High-precision sampling processing method based on multi-model temperature sensors

Abstract

The invention relates to the field of temperature acquisition systems, and discloses a high-precision sampling processing method based on multi-model temperature sensors, and the method comprises the steps: obtaining an original ADC data block of a temperature sensor, and extracting a dynamic feature fingerprint; analyzing the dynamic feature fingerprints to determine a model probability vector characterizing a known working mode and a novelty score characterizing an unknown working mode; according to the model probability vector, the estimation results of the multiple extended Kalman filters are fused through an interactive multi-model algorithm, a temperature value is obtained, and a residual sequence is generated; and updating the system model online according to the novelty score or the residual sequence. The system comprises a data acquisition and feature extraction module, a sensor state sensing module, a self-adaptive temperature estimation module and a system online evolution module. According to the method, a closed-loop processing flow with state sensing and online evolution capabilities is constructed, so that the problem that the precision of a traditional fixed model is reduced under a complex working condition is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of temperature acquisition systems, specifically a high-precision sampling processing method based on multiple types of temperature sensors. BACKGROUND

[0002] Currently, temperature sensors play a fundamental role in industrial control, environmental monitoring, and other fields. Among them, thermal resistors (such as PT100, PT1000) are based on the characteristic that resistance value changes with temperature, suitable for high-precision, low-temperature detection; thermocouples (such as K-type) are based on the characteristic that thermoelectric electromotive force changes with temperature, which can achieve wider range of temperature detection.

[0003] However, existing technologies face multiple challenges in processing high-precision sampling of multiple types of temperature sensors. Traditional temperature acquisition systems often rely on fixed algorithms and lookup table methods, lacking adaptive ability to different types of sensors and their dynamic changes under complex working conditions. For example, when the sensor itself has performance drift, is subject to electromagnetic interference, or has poor contact, etc. abnormal conditions, the processing method based on static model is difficult to accurately distinguish between real temperature changes and measurement deviations caused by external interference, resulting in significant decrease in measurement accuracy and reliability.

[0004] In addition, existing systems are usually optimized for specific types or specific temperature ranges, and it is difficult to flexibly compatible with multiple types of sensors and maintain high precision in a wide temperature range.

[0005] Therefore, the present application proposes a high-precision sampling processing method based on multiple types of temperature sensors to solve the deficiencies of the prior art. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a high-precision sampling processing method based on multiple types of temperature sensors, which solves the problem that the processing method of temperature sensors usually uses fixed calibration model and filtering algorithm. When the sensor has performance drift or encounters complex working conditions such as unexpected electromagnetic interference, the adaptability of the processing model decreases, resulting in reduced accuracy and reliability of temperature sampling.

[0007] To solve the above technical problems, the present application provides a new technical solution.

[0008] The first aspect of the present application provides a high-precision sampling processing method based on multiple types of temperature sensors, which dynamically deals with the state changes of the sensor throughout its life cycle through a closed-loop processing flow integrated with state awareness, adaptive estimation and online evolution ability, to ensure the continuous accuracy of temperature measurement results.

[0009] The method first acquires a raw ADC data block from a temperature sensor. This raw ADC data block is a continuous sequence of ADC samples. To extract dynamic information representing the sensor's current operating state, the method applies a window function to the raw ADC data block and then performs a fast Fourier transform on the windowed data to obtain a frequency domain representation. From this frequency domain representation, the energy distribution and spectral peak characteristics of key frequency bands are extracted to form a multidimensional dynamic feature fingerprint.

[0010] After obtaining the dynamic feature fingerprint, the method performs two analyses on it in parallel.

[0011] The first analysis is to match the dynamic feature fingerprint with multiple reference fingerprint templates in a preset dynamic feature fingerprint template library to calculate a model probability vector, which represents the similarity between the current state and each known working mode.

[0012] The second analysis is to input the dynamic feature fingerprint into a preset generative anomaly detection model, and obtain a novelty score by calculating the reconstruction error of the model. The novelty score is used to quantify the degree to which the current state deviates from all known working modes.

