A method for fast measurement of indoor temperature based on thermal equilibrium final state prediction

Abstract

The application relates to a kind of indoor temperature fast measurement methods based on heat balance terminal state prediction, it is related to measurement technical field, the terminal temperature can be inferred by intercepting transient data in initial very short time window, the waiting time of traditional need is shortened to three minutes, the work efficiency of heating into the house inspection is improved by several times, the positioning environment center point mechanism is introduced in the measurement process, the interference of local cold and heat source and boundary layer convection in wall corner, window edge and the like is effectively avoided, the standardization of measurement characteristics is guaranteed from physical space source, by extracting temperature change rate and temperature change acceleration after smoothing processing, the response difference caused by different initial temperature difference is eliminated, both preventing algorithm divergence by using physical law and correcting environmental perturbation error by using model, cooperate with adaptive volatility locking mechanism, not only ensure the accuracy of final output result, but also effectively reduce the reactive operation energy consumption of portable device.

Description

Technical Field

[0001] This invention relates to the field of measurement technology, and in particular to a rapid indoor temperature measurement method based on thermal equilibrium final state prediction. Background Technology

[0002] In the production management of the heating industry and residential heating services, accurate collection of indoor temperature data is the core basis for assessing heating quality and achieving precise control. Currently common mobile indoor temperature measurement methods mainly suffer from the following technical shortcomings: I. Technical Drawback of Long Thermal Equilibrium Time: Temperature sensors (such as thermistors and NTC sensors) are affected by their own heat capacity, requiring a considerable amount of time to achieve thermal equilibrium and output stable readings after entering a new environment. Existing mobile temperature measurement devices typically require 10 to 30 minutes or more to reach thermal equilibrium under different initial operating conditions, severely restricting the efficiency of large-scale in-home temperature measurement. Chinese patent application CN107290008A discloses a rapid temperature measurement method, but it still requires waiting for the sensor reading to stabilize, failing to fundamentally solve the technical problem of long thermal equilibrium time. II. Technical Deficiencies of Lack of Compensation Mechanism for Operating Condition Migration Errors: Temperature measurement personnel typically carry equipment from low-temperature outdoor environments or high-temperature environments in heated rooms to the target temperature measurement room, resulting in a significant deviation between the initial sensor temperature and the target ambient temperature. Since the sensor response curve characteristics differ significantly under different initial operating conditions, a single, fixed compensation method cannot effectively address all operating conditions. Chinese patent application CN108362345A discloses a temperature compensation method, but it only compensates for a single operating condition and lacks a multi-condition adaptive compensation mechanism.

[0003] III. Fundamental Limitations of Existing Time Series Prediction Methods: Existing technologies contain numerous temperature prediction methods, all targeting the next value of the time series or the value of the series several steps in the future. This is a typical time series extrapolation problem. As is well known to those skilled in the art, time series extrapolation methods (such as ARIMA and LSTM time series prediction) attempt to infer the short-term evolution direction of a system based on historical data. Their prediction target is the conditional expectation of the sequence at future moments. When dealing with non-stationary processes like sensor thermal response, which are driven by external forces, the prediction results diverge rapidly with the number of extrapolation steps, making it impossible to reliably provide an accurate estimate of the final thermal equilibrium state.

[0004] To address the aforementioned technical deficiencies, a solution is proposed. Summary of the Invention

[0005] The purpose of this invention is to introduce a positioning environment center point mechanism during the measurement process. By extracting the smoothed temperature change rate and temperature change acceleration, the response differences caused by different initial temperature differences are eliminated. This not only uses physical laws to prevent algorithm divergence, but also uses the model to correct environmental perturbation errors. Combined with an adaptive volatility locking mechanism, this not only ensures the accuracy of the final output results, but also effectively reduces the reactive power consumption of portable devices.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A rapid indoor temperature measurement method based on thermal equilibrium final state prediction includes the following steps: S1. Record the initial temperature value at the moment the temperature sensor enters the target environment; S2. Move the temperature sensor to position it at the center point of the environment, and continuously collect temperature time series data within a short time window [t0, tn] at the center point of the environment at a set sampling frequency, where tn is less than the time required for the temperature sensor to reach physical thermal equilibrium. S3. The temperature time series data is smoothed and filtered, and the temperature change rate and temperature change acceleration within a short time window are calculated to evaluate the dynamic response characteristics of the temperature sensor. S4. The parameter estimation results of the physical thermodynamic model based on Newton's law of cooling are fused with the multidimensional feature regression prediction results based on the machine learning model to solve the corrected temperature measurement value in real time. S5. When the fluctuation rate of the continuously calculated corrected temperature measurement value is less than the preset fluctuation threshold, the corrected temperature measurement value is locked and output as the final target ambient temperature measurement result.

