a closed-loop feedback system is built by operational amplifier to dynamically adjust the sensor operating point to suppress noise. Application scenarios: commonly used in industrial temperature control instruments, high stability temperature measurement modules, etc., especially suitable for devices with high short-term stability requirements. Its limitations are: 1. Limited compensation range: linear compensation of resistance network is only effective within a limited temperature range (such as ±20℃), and the error increases significantly beyond the range. For example, the residual error of a certain type of platinum resistance after hardware compensation is still up to ±0.2℃ at 100℃; 2. Environmental sensitivity: the compensation elements are affected by factors such as humidity and vibration, and long-term use may cause parameter drift. Experimental data shows that the additional error introduced by the compensation circuit can reach ±0.15℃ after 1000 hours of continuous work in an environment with humidity >80%; 3. Temperature gradient effect: when there is a spatial temperature difference between the sensor and the compensation element (such as rapid temperature change scenarios), the compensation accuracy decreases sharply. For example, in industrial furnace temperature monitoring, if the temperature difference between the sensor and the compensation resistance is 5℃, it will cause a measurement deviation of ±0.3℃.Its typical process includes: calibration point collection: record the output voltage value of the sensor at multiple temperature points (such as one point every 10℃) in the standard temperature field; table storage: store the calibration data in non-volatile memory (such as EEPROM); real-time interpolation: according to the current voltage value, linear interpolation is performed between adjacent calibration points to calculate the temperature. Application scenarios: widely used in digital thermometers, consumer electronics and other cost-sensitive and narrow temperature range scenarios. Its limitations are reflected in: 1. Storage space and precision contradiction: to achieve ±0.1℃ precision, at least 200 calibration points need to be set in the range of-50~150℃, occupying more than 4KB of storage space (calculated by 16-bit data). This is a significant burden for resource-constrained embedded systems (such as 8-bit MCUs); 2. Non-linear error accumulation: in the inflection point area of the sensor characteristic curve (such as the low temperature section of NTC thermistor), the linear interpolation error can reach 5%~8% of the actual value. For example, a certain type of NTC sensor has an error of ±1.2℃ at-30℃ when using table lookup method; 3. Aging failure problem: the characteristic drift of the sensor after long-term use will cause the table data to fail. Research shows that the resistance value drift rate of NTC sensor can reach 2%~5% after 3 years of use, directly leading to more than 50% reduction in lookup table precision. Polynomial fitting method establishes a function relationship model between the output voltage of the sensor and the temperature, usually using a second or third order polynomial for curve fitting. The mathematical model is: ; Where the coefficients are obtained by least squares fitting of calibration data. Application scenarios: suitable for calibration of low non-linear error sensors for scientific instruments, medical devices, etc. Its limitations are reflected in: 1. High-order non-linear compensation is insufficient: when the sensor characteristics contain high-order non-linearity (such as exponential characteristics), the fitting residual of the third-order polynomial can still reach ±0.5℃. For example, the fitting error of a certain type of infrared temperature sensor at high temperature (>300℃) is more than ±1℃; 2. Overfitting and underfitting contradiction: low-order polynomials (such as second-order) are prone to underfitting in a wide temperature range, while high-order polynomials (such as fifth-order) are prone to overfitting when calibrated with small samples. Experiments show that when the number of calibration points is less than 10, the temperature prediction error of the fifth-order polynomial expands to ±0.8℃; 3. Consumption of computing resources: third-order polynomial fitting requires at least 6 floating-point multiplication operations, which requires more than 500 clock cycles on embedded processors without hardware FPU (such as STM32F0 series), resulting in a decrease in real-time performance. Two-point calibration method establishes a linear conversion model by selecting two reference temperature points (usually 0℃ ice water mixture and 100℃ boiling water) (where the slope k and intercept are calculated by two-point calibration): ; Application scenario: mainly used for low-cost, wide-range temperature sensor calibration, such as household water heater temperature control system. Its limitations are reflected in: 1. Linear assumption failure: the non-linear error of actual sensors (such as thermocouples) in the non-calibration point area is significant. For example, the error of the two-point calibration method for K-type thermocouples at 500°C can reach ±4°C; 2. Calibrating conditions are harsh: the calibration environment needs to be strictly controlled (such as the boiling point temperature under standard atmospheric pressure), and field calibration is difficult to implement. Studies have shown that for every 1000 meters of altitude, the boiling point temperature drops by about 3.3°C, directly leading to the invalidation of calibration parameters; 3. Temperature drift without compensation: sensor zero drift (such as 0.1% FS per month) will cause error accumulation after long-term use. The additional error caused by drift of a certain type of pressure type temperature sensor after 6 months of use is ±0.25°C.