[0013] The generative anomaly detection model includes an encoder and a decoder. The novelty score is calculated using the dynamic feature fingerprint as input, and its calculation formula is:

[0014] S n =||v f -p θ (q φ (v f ))|| 2 ;

[0015] Where S n Score for novelty; v f is the dynamic feature fingerprint; q φ is the encoder of the generative anomaly detection model; p θ Decoder for the generative anomaly detection model.

[0016] Next, the method initiates an interactive multi-model algorithm based on the calculated model probability vector. This algorithm dynamically performs a weighted fusion of the estimation results of multiple extended Kalman filters from a state-space model library. Each extended Kalman filter state vector includes a temperature value and a drift parameter that characterizes the variation in the electrical characteristics of the temperature sensor itself. The method extracts the temperature value from the fused state vector as the current optimal estimate and simultaneously generates a residual sequence for the interactive multi-model algorithm.

[0017] Finally, the method updates the state space model library online according to the novelty score or the residual sequence, and outputs the final temperature value. There are two paths for the online update:

[0018] Path one: when the novelty score is higher than a preset threshold, the method triggers an active detection process, injects a disturbance signal into the excitation source of the temperature sensor or increases the ADC sampling rate, to capture enhanced data of unknown working modes. Then, unsupervised clustering is performed on the captured enhanced data to identify the modes of the unknown working modes, and a new candidate model is automatically synthesized according to the identified modes to supplement the state space model library.

[0019] Path two: when the interactive multiple model algorithm determines that the temperature sensor is stably running in a known working mode, the method uses the generated residual sequence as the basis, and uses the statistical characteristics of the residual sequence to update the noise covariance matrix of the extended Kalman filter corresponding to the known working mode in the state space model library, and update the corresponding reference fingerprint template in the dynamic feature fingerprint template library.

[0020] The second aspect of the application provides a high-precision sampling processing system based on multiple temperature sensors, which is applied to the method described above. The system comprises:

[0021] A data acquisition and feature extraction module is configured to obtain an original ADC data block of the temperature sensor, and extract a dynamic feature fingerprint from the original ADC data block.

[0022] A sensor state perception module is configured to analyze the dynamic feature fingerprint extracted by the data acquisition and feature extraction module to determine a model probability vector and a novelty score.

[0023] An adaptive temperature estimation module is configured to determine a temperature value from a state space model library by fusing estimation results of multiple extended Kalman filters through an interactive multiple model algorithm according to the model probability vector determined by the sensor state perception module, and generate a residual sequence of the interactive multiple model algorithm.

[0024] A system online evolution module is configured to update the state space model library online according to the novelty score determined by the sensor state perception module or the residual sequence generated by the adaptive temperature estimation module.

[0025] A temperature value output module is configured to output the temperature value obtained by the adaptive temperature estimation module.

[0026] The application provides a high-precision sampling processing method based on multiple temperature sensors. The method has the following advantages:

[0027] 1、The present application realizes the accurate perception of the current running state of the temperature sensor by extracting the dynamic characteristic fingerprint of the original ADC data block. It can distinguish between real temperature changes and data fluctuations caused by abnormal internal and external working conditions of the sensor, avoiding the misjudgment of state abnormalities as temperature changes, thereby improving the reliability and robustness of high-precision sampling processing in complex electromagnetic environments or when the sensor itself has performance fluctuations.

[0028] 2、The present application uses an interactive multi-model algorithm to dynamically fuse multiple extended Kalman filters, replacing the fixed filtering or lookup table model in the traditional method. By adaptively adjusting and fusing the optimal estimation result according to the real-time perceived model probability vector, the system can smoothly switch between multiple known working modes, and compared with any single static model, it can guarantee the accuracy of temperature value processing in a wider working condition range.

[0029] 3、The present application introduces the calculation and online evolution mechanism of novelty score, enabling the system to have the ability of autonomous learning and expansion of unknown working modes. When the system identifies a never-before-seen sensor state, it can actively trigger detection and model synthesis to include the new mode into its state space model library. This design makes the system not limited to the pre-set working conditions at the factory, has the potential to cope with future unknown disturbances or new types of sensor access, improves the applicability and life cycle of the system.