[0007] As a preferred solution, the specific process for locating the center point of the target environment is as follows: S21. When the temperature measuring terminal equipped with the temperature sensor enters the target environment, the relative distance between the temperature measuring terminal and the four walls of the target environment is measured by the omnidirectional lidar carried on the temperature measuring terminal, and the two-dimensional spatial data of the target environment is reconstructed. S22. Filter the reconstructed two-dimensional spatial data, identify and remove temporary obstacles in the target environment, and extract the boundary coordinates of the effective measurement space composed of solid walls. S23. Calculate the environmental center point Pc(xc, yc, zc) of the three-dimensional effective space, where the projection center of the four main walls is set as the horizontal and vertical coordinates (xc, yc) of the environmental center point Pc, and the zc of the core thermal sensing area of ​​human daily activities above the ground is set as the vertical coordinate of the environmental center point Pc. S24. Move the temperature measurement terminal to the coordinate point (xc, yc). The omnidirectional lidar continuously performs real-time positioning and mapping. When the Euclidean distance between the center point of the temperature measurement terminal and the center point of the environment Pc in three-dimensional space is less than the preset tolerance threshold, the process of locating the center of the environment is completed.

[0008] As a preferred solution, the specific process of S3 is as follows: The temperature time series data is smoothed and filtered to obtain the temperature continuous function T(t). Within a short time window, the temperature change rate v(t) and temperature change acceleration a(t) at the current moment are calculated in real time by the local polynomial differentiation method.

[0009] As a preferred approach, the specific process for obtaining the parameter estimation results of the physical thermodynamic model based on Newton's law of cooling is as follows: S41. Based on Newton's law of cooling, the internal temperature distribution of the temperature sensor is assumed to be uniform, and its temperature-time response equation is assumed to be... It is represented as follows; ; in, Initial temperatures under different initial operating conditions; The final temperature of the environment to be predicted. The system time constant of the target environment where the temperature sensor is located; S42. Calculate the essential parameters of the current dynamic response characteristics according to the following formula. ; ; ; Therefore, we get: ; S43, will Substitute back into the first derivative equation In this way, the final temperature of the environment can be obtained. The transient analytical estimate at time t; ; S44. Execute the following dynamic parameter estimation algorithm within a short time window [t0, tn]; By setting a sliding calculation window and constructing a weighted least squares objective function for the final temperature based on transient analytical estimates, a weighted least squares objective function is defined. ; ; in, These are preset weighting coefficients; The weighted least squares objective function is solved. As the number of effective sampling points in the sliding calculation window increases, the final temperature estimation result gradually stabilizes. The estimated value obtained is used as the first final state prediction value, that is, the physical thermodynamic model parameter estimation result.

[0010] As a preferred approach, the specific process for obtaining multidimensional feature regression prediction results based on machine learning models is as follows: Construct a multidimensional feature vector for input to a machine learning model, the multidimensional feature vector comprising: Boundary state characteristics: Obtain the initial temperature value and the transient temperature value at the current sampling time to define the basic temperature difference range of the current heat transfer process; Dynamic response characteristics are used to obtain the current temperature change rate and temperature change acceleration, which are used to reflect the real-time fluctuation of the environmental convective heat transfer coefficient and the decay trend of the sensor's thermal inertia. Temporal process characteristics: Obtain the actual time taken for the temperature sensor to travel from entering the target environment to the current sampling point, which is used to assist the machine learning model in recognizing the specific physical stage of the heat exchange process. An algorithm based on decision tree ensemble is used to fit the complex nonlinear mapping relationship between multidimensional feature vectors and environmental final temperature. A machine learning model that has been trained and stored in read-only memory is loaded, and new multidimensional feature vectors are continuously generated within a short time window and input into the machine learning model frame by frame. Through preset nonlinear regression rules and node branch judgment, the second final state prediction value is calculated and output in real time, which is the multidimensional feature regression prediction result based on machine learning.