[0003] It can be seen that the main defects of the prior art are: firstly, environmental interference sensitivity problem: the existing method generally lacks a synergistic suppression mechanism for multiple environmental interference factors (humidity, electromagnetic noise, air pressure fluctuation); humidity effect: humidity changes by changing the dielectric constant of the sensor packaging material, resulting in changes in parasitic capacitance. For example, the output of a certain type of MEMS temperature sensor drifts by 0.12℃ when the humidity rises from 30% to 90%; electromagnetic interference: the wideband electromagnetic noise (typical intensity 50mVpp) generated by frequency converters, motors and other equipment in industrial sites will be coupled to the sensor signal line, and traditional RC filtering is difficult to completely suppress, resulting in ±0.05~0.1℃ random error; air pressure effect: for sensors that rely on heat conduction (such as thermocouples), air pressure fluctuations (such as ±10kPa) will change the medium heat transfer efficiency, resulting in ±0.08℃ additional error. Secondly, insufficient compensation for nonlinear errors: traditional methods have theoretical limitations in dealing with complex nonlinear characteristics; segmented linearization defects: when the wide temperature range is divided into multiple subintervals for local linearization, the discontinuity of the derivative at the interval boundary will cause error jumps. For example, the error jump of a certain type of infrared sensor at the 100℃ segmentation point is ±0.3℃; high-order term neglect problem: polynomial fitting usually only considers nonlinear terms up to the third order, while actual sensors (such as semiconductor temperature sensors) may contain high-order nonlinear components above the fifth order, with residual error up to ±0.2℃. Thirdly, calibration dependence and cost problem: the existing calibration technology has the problem of excessive dependence on laboratory conditions; standard equipment dependence: high-precision calibration requires the use of professional equipment such as constant temperature tanks (cost >10,000), standard platinum resistance thermometers (precision ±0.01℃), resulting in a single calibration cost of more than 200; long calibration period: traditional multi-point calibration method requires stable for several hours in multiple temperature zones, and full-range calibration takes 8~12 hours, which cannot meet the demand of on-site rapid calibration. Fourthly, the lack of dynamic temperature drift compensation: the existing algorithm lacks real-time compensation capability for the following dynamic factors; self-heating effect: self-heating caused by sensor working current (such as 0.1℃ temperature rise caused by 1mA excitation current in RTD); thermal hysteresis effect: the asymmetric response of the sensor during temperature rise and fall, the hysteresis error of a certain type of thin film platinum resistance can reach ±0.07℃; aging drift: long-term drift caused by structural relaxation of sensor sensitive material, typical rate is 0.05℃ / year.Fifth, the resource occupation of the embedded system is too high: the traditional algorithm faces severe resource constraints in embedded implementation; memory consumption: the look-up table method + interpolation algorithm needs to occupy more than 20KB of Flash storage space (based on 200 calibration points), which exceeds the available resources of a low-end MCU (such as STM32F030, Flash = 32KB); calculation load: a third-order polynomial fitting needs more than 1ms of calculation time on a single-chip microcomputer without a hardware floating-point unit (based on a 16MHz main frequency), which cannot meet the real-time requirements of high-speed sampling (>100SPS) systems; power consumption problem: the continuous high-load operation (>60% utilization rate) of the CPU caused by complex algorithms shortens the standby time of the device by more than 50%.