[0030] 4、The present application uses the residual sequence generated by the interactive multi-model algorithm in the stable running state to refine the state space model library online. It can continuously and subtly correct the model parameters to compensate for the aging or slow performance drift caused by long-term use of the sensor. This self-correcting ability ensures that the high-precision characteristics of the system do not significantly decrease over time, realizing high-precision tracking of the temperature sensor throughout its life cycle. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 is the system function module diagram of the present application;

[0032] Figure 2 is the method flowchart of the present application;

[0033] Figure 3 is the detailed flowchart of the online evolution step of the present application;

[0034] Figure 4 is the interactive diagram of the interactive multi-model algorithm and the state space model library of the present application.

[0035] Wherein, 10, data acquisition and feature extraction module; 20, sensor state perception module; 30, adaptive temperature estimation module; 40, system online evolution module; 50, temperature value output module; 60, state space model library; 70, dynamic feature fingerprint template library. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the specification of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0037] Referring to Figure 1 One embodiment of the present application provides a high-precision sampling processing system based on multiple types of temperature sensors. The system can be deployed in a computing platform based on a Phytium processor and a Kylin operating system, and interacts with temperature sensor hardware through a driver interface. The system includes a data acquisition and feature extraction module 10, a sensor state perception module 20, an adaptive temperature estimation module 30, a system online evolution module 40, and a temperature value output module 50.

[0038] The data acquisition and feature extraction module 10 directly communicates with the temperature sensor hardware, and the sensor state perception module 20 is connected to the output end of the data acquisition and feature extraction module 10. The input end of the adaptive temperature estimation module 30 is connected to the output end of the sensor state perception module 20 and a state space model library 60. The input end of the system online evolution module 40 is connected to the output end of the sensor state perception module 20 and the output end of the adaptive temperature estimation module 30, and its output end is connected to the state space model library 60 and a dynamic feature fingerprint template library 70. The temperature value output module 50 is connected to the output end of the adaptive temperature estimation module 30.

[0039] Referring to Figure 2 Corresponding to the above system, the present application provides a high-precision sampling processing method based on multiple types of temperature sensors. The method dynamically deals with the state changes of the sensor in the whole life cycle through a closed-loop processing flow. The method can include the following steps:

[0040] S1, the data acquisition and feature extraction module 10 acquires the original ADC data block of the temperature sensor, and extracts the dynamic feature fingerprint representing the current running state of the temperature sensor according to the original ADC data block.

[0041] S2, the sensor state awareness module 20 analyzes the dynamic feature fingerprint extracted in S1 to determine a model probability vector representing the likelihood of the temperature sensor being in each of a plurality of known operating modes, and to determine a novelty score representing the likelihood of the temperature sensor being in an unknown operating mode.

[0042] S3, the adaptive temperature estimation module 30 obtains the temperature value of the temperature sensor from the state space model library 60 according to the model probability vector determined in S2, fuses the estimation results of a plurality of extended Kalman filters through an interacting multiple model algorithm, and generates a residual sequence of the interacting multiple model algorithm.

[0043] S4, the system online evolution module 40 performs online update on the state space model library 60 and the dynamic feature fingerprint template library 70 according to the novelty score determined in S2 or the residual sequence generated in S3, and the temperature value output module 50 outputs the temperature value obtained in S3.

[0044] Reference Figure 1 and Figure 2 The following will describe step S1 in the method flow of the present application in detail, which is performed by the data acquisition and feature extraction module 10.

[0045] At the beginning of the method, an initial step of interacting with hardware is first performed. The data acquisition and feature extraction module 10 identifies the currently accessed temperature sensor model by reading the value of a preset hardware register (register address 0x18 in this embodiment). For example, the value 0x01 represents a PT100 thermal resistance, 0x02 represents a PT1000 thermal resistance, and 0x03 represents a K-type thermocouple. Meanwhile, the module sets the specific physical channel for subsequent sampling according to the parameters passed by the upper application program through a standard interface function such as ioctl.

[0046] After determining the sensor model and the sampling channel, the data acquisition and feature extraction module 10 writes a specific value (for example, 0x55aa) to a control register (register address 0x2054 in this embodiment) to start the analog-to-digital converter (ADC) for high-speed continuous sampling. After the ADC completes the conversion, an interrupt signal is generated. Unlike the prior art which only reads a few sampling points to obtain the average value, the method of the present application continuously reads a high-speed ADC sampling sequence from a specified memory address (for example, the first address 0x04) after receiving the interrupt signal, to form an original ADC data block. The data block retains the complete dynamic information of the signal in the time sequence.