[0011] As a preferred approach, the specific process for obtaining the final target ambient temperature measurement result is as follows: S51. During the continuous sampling process within a short time window, obtain the first final state prediction value estimated by the physical thermodynamic model based on Newton's law of cooling, and the second final state prediction value obtained by multidimensional feature regression prediction based on the machine learning model. S52. Compare the theoretical fitting curve of the current physical thermodynamic model with the actual collected temperature time series, extract the recent fitting residual, and at the same time extract the confidence interval width of the machine learning model output distribution. S53. Based on the evaluation results of the recent fitting residuals and confidence interval width, assign weight coefficients to the first final state prediction value and the second final state prediction value in real time, and output the corrected temperature measurement value under the current sampling period in real time through weighted comprehensive processing. S54. Construct a sliding statistical window, which is used to cache the corrected temperature measurement values ​​output from multiple consecutive sampling periods in the recent period, and calculate the dispersion index of all historical prediction values ​​within the sliding statistical window in real time. S55. When the predicted value volatility index within the sliding statistical window continues to decrease and is less than the preset volatility threshold within multiple consecutive time periods, it is determined that the temperature prediction model has fully converged. At this time, the locking mechanism is triggered to stop the physical temperature sampling of the target environment and cut off the high-energy-consuming prediction calculation process. The average predicted value within the sliding statistical window at the time of triggering the locking is obtained as the final target environment temperature measurement result.

[0012] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: This rapid indoor temperature measurement method based on thermal equilibrium final state prediction can infer the final state temperature by extracting transient data within an extremely short initial time window, reducing the traditional waiting time of more than ten minutes to within three minutes, thus significantly improving the efficiency of heating system inspections. The method incorporates a location-based environmental center point mechanism during measurement, effectively avoiding interference from local heat sources such as corners and windows, as well as boundary layer convection, ensuring standardization of measurement characteristics from the physical space source. By extracting smoothed temperature change rates and accelerations, it eliminates response differences caused by varying initial temperature differences. It utilizes physical laws to prevent algorithm divergence and uses models to correct environmental perturbation errors. Combined with an adaptive volatility locking mechanism, this not only ensures the accuracy of the final output but also effectively reduces the reactive power consumption of portable devices. Detailed Implementation

[0013] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0014] Example: A rapid indoor temperature measurement method based on thermal equilibrium final state prediction includes the following steps: S1. At the initial moment t0 when the temperature measuring terminal equipped with the temperature sensor enters the target environment, record the initial temperature value of the temperature sensor. S2. Place the temperature sensor in the target environment, locate the center point of the environment, and continuously collect temperature time series data within a short time window [t0, tn] at the center point of the environment at a set sampling frequency, where tn is less than the time required for the temperature sensor to reach physical thermal equilibrium. The specific process for locating the center point of the target environment is as follows: S21. When the temperature measuring terminal enters the target environment (such as a heated room in a residential building), the omnidirectional lidar carried on the temperature measuring terminal emits laser beams in all directions and receives reflected signals. Based on the time-of-flight principle, the relative distance between the temperature measuring terminal and the surrounding walls of the target environment is calculated, and the two-dimensional spatial data of the target environment is reconstructed. S22. Filter the reconstructed two-dimensional spatial data, identify and remove temporary obstacles in the target environment, and extract the boundary coordinates of the effective measurement space composed of solid walls. S23. Calculate the environmental center point Pc(xc, yc, zc) in the three-dimensional effective space using the centroid interpolation algorithm. Specifically: The projection centers of the four main walls are set as the horizontal and vertical coordinates (xc, yc) of the environmental center point Pc. The zc of the core thermal sensing area for daily human activities above the ground is set as the vertical coordinate of the environmental center point Pc. zc is generally an empirical value, such as 1.1m or 1.5m above the ground. S24. Move the temperature measuring terminal to the coordinate point (xc, yc). This can be done manually or automatically. For example, plan the optimal collision-free path from the current initial position of the temperature measuring terminal to the environmental center point Pc. The temperature measuring terminal with the driving structure will then move along the optimal collision-free path. During the movement, the omnidirectional lidar will continuously perform real-time positioning and mapping. When the temperature measuring terminal moves to the environmental center point Pc and the Euclidean distance between the center point of the temperature measuring terminal and the environmental center point Pc in three-dimensional space is less than the preset tolerance threshold, the temperature measuring terminal will stop moving, completing the process of locating the environmental center.