[0004] Therefore, the current mainstream temperature correction technology is limited by the single-dimensional compensation strategy, static mathematical model and discretization processing mode, and presents systematic short boards in dealing with complex industrial scenes. The essence of these defects lies in that the traditional technology simplifies temperature measurement as an isolated system of "sensor-linear model", ignoring environmental coupling, dynamic time-varying, resource constraints and other realistic factors. As a result, the existing scheme faces an adaptability crisis in emerging scenes such as smart medical care, and it is urgent to realize technological leap through innovative architecture of multidisciplinary cross. SUMMARY

[0005] The purpose of the present application is to provide a high-precision temperature measurement equipment correction system that can improve the accuracy, efficiency and reliability of temperature measurement equipment correction.

[0006] The present application provides a high-precision temperature measurement equipment correction system, which comprises a hardware layer, an algorithm layer and an embedded layer connected to each other; the hardware layer is used to acquire temperature data, humidity data and pressure data; the hardware layer comprises a temperature sensor array, a humidity sensor, a pressure sensor and an analog-to-digital converter; the algorithm layer is used to correct the temperature data according to the temperature data, humidity data and pressure data, and the algorithm layer comprises a data preprocessing module, a hybrid correction module and a dynamic compensation module; the embedded layer is used for task scheduling and outputting temperature data.

[0007] Further, the above-mentioned analog-to-digital converter is a 24-bit sigma-delta type analog-to-digital converter, which adopts a rate of 200 SPS.

[0008] Further, the above-mentioned data preprocessing module is used for sliding average filtering and outlier rejection of input data, and the preprocessed data is obtained according to an environmental interference model.

[0009] Further, the above-mentioned environmental interference model is, for example, the formula: , wherein, is the temperature measurement deviation compensation amount caused by environmental interference, is humidity, P is air pressure, coefficient , , is an experimental calibration coefficient.

[0010] Further, the mixed correction module is used for mixed correction of the preprocessed data to obtain error feedback correction data.

[0011] Further, the mixed correction of the preprocessed data to obtain error feedback correction data comprises: confirming that the temperature is less than a preset temperature threshold, and performing 3-order polynomial correction on the preprocessed data to obtain error feedback correction data; and confirming that the temperature is less than the preset temperature threshold, and performing correction on the preprocessed data by using a 3-order polynomial correction method and a neural network compensation method to obtain error feedback correction data.

[0012] Further, the temperature threshold is 100 .

[0013] Further, the dynamic compensation module is used for dynamic compensation of the error feedback correction data by using a sliding window and a difference integrated moving average autoregressive model to obtain corrected data.

[0014] The application further provides an application of the high-precision temperature measuring equipment correction system, which is applied to the field of temperature measuring equipment correction.

[0015] The high-precision temperature measuring equipment correction system has the following beneficial effects: The application aims at the key bottlenecks of traditional temperature measurement technology in nonlinear modeling, environmental anti-interference, dynamic compensation and embedded implementation, utilizes a three-level collaborative system of "sensing-algorithm-embedded", utilizes the algorithm layer to implement three-level processing, that is, the data preprocessing module performs sliding average filtering and outlier rejection, the mixed correction module processes nonlinear error and the dynamic compensation module realizes drift suppression through time series prediction, utilizes the temperature-humidity-pressure multi-physical field coupling correction mechanism, realizes the quality change improvement of anti-interference ability in complex industrial scenes through the quantitative modeling and dynamic decoupling of environmental interference, utilizes the collaborative correction framework of segmented polynomial and lightweight neural network to realize high-precision approximation in a wide temperature range under the constraint of embedded resources, solves the contradiction that the traditional method cannot be compatible with precision and efficiency, constructs an adaptive compensation engine based on time series prediction, through the closed-loop mechanism of advance prediction and feedback correction, the temperature mutation response speed is improved to milliseconds, breaking through the dynamic response limit of the traditional steady-state model, through the collaborative optimization of memory management, calculation acceleration and power consumption control, the embedded layer constructs a multi-task system based on RTOS, adopts the memory pool management technology to reduce resource occupation, realizes low-power operation through the event triggering mechanism, solves the key bottleneck problems of traditional temperature measurement equipment in precision, stability, environmental adaptability and resource efficiency, realizes the generational breakthrough in the three dimensions of precision, efficiency and reliability, embodies the characteristics of high precision, high reliability, high efficiency, low resource consumption, low environmental impact and wide adaptability, through the multi-dimensional optimization of "precision-energy efficiency-cost-environmental protection", provides a new generation of temperature sensing infrastructure for industrial intelligent upgrading, can realize significant performance breakthrough and comprehensive benefit improvement in the fields of industry, medical treatment, energy and the like. BRIEF DESCRIPTION OF DRAWINGS