[0047] After obtaining the raw ADC data block, the data acquisition and feature extraction module 10 performs a series of processing to extract a multi-dimensional dynamic feature fingerprint from it. The process first applies a window function, such as Hamming window or Hanning window, to the raw ADC data block. The purpose of applying a window function is to smooth the two ends of the data block, so as to reduce the spectral leakage effect when performing Fourier transform later.

[0048] Next, the raw ADC data block after applying the window function is subjected to Fast Fourier Transform (FFT) to convert it from time domain representation to frequency domain representation. The frequency domain representation reveals the amplitude and phase information of each frequency component contained in the original signal.

[0049] Finally, a set of pre-defined spectral features are extracted from the frequency domain representation to form the multi-dimensional dynamic feature fingerprint. The extracted features include:

[0050] The energy distribution in multiple critical frequency bands, such as the energy values of power frequency (50Hz / 60Hz) and its harmonic frequency bands, the total energy of high frequency noise bands;

[0051] and the main spectral peak features in the spectrum, such as the center frequency, amplitude, width and number of spectral peaks.

[0052] These quantified features collectively form a numerical vector, i.e. the dynamic feature fingerprint, which can comprehensively represent the comprehensive running state of the current temperature sensor system under the electrical and physical environment.

[0053] Referring to Figure 1 and Figure 2 , the step S2 in the method flow of the present application will be described in detail below, which is performed by the sensor state perception module 20. The sensor state perception module 20 receives the dynamic feature fingerprint generated by the data acquisition and feature extraction module 10, and performs two parallel analysis processes on it.

[0054] The first analysis is known working mode matching. The sensor state perception module 20 matches the received dynamic feature fingerprint with a plurality of reference fingerprint templates in a pre-defined dynamic feature fingerprint template library 70. The dynamic feature fingerprint template library 70 pre-stores reference fingerprint templates corresponding to multiple known working modes, such as standard fingerprints of the temperature sensor under different states, such as normal working, suffering from electromagnetic interference of a specific frequency, or contact failure. By calculating the similarity (e.g. by calculating the Euclidean distance or cosine similarity) between the current dynamic feature fingerprint and each reference fingerprint template in the library, and normalizing all similarity results (e.g. by Softmax function), a model probability vector is finally calculated. Each element in the model probability vector corresponds to the probability of the current sensor state being a certain known working mode.

[0055] The second analysis is unknown working mode detection. The sensor state perception module 20 inputs the same dynamic feature fingerprint into a preset, trained generative anomaly detection model. In this embodiment, the model is an autoencoder, which includes an encoder and a decoder. Based on the data distribution of all known working modes learned by the model during the training phase, the model attempts to reconstruct the input dynamic feature fingerprint.

[0056] The novelty score is calculated according to the reconstruction error of the input dynamic feature fingerprint by the generative anomaly detection model. A higher reconstruction error value indicates that the input fingerprint deviates from the data distribution of all known modes learned by the model, thereby representing the possibility of the sensor being in an unknown working mode. The calculation formula of the novelty score is:

[0057] S n =||v f -p θ (q φ (v f ))|| 2 ;

[0058] In the formula, S n is the novelty score; v f is the dynamic feature fingerprint; q φ is the encoder of the generative anomaly detection model; and p θ is the decoder of the generative anomaly detection model.

[0059] After completing the above two parallel analysis processes, the sensor state perception module 20 outputs the calculated model probability vector and novelty score to the adaptive temperature estimation module 30 and the system online evolution module 40, which are used for subsequent temperature estimation and system model updating.

[0060] Referring to Figure 1 , Figure 2 and Figure 4 , the step S3 in the method flow of the present application will be described in detail below. The step is performed by the adaptive temperature estimation module 30. The module receives the model probability vector delivered by the sensor state perception module 20, and performs adaptive temperature estimation based on the vector.