[0015] Furthermore, the temperature measuring terminal is equipped with an active airflow guiding component, which includes a miniature fan and an airflow guiding channel connected to the temperature sensing end of the temperature sensor. After the temperature measuring terminal enters the target environment, the miniature fan is controlled to start at a preset speed, causing the target environment air to flow along the airflow guiding channel past the temperature sensing end of the temperature sensor or the surface of the temperature measuring chip, thereby enhancing the convective heat transfer efficiency between the target environment air and the temperature sensor and shortening the response time of the temperature sensor from the initial temperature to the target environment temperature.

[0016] Through the aforementioned active airflow guidance method, the temperature sensor can acquire transient temperature time-series data that better reflects the temperature change trend of the target environment within a short time window. This reduces the response lag caused by sensor thermal inertia and provides a stable data foundation for subsequent rapid measurements based on temperature change rate, temperature change acceleration, and final-state temperature prediction models. Simultaneously, a heat insulation structure is installed between the micro-fan and the temperature sensor, and the speed of the micro-fan is limited to a preset range to reduce the impact of fan motor heating and local airflow disturbances on the temperature sampling results.

[0017] S3. The temperature time series data is smoothed and filtered, and the temperature change rate and temperature change acceleration within a short time window are calculated to evaluate the dynamic response characteristics of the temperature sensor under different initial operating conditions. The specific process of S3 is as follows: The temperature time series data is smoothed and filtered to obtain a continuous temperature function T(t). Within a short time window, the temperature change rate v(t) and temperature change acceleration a(t) are calculated in real time using the local polynomial differentiation method (local polynomial fitting) or Savitzky-Golay, where: Temperature change rate v(t): the first derivative of temperature with respect to time t. , is used to represent the intensity of heat exchange between the current temperature sensor and the environment. The absolute value of the temperature change rate is proportional to the temperature difference between the current temperature sensor and the final temperature of the environment, or approximately related under a preset first-order thermal response model. Temperature-dependent acceleration a(t): the second derivative of temperature with respect to time t. This is used to indicate the rate of decay of the heat exchange driving force, reflecting the dynamic changes in the overall heat capacity of the temperature sensor and the convective heat transfer coefficient of the environment.

[0018] S4. Construct a temperature prediction model based on a dual-path final state temperature prediction architecture, and fuse the parameter estimation results of the physical thermodynamic model based on Newton's law of cooling with the multidimensional feature regression prediction results based on the machine learning model to solve the corrected temperature measurement value in real time. The specific process for obtaining the parameter estimation results of the physical thermodynamic model based on Newton's law of cooling is as follows: S41. The transient heat transfer process of the temperature sensor after entering the target environment conforms to Newton's law of cooling. Based on Newton's law of cooling, the internal temperature distribution of the temperature sensor is assumed to be uniform, and its temperature-time response equation is... It is expressed as follows: ; in, Initial temperatures under different initial operating conditions; The final temperature of the environment to be predicted. The system time constant of the target environment where the temperature sensor is located; S42. Calculate the essential parameters of the current dynamic response characteristics according to the following formula. : ; ; Therefore, we get: ; S43, will Substitute back into the first derivative equation In this way, the final temperature of the environment can be obtained. The transient analytical estimate at time t: ; S44. Execute the following dynamic parameter estimation algorithm within a short time window [t0, tn]: Define a sliding computation window (e.g., containing k consecutive sampling points), and construct a weighted least squares objective function for the final temperature based on the transient analytical estimate. : ; in, For The weighting coefficients corresponding to the sampling points are based on the weighting coefficients of the first sampling point. At least one of the following is determined for each sampling point: the absolute value of the temperature change rate, the stability of the temperature change acceleration, and the recent fitting residual; wherein, the recent fitting residual is the residual between the actual sampled temperature value and the fitted temperature value of the physical thermodynamic model within the sliding calculation window. When the temperature change acceleration is stable and the recent fitting residual does not exceed a preset range, the larger the absolute value of the temperature change rate, the larger the corresponding weighting coefficient.