[0016] The application will be further described below in combination with the drawings and examples, and the drawings are as follows: Figure 1 It is the high-precision temperature measurement equipment correction system composition block diagram provided by the application; Figure 2 It is the system overall architecture schematic diagram provided by the application; Figure 3 It is the segmented strategy and neural network fusion architecture schematic diagram provided by the application; Figure 4 It is the dynamic supplementary algorithm schematic diagram provided by the application; Figure 5 It is the embedded optimization schematic diagram provided by the application; Figure 6 It is the overall technical implementation flowchart provided by the application. DETAILED DESCRIPTION

[0017] In order to have a more clear understanding of the technical features, purposes and effects of the application, the specific embodiments of the application will be described in detail with reference to the drawings.

[0018] Figure 1 A schematic diagram of the high-precision temperature measurement device correction system of the present embodiment is shown. In the high-precision temperature measurement device correction system of the present embodiment, a hardware layer, an algorithm layer and an embedded layer are electrically connected to each other; the hardware layer is configured to acquire temperature data, humidity data and pressure data; the hardware layer comprises a temperature sensor array, a humidity sensor, a pressure sensor and an analog-to-digital converter; the algorithm layer is configured to correct the temperature data based on the temperature data, the humidity data and the pressure data, and the algorithm layer comprises a data preprocessing module, a hybrid correction module and a dynamic compensation module; and the embedded layer is configured to perform task scheduling and output temperature data.

[0019] In an exemplary embodiment, the analog-to-digital converter is a 24-bit sigma-delta analog-to-digital converter, which adopts a rate of 200 SPS; In an exemplary embodiment, the data preprocessing module is configured to perform sliding average filtering and outlier rejection on the input data, and obtain preprocessed data based on an environmental interference model; In an exemplary embodiment, the environmental interference model is as shown in the following formula: , wherein, is a temperature measurement deviation compensation amount caused by environmental interference, is humidity, is air pressure, and the coefficient , , is an experimental calibration coefficient; In an exemplary embodiment, the hybrid correction module is configured to perform hybrid correction on the preprocessed data to obtain error feedback correction data; In an exemplary embodiment, the hybrid correction on the preprocessed data to obtain error feedback correction data comprises: confirming that the temperature is less than a preset temperature threshold, performing 3-order polynomial correction on the preprocessed data to obtain error feedback correction data; and confirming that the temperature is less than the preset temperature threshold, performing correction on the preprocessed data by using a 3-order polynomial correction method and a neural network compensation method to obtain error feedback correction data. In an exemplary embodiment, the temperature threshold is 100 ; In an exemplary embodiment, the dynamic compensation module performs dynamic compensation on the error feedback correction data by using a sliding window and a difference integrated moving average autoregressive model to obtain corrected data. The present embodiment provides an application of the above-mentioned high-precision temperature measurement device correction system, which is applied to temperature measurement device correction.

[0020] In some embodiments, the high-precision temperature measurement device correction system described above can also be implemented in the following way.