[0061] The core of the adaptive temperature estimation module 30 is an interactive multiple model (IMM) algorithm. The algorithm schedules multiple parallel extended Kalman filters (EKF) from a state space model library 60. The state space model library 60 stores multiple independent EKF models, each of which corresponds to a known sensor working mode, such as a normal working mode, a specific electromagnetic interference mode, etc.

[0062] Each extended Kalman filter is based on a nonlinear state-space model that describes the dynamic behavior of the sensor system in a particular operating mode. In one embodiment, the state vector of each EKF model includes two components: the temperature value of the object under test, and a drift parameter that characterizes changes in the electrical characteristics of the sensor due to aging or environmental changes.

[0063] Each EKF model describes the system through a state transition equation and an observation equation. At discrete time point k, the state-space representation is:

[0064] State transition equation:

[0065] x k = f(x k-1 ) + w k-1 ;

[0066] Observation equation:

[0067] z k = h(x k ) + v k ;

[0068] where k is the discrete time step; x k is the state vector at time point k, x k = [T k , d k ] T , where T k is the temperature value and d k is the drift parameter; f(·) is a nonlinear state transition function that describes the evolution of the state vector from time point k-1 to k; w k-1 is a process noise vector that is Gaussian distributed with zero mean and covariance matrix Q, representing the uncertainty of the model; z k is the observation vector at time point k, corresponding to the preliminary processed measurement values obtained from the ADC; h(·) is a nonlinear observation function that describes the relationship between the state vector x k and the observation vector z k ; v k is an observation noise vector that is Gaussian distributed with zero mean and covariance matrix R, representing the error of the measurement.

[0069] The IMM algorithm fuses the state estimation results of all the EKF models in the state space model library 60 in a processing cycle, using the model probability vector provided by the sensor state awareness module 20 as the weight. This fusion process generates an optimal combined state estimation. The adaptive temperature estimation module 30 extracts the temperature value component from the combined state estimation as the final output temperature value at the current time. At the same time, the module also calculates the difference between the current observation value and the fused predicted observation value to form the residual sequence of the IMM algorithm. The residual sequence and the final temperature value are output together, where the residual sequence will be passed to the system online evolution module 40.

[0070] With reference to Figure 1 , Figure 2 and Figure 3 , the step S4 in the method flow of the present application will be described in detail below. This step is executed by the system online evolution module 40. The module updates the system model online according to the novelty score determined by the sensor state awareness module 20, or the residual sequence generated by the adaptive temperature estimation module 30. This online update mechanism is implemented through two parallel paths.

[0071] The first evolution path is the learning and supplement of unknown operating modes. With reference to Path One in Figure 3 , when the novelty score S n continues to be higher than a preset threshold, the system online evolution module 40 determines that the current sensor may be in an unknown, unmodeled operating mode. At this time, the module triggers an active detection process. This process actively stimulates the sensor system by injecting a preset disturbance signal into the excitation source of the temperature sensor, or instantaneously increasing the sampling rate of the ADC, etc., to capture enhanced data that can fully represent the unknown operating mode.

[0072] After obtaining the enhanced data, the system online evolution module 40 performs unsupervised clustering analysis (e.g., using algorithms such as DBSCAN or Gaussian Mixture Model) on the dynamic characteristic fingerprint set corresponding to the enhanced data to identify the data distribution pattern of the unknown operating mode. According to the identified pattern, the module automatically synthesizes a new candidate model, which includes the state space equation parameters of a new EKF and a reference fingerprint template representing the new pattern. Finally, this new EKF model is supplemented to the state space model library 60, and the new reference fingerprint template is supplemented to the dynamic characteristic fingerprint template library 70. Through this path, the system realizes autonomous learning of new operating modes and expansion of the model library.

[0073] The second evolution path is the online refinement of known operating modes. With reference to Figure 3Path two in FIG. 1, when the interacting multiple model algorithm in adaptive temperature estimation module 30 judges that the temperature sensor is running stably in a certain known working mode (i.e. the probability value corresponding to the mode in the model probability vector is continuously close to 1), the online evolution module 40 starts the online refinement process. This process is based on the residual sequence generated by the adaptive temperature estimation module 30.