[0019] The weighted least squares objective function is minimized. As the number of effective sampling points in the sliding calculation window increases, the final temperature estimation result gradually stabilizes. The estimated value obtained is used as the first final state prediction value, that is, the physical thermodynamic model parameter estimation result.

[0020] The specific process for obtaining multidimensional feature regression prediction results based on machine learning models is as follows: Construct a multidimensional feature vector for input to a machine learning model, the multidimensional feature vector comprising: Boundary state characteristics: Obtain the initial temperature value and the transient temperature value at the current sampling time to define the basic temperature difference range of the current heat transfer process; Dynamic response characteristics are used to obtain the current temperature change rate and temperature change acceleration, which are used to reflect the real-time fluctuation of the environmental convective heat transfer coefficient and the decay trend of the sensor's thermal inertia. Timing process characteristics: Obtain the actual time taken for the temperature sensor to travel from entering the target environment to the current sampling point, which is used to help the machine learning model recognize the specific physical stage of the heat exchange process (such as the initial intense heat exchange period or the middle smooth transition period). Before feeding the multidimensional feature vectors into the model, normalization or standardization is performed to map the values ​​of all dimensions to a unified data range. During the equipment's research and development and manufacturing phases, the system completed the offline training process for the model. The large-scale dataset used for training contained multi-dimensional feature data sequences of the temperature measuring terminal within a short time window under hundreds of different initial operating conditions (such as entering the target indoor environment from extremely cold outdoor conditions, normal temperature corridor conditions, and high temperature vehicle conditions). Simultaneously, the label data in the training set represented the actual physical thermal equilibrium temperature reached by the sensor after a prolonged period of placement. By minimizing the regression error between the predicted output and the true label, the model fully learned and mastered the implicit relationship between short-term dynamic characteristics and the final steady-state temperature.

[0021] An algorithm based on decision tree ensemble is used to fit the complex nonlinear mapping relationship between multidimensional feature vectors and environmental final temperature. A machine learning model that has been trained and stored in read-only memory is loaded, and new multidimensional feature vectors are continuously generated within a short time window and input into the machine learning model frame by frame. Through preset nonlinear regression rules and node branch judgment, the second final state prediction value is calculated and output in real time, which is the multidimensional feature regression prediction result based on machine learning.

[0022] S5. When the fluctuation rate of the continuously calculated corrected temperature measurement value is less than the preset fluctuation threshold, the corrected temperature measurement value is locked and output as the final target ambient temperature measurement result.

[0023] The specific process for obtaining the final target ambient temperature measurement result is as follows: S51. During the continuous sampling process within a short time window, obtain the first final state prediction value estimated by the physical thermodynamic model based on Newton's law of cooling, and the second final state prediction value obtained by multidimensional feature regression prediction based on the machine learning model. S52. Compare the theoretical fitting curve of the current physical thermodynamic model with the actual collected temperature time series, extract the recent fitting residual, and at the same time extract the confidence interval width of the machine learning model output distribution. The smaller the fitting residual and the narrower the confidence interval, the better the prediction performance of the corresponding path under the current specific working conditions.

[0024] S53. Based on the evaluation results of the recent fitting residuals and confidence interval width, assign weight coefficients to the first final state prediction value and the second final state prediction value in real time. Multiply the first final state prediction value with its corresponding dynamic weight, and add the product of the second final state prediction value with its dynamic weight. Through weighted comprehensive processing, output the corrected temperature measurement value under the current sampling period in real time. The specific principle is as follows: assign a larger weight to the prediction path with the smaller prediction error and higher confidence; conversely, reduce the weight of the other path. Furthermore, the sum of the weight coefficients of the two paths is always maintained as the overall benchmark.

[0025] S54. Construct a sliding statistical window, which is used to cache the corrected temperature measurement values ​​output from multiple consecutive sampling periods in the recent period. After each new corrected temperature measurement value enters the sliding statistical window, the dispersion index of all historical prediction values ​​in the sliding statistical window is calculated in real time. The dispersion index is characterized by the difference between the maximum and minimum predicted values ​​within the sliding statistical window, reflecting whether the prediction algorithm is currently in a convergent state. S55. As sampling proceeds, when the predicted value volatility index within the sliding statistical window continues to decrease and is less than the preset volatility threshold within multiple consecutive time periods, it is determined that the temperature prediction model has fully converged. At this point, a locking mechanism is triggered to stop sampling the physical temperature of the target environment and cut off the energy-intensive prediction calculation process. The average predicted value within the sliding statistical window at the moment of triggering the locking is obtained as the final target environment temperature measurement result and reported to the heating management platform and the user's mobile terminal corresponding to the target environment through the wireless communication module.