[0021] This embodiment constructs a "sensing-algorithm-embedded" three-level collaborative system (as shown in Figure 2 The hardware layer uses a multi-sensor array (temperature / humidity / pressure) to access a 24-bit sigma-delta ADC to collect raw data at a rate of 200 SPS. The algorithm layer implements three levels of processing, including a data preprocessing module for sliding average filtering and outlier rejection, a mixed correction model (piecewise polynomial + neural network) for processing nonlinear errors, and a dynamic compensation engine for drift suppression through time series prediction. The embedded layer is based on RTOS and uses memory pool management technology to reduce resource occupation by 60%. The event triggering mechanism is used to achieve low-power operation. Now, combined with Figure 2 , the following describes each layer: 1. Hardware layer: integrate multiple types of environmental sensors and realize signal digitization through high-precision ADC; 2. Algorithm layer: build a complete processing chain including data cleaning, mixed model correction, and dynamic compensation; 3. Embedded layer: use real-time operating system for task scheduling, and integrate low-power communication and wake-up mechanism.

[0022] This architecture realizes a complete closed loop from physical signal collection to calibration output. Through the iterative optimization mechanism of anomaly detection-mixed correction-parameter self-learning, the system can still maintain a measurement accuracy of ±0.05℃ in complex environments, while the memory occupation of the embedded system is controlled within 40KB.

[0023] The technical innovation of the embodiment takes "multi-dimensional perception-intelligent correction-dynamic optimization" as the core logic, and overcomes the key bottlenecks in non-linear modeling, environmental anti-interference, dynamic compensation and embedded implementation of traditional temperature measurement technology through a systematic architecture design of multi-disciplinary cross-fusion. Different from the single optimization path of the prior art, the embodiment builds a full-chain innovation system from the physical perception layer, the algorithm processing layer to the embedded execution layer, and its core breakthroughs are reflected in the following four dimensions: first, a multi-source data fusion architecture: breaking through the traditional single sensor dependence mode, a temperature-humidity-pressure multi-physical field coupling correction mechanism is created, through quantitative modeling and dynamic decoupling of environmental interference, the anti-interference capability in complex industrial scenes is improved; second, a mixed nonlinear correction model: a segmented polynomial and lightweight neural network collaborative correction framework is proposed, which realizes high-precision approximation in a wide temperature range (-200℃~1800℃) under embedded resource constraints, and solves the contradiction between precision and efficiency of traditional methods; third, a dynamic compensation algorithm: an adaptive compensation engine based on time series prediction is constructed, through the closed-loop mechanism of advance prediction and feedback correction, the temperature mutation response speed is improved to milliseconds, breaking through the dynamic response limit of traditional steady-state models; fourth, an embedded optimization technology: a lightweight implementation scheme for resource-constrained systems is developed, through the collaborative optimization of memory management, computation acceleration and power consumption control, the system simultaneously reaches the industry benchmark level in resource occupation (RAM<10%), real-time (delay<2ms) and energy efficiency (power consumption<1.5mA). The synergistic effect forms three technical breakthroughs: at the theoretical level, a temperature correction differential equation model of multi-physical field coupling is established, breaking through the mathematical limitations of traditional linearization assumptions; at the method level, a parameter self-learning mechanism of mixed correction model is developed, realizing the paradigm shift from laboratory calibration to on-site adaptation; at the engineering level, a modular embedded architecture is designed, which is the first to realize the full-function integration of a high-precision temperature measurement system with a precision of sub-0.1℃ on a low-cost MCU; fundamentally solving the tetrahedron contradiction of "precision-dynamic-environment-resource".

[0024] 1. Compared with the traditional hardware compensation method, the embodiment adopts a multi-source data fusion architecture; A temperature-humidity-pressure multi-dimensional sensing matrix is constructed, and multi-sensor data is fused through Kalman filtering; an environmental interference model is established: ; Among them, is humidity, is air pressure, and the coefficients , , are coefficients obtained through experimental calibration.