[0074] This module calculates the sample covariance of the residual sequence in a sliding time window. In theory, in an exactly matched model, the residual sequence should be zero-mean white noise. When the actually calculated sample covariance deviates from the theoretical value preset in the EKF model, it indicates that there is a mismatch in the model parameters (especially the process noise covariance matrix). The system online evolution module 40 uses this deviation to update the noise covariance matrix of the EKF model corresponding to the known working mode in the state space model library 60. A recursive formula for updating the process noise covariance matrix can be represented as:

[0075]

[0076] In the formula, Q k is the updated process noise covariance matrix at time point k; Q k-1 is the process noise covariance matrix before updating; δ k is a learning rate or forgetting factor, used to adjust the amplitude of updating; K k is the Kalman gain matrix of the EKF calculated at time point k; is the sample covariance matrix of the residual sequence calculated in a time window around time point k.

[0077] At the same time, the system online evolution module 40 collects the dynamic feature fingerprints generated by the data acquisition and feature extraction module 10 during this stable running period, and uses these fingerprints to update (for example, by calculating a moving average) the reference fingerprint template corresponding to the known working mode in the dynamic feature fingerprint template library 70, to ensure that the template can accurately reflect the features of the sensor in the current state, and compensate for the slow drift caused by factors such as sensor aging.

[0078] In a specific embodiment of the present application, the state space model library 60 is initialized according to the known sensor model (such as PT100, PT1000, K-type thermocouple) and its physical or empirical characteristics under normal working conditions when the system is deployed, to construct a set of basic extended Kalman filter models.

[0079] When the system identifies the accessed sensor as PT100, a normal working EKF model specially designed for it will be loaded in the state space model library 60. In this model, the construction of its observation function h(·) is based on the calibration table defined in IEC60751 international standard. Specifically, the raw measurement value of the ADC is first converted to resistance value through a pre-set linear fitting equation y = kx + b (where k and b are correction coefficients obtained from multi-point calibration). Subsequently, the non-linear relationship between the resistance value and temperature value (defined by the calibration table) is used as the core to construct the observation function h(·). This basic model describes the behavior of PT100 in an ideal, interference-free environment.

[0080] When the sensor is PT1000, a normal working EKF model specially designed for it will also be constructed. Similar to PT100, its observation function h(·) is also based on the resistance-temperature relationship. In particular, to handle its non-linear characteristics in a specific temperature range (e.g. 300-660°C), its observation function h(·) can employ piecewise linear function or polynomial function to accurately describe in this range, and the coefficients of the function are also determined through pre-calibration.

[0081] When the sensor is K-type thermocouple, the normal working EKF model constructed for it is more complex. Its observation function h(·) not only describes the relationship between the measured thermoelectric force and the measuring end temperature, but also needs to take the cold end temperature as one of the input variables. In this embodiment, the system uses hardware resistance compensation method and software lookup table method to compensate the cold end temperature, and the corresponding relationship between the compensated net thermoelectric force and the actual temperature (according to the standard calibration table) is used to construct the observation function h(·) of the EKF model.

[0082] Although the embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A high-precision sampling and processing method based on multiple temperature sensors, characterized in that: The method comprises the following steps: S1. Obtaining an original ADC data block of a temperature sensor, and extracting a dynamic feature fingerprint representing a current operating state of the temperature sensor based on the original ADC data block; S2. Analyze the extracted dynamic feature fingerprint to determine a model probability vector for characterizing the likelihood that the temperature sensor is in each of a plurality of known operating modes, and determine a novelty score for characterizing that the temperature sensor is in an unknown operating mode; S3. Based on the model probability vector, from a state space model library, using an interactive multi-model algorithm to fuse estimation results of multiple extended Kalman filters to obtain a temperature value of the temperature sensor, and generate a residual sequence of the interactive multi-model algorithm; S4. Update the state space model library online according to the novelty score or residual sequence, and output the temperature value.

2. The high-precision sampling and processing method based on multiple temperature sensors according to claim 1, characterized in that: In step S1, the steps of obtaining the original ADC data block of the temperature sensor and extracting the dynamic feature fingerprint representing the current operating state of the temperature sensor according to the original ADC data block include: The original ADC data block is a continuous ADC sampling sequence; The process of extracting the dynamic feature fingerprint is as follows: A window function is applied to the original ADC data block, and then a fast Fourier transform is performed on the original ADC data block after the window function is applied to obtain a frequency domain representation. Finally, the energy distribution and spectral peak characteristics of the key frequency band are extracted from the frequency domain representation to form a multi-dimensional dynamic feature fingerprint.