[0026] This invention can infer the final temperature by capturing transient data within an initial, extremely short time window, reducing the traditional waiting time of more than ten minutes to within three minutes, thus significantly improving the efficiency of heating system inspections. The invention introduces a mechanism to locate the center point of the environment during measurement, effectively avoiding interference from local heat sources such as corners and windows, as well as boundary layer convection. This ensures the standardization of measurement characteristics from the physical space source. By extracting the smoothed temperature change rate and acceleration, it eliminates response differences caused by varying initial temperature differences. It utilizes physical laws to prevent algorithm divergence and uses a model to correct environmental perturbation errors. Combined with an adaptive volatility locking mechanism, this not only ensures the accuracy of the final output but also effectively reduces the reactive power consumption of portable devices.

[0027] The setting of the interval and threshold size is for the purpose of facilitating comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value.

[0028] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation. The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A rapid indoor temperature measurement method based on thermal equilibrium final state prediction, characterized in that, Includes the following steps: S1. Record the initial temperature value at the moment the temperature sensor enters the target environment; S2. Move the temperature sensor to its location at the center point of the environment, and continuously collect temperature time series data within a short time window [t0, tn] at the center point of the environment at a set sampling frequency, where tn is less than the time required for the temperature sensor to reach physical thermal equilibrium; wherein, the temperature measuring terminal equipped with the temperature sensor is provided with an active airflow guiding component, the active airflow guiding component including a miniature fan and an airflow guiding channel connected to the temperature sensing end of the temperature sensor; after the temperature measuring terminal enters the target environment, control the miniature fan to start at a preset speed, so that the target environment air flows along the airflow guiding channel through the temperature sensing end of the temperature sensor; a heat insulation structure is provided between the miniature fan and the temperature sensor, and the speed of the miniature fan is limited to a preset range; S3. The temperature time series data is smoothed and filtered, and the temperature change rate and temperature change acceleration within a short time window are calculated to evaluate the dynamic response characteristics of the temperature sensor. S4. The parameter estimation results of the physical thermodynamic model based on Newton's law of cooling are fused with the multidimensional feature regression prediction results based on the machine learning model to solve for the corrected temperature measurement value in real time. Specifically, during continuous sampling within a short time window, the first final state prediction value estimated by the physical thermodynamic model based on Newton's law of cooling and the second final state prediction value predicted by the multidimensional feature regression based on the machine learning model are obtained. The theoretical fitting curve of the current physical thermodynamic model is compared with the actual collected temperature time series to extract the recent fitting residual, and the confidence interval width of the machine learning model output distribution is also extracted. Based on the evaluation results of the recent fitting residual and the confidence interval width, weight coefficients are assigned to the first and second final state prediction values ​​in real time. Through weighted comprehensive processing, the corrected temperature measurement value under the current sampling period is output in real time. S5. Construct a sliding statistical window. The sliding statistical window is used to cache the corrected temperature measurement values ​​output from multiple consecutive sampling periods in the recent period and to calculate the dispersion index of all historical prediction values ​​in the sliding statistical window in real time. When the fluctuation index of the prediction value in the sliding statistical window continues to decrease and is less than the preset fluctuation threshold in multiple consecutive time periods, it is determined that the temperature prediction model has fully converged. At this time, the locking mechanism is triggered to stop the physical temperature sampling of the target environment and cut off the high-energy-consuming prediction calculation process. The average prediction value in the sliding statistical window at the time of triggering the locking is obtained as the final target environment temperature measurement result.