[0025] Compared with the traditional hardware compensation method which can only compensate for fixed environmental interference, the embodiment realizes dynamic compensation: the humidity influence suppression rate is increased by 82% (from ±0.12℃ to ±0.02℃); the pressure fluctuation compensation range is expanded to ±50kPa (the traditional method is only ±10kPa). In addition, the embodiment adopts a sensor self-checking mechanism, which automatically switches to a pressure-temperature correlation model when the humidity sensor fails, and the system reliability is improved by 60%.

[0026] 2. Compared with the polynomial fitting / table lookup method, the embodiment adopts a hybrid nonlinear correction model; For the wide temperature range nonlinear problem, the embodiment combines segmented polynomials with lightweight neural networks. Under the condition of limited hardware resources, a 5-layer micro network based on Q16 fixed-point numbers (input layer 5 nodes, hidden layer 8 / 4 nodes) is designed, and the quantization error is controlled to be less than or equal to 0.01℃ through a dynamic scaling factor. The input layer receives temperature / environmental parameters, the hidden layer reduces the dimension of parameter quantity, and the output is a compensation value. Only 500 instructions are required for a single inference, and the computational load is reduced by 95% compared to the floating-point scheme, and the RAM occupancy is compressed to 1.2KB. An improved L-BFGS optimization algorithm is used for weight training, and the storage of the Hessian inverse matrix is reduced from O(n²) to O(mn) (fixed memory depth m=5), and the matrix operation is accelerated using DSP instructions, and the time consumption of a single iteration is shortened from 18ms to 0.9ms. Through the cooperation of the above-mentioned 5-layer micro network based on Q16 fixed-point numbers and the improved L_BFGS optimization algorithm, the Q16 network reduces storage by 90%, the improved L-BFGS improves training efficiency by 80%, and solves the resource-precision contradiction in embedded scenarios.

[0027] The measured data shows that in the low temperature region of-50℃, the residual error of the third-order polynomial fitting is less than or equal to 0.03℃; in the high temperature region of 300℃, the error after neural network compensation is less than or equal to 0.05℃. Compared with the traditional table lookup method, the storage space requirement is reduced by 57% (from 42KB to 18KB), and when the sensor characteristics drift by 2% due to aging, the model self-adaptive adjustment can control the error increase to within 0.008℃, which is significantly better than the 0.15℃ error deterioration of the fixed table scheme.

[0028] The segmented strategy and neural network fusion architecture is as shown in Figure 3 In the low temperature region (T<100℃), a 3-order polynomial is used for fast calculation (only 15μs@48MHz); in the high temperature region (T≥100℃), a 5-order polynomial is used for basic compensation, and a 3-layer neural network (structure: 5-8-4-1) is used for residual error compensation.

[0029] Wide temperature range adaptability: maximum error ≤0.08℃ in the range of -50~300℃ (traditional 5-order polynomial ±0.35℃); resource optimization: Flash occupancy is only 18.6KB (table lookup method needs 42KB to achieve the same accuracy); online learning ability: neural network weights are updated automatically every 24 hours, and long-term drift rate is reduced to 0.005℃ / year.

[0030] 3. Compared with static calibration method, the embodiment adopts dynamic compensation algorithm; The embodiment realizes the prediction and advance compensation of temperature drift by constructing a sliding window (length N=2000) difference integrated moving average autoregressive model (ARIMA) time series model. The key technologies include: (1) Automatic differentiation order selection (BIC criterion optimization): based on the Bayesian information criterion (BIC), the differentiation order d of the ARIMA model is dynamically optimized. By real-time calculation:

[0031] (wherein is the residual variance, k is the parameter quantity, and n is the sample number), the system traverses d∈{0,1,2} and selects the optimal order corresponding to the minimum BIC value. This method breaks through the limitation of traditional fixed order, adaptively eliminates non-stationarity, improves the temperature prediction accuracy by 23%, and avoids model distortion caused by over / under differentiation; (2) Parameter online updating (model retraining every 60 seconds): the sliding window recursive mechanism is used to realize 60-second-level model retraining. The latest 2000 samples (10-second data window) are retained, combined with the decay factor λ=0.98λ=0.98 weighted new data, and the ARIMA coefficients are updated online through the recursive least squares method (RLS). The calculation complexity is compressed to , so that the response delay of the model is less than 0.8 seconds when the temperature suddenly changes (>5℃ / s), and the efficiency is improved by 15 times compared with the traditional offline update; (3) Fuzzy control of compensation amount (membership function width=0.02℃): based on the prediction error e and its change rate Δe, a fuzzy rule base is constructed to strictly constrain the membership function width to 0.02℃ to realize accurate nonlinear mapping: when |e|≤0.01℃∣e∣≤0.01℃, output zero compensation; 0.01<|e|≤0.03℃0.01<∣e∣≤0.03℃, start proportional compensation; |e|>0.03℃∣e∣>0.03℃, trigger ±0.05℃ saturated compensation. This design compresses the temperature fluctuation standard deviation to 0.04℃, and the overshoot suppression rate is >90%, completely solving the overcompensation problem of the static calibration method.

[0032] Industrial field test shows that in the 10℃ / s rapid temperature change scenario, the measurement stabilization time is shortened from 3.2 seconds of the traditional method to 0.8 seconds, and the overshoot is reduced by 91% (from 0.25℃ to 0.02℃). At the same time, the compensation efficiency of the sensor self-heating effect reaches 92%, making the zero drift after 8 hours of continuous work ≤0.005℃, which is better than 0.08℃ of the uncompensated system.

[0033] As shown in Figure 4 , the sliding window, ARIMA model and compensation trigger realize the following functions: sliding window: cache the last 10 seconds of data (2000 sample points); ARIMA model: automatic optimization of parameters (p=2, d=1, q=1); compensation trigger: activate the compensation engine when the prediction error >0.03℃.

[0034] Dynamic response capability: the stabilization time is shortened from 3.2s to 0.8s when the temperature suddenly changes (10℃ / s); drift suppression: the self-heating effect compensation efficiency reaches 92% (traditional method without compensation); calculation efficiency: only 1.2ms is needed for a single prediction on the Cortex-M4 core (traditional FEM simulation needs >5s).

[0035] 4. Compared with the traditional implementation, the embodiment adopts embedded optimization technology to realize; The embodiment breaks through the resource bottleneck through a three-level optimization strategy: memory management: use block memory pool (12KB fixed + 4KB dynamically allocated) to reduce memory fragmentation; operation acceleration: design a mixed precision calculation framework (temperature value 32-bit floating point, weight parameter 16-bit fixed point); power consumption control: intermittent working mode based on event trigger (sleeping ratio 83%). The actual measurement on the STM32F407 platform shows that the complete algorithm process only occupies 9.8KB RAM (traditional scheme needs 128KB), and the CPU utilization rate is reduced from 68% to 22%. In terms of power consumption, the average working current is reduced from 8.7mA to 1.2mA, making the device using CR2032 button cell battery last from 7 days to 48 days, meeting the low power consumption demand of industrial Internet of Things devices. The technical implementation is as shown in Figure 5 , which shows the superiority comparison: 1. resource occupation: RAM usage is reduced from 128KB to 9.8KB (reduced by 92%); CPU utilization rate is reduced from 68% to 22% (sampling rate remains 200SPS); 2. real-time improvement: interrupt delay is shortened from 15μs to 1.8μs; 3. power consumption optimization: average working current is reduced from 8.7mA to 1.2mA.

[0036] 5. Overall technical implementation; Organic synergy is achieved through hierarchical and modular technical architecture, and the overall technical implementation path is as shown in Figure 6The flow is taken as the input starting point of multi-source data fusion, processed by a mixed model nonlinear correction and dynamic compensation engine, and finally outputs the calibration result through the embedded optimization layer to form a closed-loop optimization system of "perception-correction-output". The data interaction and function coupling between the modules embody the following core logic: data-driven decision: real-time fusion of environmental parameters and temperature signals provides dynamic input reference for the correction model; model adaptive switching: temperature interval judgment and error feedback constitute the autonomous decision mechanism of the mixed model; dynamic allocation of resources: the task scheduler of the embedded layer dynamically adjusts the memory and computing resources according to the algorithm complexity. The comparison structure is shown in Table 1. Table 1: Summary of technical advantages