3. The high-precision sampling and processing method based on multiple temperature sensors according to claim 1, characterized in that: In step S2, the step of analyzing the extracted dynamic feature fingerprint includes: Matching the dynamic feature fingerprint with a plurality of reference fingerprint templates in a dynamic feature fingerprint template library to calculate the model probability vector; The dynamic feature fingerprint is input into a generative anomaly detection model, and the novelty score is calculated based on the reconstruction error of the generative anomaly detection model.

4. The high-precision sampling and processing method based on multiple temperature sensors according to claim 3 is characterized in that: The generative anomaly detection model includes an encoder and a decoder. The novelty score is calculated using the dynamic feature fingerprint as input. The calculation formula of the novelty score is: S n =||v f -p θ (q φ (v f ))|| 2 ; Where S n Score for novelty; v f is the dynamic feature fingerprint; q φ is the encoder of the generative anomaly detection model; p θ Decoder for the generative anomaly detection model.

5. The high-precision sampling and processing method based on multiple temperature sensors according to claim 1, characterized in that: In step S3, the steps of obtaining the temperature value of the temperature sensor by fusing the estimation results of multiple extended Kalman filters from a state space model library through an interactive multi-model algorithm according to the model probability vector and generating a residual sequence of the interactive multi-model algorithm include: The state vector of each extended Kalman filter is estimated, wherein the state vector includes a temperature value and a drift parameter for characterizing a change in an electrical characteristic of the temperature sensor itself, and the temperature value is extracted from the fused state vector.

6. The high-precision sampling and processing method based on multiple temperature sensors according to claim 1, characterized in that: In step S4, the step of updating the state space model library online according to the novelty score or the residual sequence and outputting the temperature value includes: When the novelty score is higher than a preset threshold, the online update includes: triggering an active detection process by injecting a disturbance signal into the excitation source of the temperature sensor or increasing the ADC sampling rate to capture enhanced data of the unknown working mode.

7. The high-precision sampling and processing method based on multiple temperature sensors according to claim 6, characterized in that: After the active detection process is triggered, the online update further includes: performing unsupervised clustering on the enhanced data captured during the active detection process to identify patterns of the unknown operating mode; According to the identified pattern of the unknown working mode, a new candidate model is automatically synthesized, and the new candidate model is added to the state space model library.

8. The high-precision sampling and processing method based on multiple temperature sensors according to claim 7, characterized in that: When the interactive multi-model algorithm determines that the temperature sensor is stably operating in a known working mode, the online update uses the residual sequence as a basis to perform online refinement on the model corresponding to the known working mode in the state space model library.

9. The high-precision sampling and processing method based on multiple temperature sensors according to claim 8, characterized in that: The online refinement includes: utilizing the statistical characteristics of the residual sequence to update the noise covariance matrix of the extended Kalman filter corresponding to the known working mode in the state space model library, and updating the reference fingerprint template corresponding to the known working mode in a dynamic feature fingerprint template library.

10. A high-precision sampling and processing system based on multiple temperature sensors, applied to the method according to any one of claims 1 to 9, characterized in that: The system comprises: The data acquisition and feature extraction module is used to obtain the original ADC data block of the temperature sensor and extract the dynamic feature fingerprint based on the original ADC data block; a sensor state perception module, configured to analyze the dynamic feature fingerprint extracted by the data acquisition and feature extraction module to determine a model probability vector and a novelty score; An adaptive temperature estimation module is configured to obtain a temperature value by fusing the estimation results of multiple extended Kalman filters from a state space model library using an interactive multi-model algorithm based on the model probability vector determined by the sensor state perception module, and generate a residual sequence of the interactive multi-model algorithm; A system online evolution module, configured to update the state space model library online according to the novelty score determined by the sensor state perception module or the residual sequence generated by the adaptive temperature estimation module; The temperature value output module is used to output the temperature value obtained by the adaptive temperature estimation module.

Images