2. The rapid indoor temperature measurement method based on thermal equilibrium final state prediction according to claim 1, characterized in that, The specific process for locating the center point of the target environment is as follows: S21. When the temperature measuring terminal equipped with the temperature sensor enters the target environment, the relative distance between the temperature measuring terminal and the four walls of the target environment is measured by the omnidirectional lidar carried on the temperature measuring terminal, and the two-dimensional spatial data of the target environment is reconstructed. S22. Filter the reconstructed two-dimensional spatial data, identify and remove temporary obstacles in the target environment, and extract the boundary coordinates of the effective measurement space composed of solid walls. S23. Calculate the environmental center point Pc(xc, yc, zc) of the three-dimensional effective space, where the projection center of the four main walls is set as the horizontal and vertical coordinates (xc, yc) of the environmental center point Pc, and the zc of the core thermal sensing area of ​​human daily activities above the ground is set as the vertical coordinate of the environmental center point Pc. S24. Move the temperature measurement terminal to the coordinate point (xc, yc). The omnidirectional lidar continuously performs real-time positioning and mapping. When the Euclidean distance between the center point of the temperature measurement terminal and the center point of the environment Pc in three-dimensional space is less than the preset tolerance threshold, the process of locating the center of the environment is completed.

3. The rapid indoor temperature measurement method based on thermal equilibrium final state prediction according to claim 1, characterized in that, The specific process of S3 is as follows: The temperature time series data is smoothed and filtered to obtain the temperature continuous function T(t). Within a short time window, the temperature change rate v(t) and temperature change acceleration a(t) at the current moment are calculated in real time by the local polynomial differentiation method.

4. The rapid indoor temperature measurement method based on thermal equilibrium final state prediction according to claim 1, characterized in that, The specific process for obtaining the parameter estimation results of the physical thermodynamic model based on Newton's law of cooling is as follows: S41. Based on Newton's law of cooling, the internal temperature distribution of the temperature sensor is assumed to be uniform, and its temperature-time response equation is assumed to be... It is represented as follows; ； in, Initial temperatures under different initial operating conditions; The final temperature of the environment to be predicted. The system time constant of the target environment where the temperature sensor is located; t For time; S42. Calculate the essential parameters of the current dynamic response characteristics according to the following formula. ; ； ； Therefore, we get: ; S43, will Substitute back into the first derivative equation In this way, the final temperature of the environment can be obtained. exist t The transient analytical estimate at time t; ； S44. Execute the following dynamic parameter estimation algorithm within a short time window [t0, tn]; By setting a sliding computation window and constructing a weighted least squares objective function for the final temperature based on transient analytical estimates, a weighted least squares objective function is defined. ; ； in, For the preset first The weight coefficients corresponding to each sampling point The number of consecutive samples. For time The rate of change of temperature, For time Changing temperature acceleration, For the first The time for each sampling point; The weighted least squares objective function is solved. As the number of effective sampling points within the sliding calculation window increases, the final state temperature estimation result gradually stabilizes. The estimated value obtained is used as the first final state prediction value, i.e., the physical thermodynamic model parameter estimation result. The recent fitting residual is the residual between the actual sampled temperature value within the sliding calculation window and the fitted temperature value of the physical thermodynamic model. When the temperature change acceleration is stable and the recent fitting residual does not exceed the preset range, the larger the absolute value of the temperature change rate, the larger the corresponding weight coefficient.

5. The rapid indoor temperature measurement method based on thermal equilibrium final state prediction according to claim 4, characterized in that, The specific process for obtaining multidimensional feature regression prediction results based on machine learning models is as follows: Construct a multidimensional feature vector for input to a machine learning model, the multidimensional feature vector comprising: Boundary state characteristics: Obtain the initial temperature value and the transient temperature value at the current sampling time to define the basic temperature difference range of the current heat transfer process; Dynamic response characteristics are used to obtain the current temperature change rate and temperature change acceleration, which are used to reflect the real-time fluctuation of the environmental convective heat transfer coefficient and the decay trend of the sensor's thermal inertia. Temporal process characteristics: Obtain the actual time taken for the temperature sensor to travel from entering the target environment to the current sampling point, which is used to assist the machine learning model in recognizing the specific physical stage of the heat exchange process. An algorithm based on decision tree ensemble is used to fit the complex nonlinear mapping relationship between multidimensional feature vectors and environmental final temperature. A machine learning model that has been trained and stored in read-only memory is loaded, and new multidimensional feature vectors are continuously generated within a short time window and input into the machine learning model frame by frame. Through preset nonlinear regression rules and node branch judgment, the second final state prediction value is calculated and output in real time, which is the multidimensional feature regression prediction result based on machine learning.

Images