[0037] Experiments show that the present application solves the key bottleneck problems of traditional temperature measurement equipment in precision, stability, environmental adaptability and resource efficiency, etc., and realizes significant performance breakthrough and comprehensive benefit improvement in the fields of industry, medicine, energy, etc. High precision: based on the synergistic effect of multi-source data fusion and mixed nonlinear correction model, the full-range temperature measurement precision is improved by more than 80% compared with traditional methods. In the range of-50~300℃, the maximum absolute error is ≤±0.08℃, and the precision improvement in the high and low temperature limit area is 93% (traditional method error ±1.2℃→the present application ±0.08℃); high reliability: the combination of dynamic compensation algorithm and embedded optimization technology makes the anti-interference ability of the system in complex industrial environment improved by more than 5 times; high efficiency: the whole process efficiency from production operation to equipment maintenance is significantly optimized; low resource consumption: the breakthrough optimization of embedded system makes the resource occupation greatly reduced; low environmental impact: precise temperature control technology brings significant environmental benefits; wide adaptability: the present scheme covers multiple industry scenarios through modular design.

[0038] The embodiments of the present application are described above in combination with the drawings, but the present application is not limited to the above specific embodiments, and the above specific embodiments are only illustrative and not limiting. Those skilled in the art can make many forms under the inspiration of the present application without departing from the purpose of the present application and the scope protected by the claims, and these all belong to the protection of the present application.

Claims

1. A high-precision temperature measuring device correction system, characterized by, The temperature measurement device comprises a hardware layer, an algorithm layer and an embedded layer which are electrically connected with each other; the hardware layer is used for acquiring temperature data, humidity data and pressure data; the hardware layer comprises a temperature sensor array, a humidity sensor, a pressure sensor and an analog-to-digital converter; the algorithm layer is used for correcting the temperature data according to the temperature data, the humidity data and the pressure data, and the algorithm layer comprises a data preprocessing module, a hybrid correction module and a dynamic compensation module; and the embedded layer is used for task scheduling and outputting temperature data.

2. The high-precision temperature measuring device correction system according to claim 1, characterized in that, The analog-to-digital converter is a 24-bit sigma-delta analog-to-digital converter, and the rate thereof is 200 SPS.

3. The high-precision temperature measuring device correction system according to claim 1, characterized in that, The data preprocessing module is used for performing sliding average filtering and abnormal value elimination on input data, and obtaining preprocessed data according to an environmental interference model.

4. The high-precision temperature measuring device correction system according to claim 3, characterized in that, The environmental interference model is, for example, a formula as follows: ， wherein, is the temperature measurement deviation compensation amount caused by environmental interference, is the humidity, is the air pressure, coefficient , , is the experimental calibration coefficient.

5. The high-precision temperature measuring device correction system according to claim 1, characterized in that, The hybrid correction module is used for performing hybrid correction on the preprocessed data to obtain error feedback correction data.

6. The high-precision temperature measuring device correction system according to claim 5, wherein The hybrid correction on the preprocessed data to obtain error feedback correction data comprises: When the temperature is less than a preset temperature threshold, performing 3-order polynomial correction on the preprocessed data to obtain error feedback correction data; When the temperature is less than a preset temperature threshold, performing correction on the preprocessed data by using a 3-order polynomial correction method and a neural network compensation method to obtain error feedback correction data.

7. The high-precision temperature measuring device correction system according to claim 6, characterized in that The temperature threshold is 100 .

8. The high-precision temperature measuring device correction system according to claim 1, wherein The dynamic compensation module performs dynamic compensation on the error feedback correction data by using a sliding window and a difference integrated moving average autoregressive model to obtain corrected data.

9. Use of a high-precision temperature measuring device correction system according to any one of claims 1 to 8, characterized in that The application is applied to the field of temperature measurement device correction.

Images