CN-121977629-A - Real-time fusion and error compensation method for multi-sensor physical experiment data

CN121977629ACN 121977629 ACN121977629 ACN 121977629ACN-121977629-A

Abstract

The invention relates to the technical field of sensor error compensation, and discloses a method for real-time fusion and error compensation of multi-sensor physical experiment data, which comprises the following steps: s1, a system initialization and sensor reference calibration stage; s2, a multi-source data synchronous acquisition stage; s3, data preprocessing and abnormal purification; s4, a dynamic weight self-adaptive fusion stage; s5, model driving error compensation; s6, performing real-time iterative optimization and data output; the invention aims to provide a real-time fusion and error compensation method for multi-sensor physical experiment data, which aims to solve the problems that in the existing multi-sensor physical application, a fixed weight distribution mode is adopted for data processing, the contribution degree cannot be dynamically adjusted according to the quality of sensor data, the preprocessing means is single, abnormal data caused by environmental interference is difficult to effectively influence, and the dimension conflict of heterogeneous sensor data cannot be properly solved.

Inventors

ZHU YICHEN
Yin Yongyan
TIAN JIAYING
GUO YING

Assignees

陕西理工大学

Dates

Publication Date: 20260505
Application Date: 20260202

Claims (7)

1. The real-time fusion and error compensation method for the multi-sensor physical experiment data is characterized by comprising the following steps of: S1, in the system initialization and sensor reference calibration stage, loading multi-sensor basic parameters, completing sensor zero point and measuring range calibration by adopting a standard calibration source, realizing multi-sensor time synchronization, and setting three reference thresholds of data effective range, abnormal fluctuation and sensor reliability; S2, in a multi-source data synchronous acquisition stage, the multi-channel synchronous acquisition unit sets acquisition frequency according to the characteristics of the sensor, synchronously acquires multi-dimensional physical experiment data, associates exclusive labels for each frame of data and stores the exclusive labels in the annular cache; s3, in the data preprocessing and abnormal purifying stage, abnormal data are removed by adopting a3 sigma rule, missing values are complemented, unified data dimension is normalized through a Z-score, and the real-time credibility of the sensor is calculated; S4, in a dynamic weight self-adaptive fusion stage, extracting time domain characteristics of sensor data, calculating dynamic weights based on an information entropy theory, and completing multi-source data weighted fusion by combining the dynamic weights, standardized data and sensor credibility; S5, in a model driving error compensation stage, an LSTM error prediction model is constructed to predict a system error, and error compensation of the fusion data is realized through reverse offset; S6, in the stage of real-time iterative optimization and data output, the absolute error is calculated by comparing the compensated result with an experimental standard value, and the model parameters are iteratively optimized according to the error threshold value, and data is output in real time through a designated interface or stored in a database.
2. The real-time fusion and error compensation method for the multi-sensor physical experiment data is characterized in that the system initialization and sensor reference calibration phase comprises the following steps of loading multi-sensor basic parameters including sensor type, rated measurement range, precision grade and response time, establishing a sensor parameter library, providing a high-precision zero standard and a measuring range standard by a standard calibration source, enabling the zero standard error to be less than or equal to 0.001% FS, enabling the measuring range standard to cover the full range of the sensor, achieving multi-sensor time synchronization by adopting a PTP protocol, enabling the synchronization error to be less than or equal to 1ms, enabling the reference threshold to comprise a data effective range threshold, an abnormal fluctuation threshold (single data jump is less than or equal to 5% of the measuring range), and enabling the sensor reliability to be the lowest threshold (more than or equal to 0.6).
3. The real-time fusion and error compensation method for the multi-sensor physical experiment data is characterized in that the multi-source data synchronous acquisition phase comprises the following steps that a multi-channel synchronous acquisition unit supports analog signal and digital signal access, acquisition frequency is set according to sensor characteristics, a high-frequency sensor is 100Hz, a medium-speed sensor is 50Hz, a low-frequency sensor is 10Hz, a data tag comprises a sensor ID, an acquisition time stamp and real-time environment parameters, the buffering capacity of an annular buffer memory module is more than or equal to 1000 frames, and an old data automatic coverage mechanism is adopted.
4. The method for real-time fusion and error compensation of multi-sensor physical experiment data according to claim 1, wherein the data preprocessing and anomaly purifying stage comprises the following steps of adopting 3 sigma criterion for outlier rejection, and determining the following formula: wherein For the current data, μ is the mean of the effective data of nearly 100 groups of sensors, For corresponding standard deviation, the abnormal data is complemented by a linear interpolation method, the continuous 3 frames of abnormal data are marked as suspected faults of the sensor, the data normalization adopts a Z-score normalization algorithm, and the formula is as follows: wherein In order to make the data after the normalization, The reliability of the sensor is calculated by a double mechanism of homology comparison and reference comparison, and the formula is as follows: wherein For the reliability of the sensor, the range of the value is 0-1, The data is measured in real time for the current sensor, As a reference value, the similar sensors take the data average value of the homologous sensors, the heterogeneous sensors take the experimental reference standard value, For the upper limit of the nominal measurement of the sensor, The lower limit is rated for the sensor.
5. The method for real-time fusion and error compensation of multi-sensor physical experiment data according to claim 1, wherein the dynamic weight self-adaptive fusion stage comprises the following steps of extracting time domain features including mean value, variance and peak value, calculating dynamic weight in two steps, calculating information entropy in the first step, Wherein For the ith sensor information entropy, the value range is 0-lnm, For a number of valid data samples of approximately 50 sets, For the jth data duty cycle of the ith sensor, the calculation is as follows: , normalized values for the jth data of the ith sensor, When=0 Taking 0, calculating dynamic weight in the second step, Wherein For the i-th sensor dynamic weight, the value range is 0-1, the sum of all the sensor weights is 1, To account for the total number of sensors involved in the fusion, The weighted fusion adopts the formula: wherein In order to fuse the results of the fusion, For the i-th sensor dynamic weight, Normalized values for the ith sensor data, And (5) the real-time credibility of the ith sensor.
6. The method for real-time fusion and error compensation of multi-sensor physical experiment data according to claim 1, wherein said model driving error compensation stage comprises the following steps of inputting characteristics of LSTM error prediction model including real-time environment parameters, sensor working time length, historical deviation of fusion result and experiment standard value, and outputting system error prediction value The model structure is an input layer, a hidden layer and an output layer, the training parameter is a learning rate eta=0.001, the iteration times=1000, the loss function is a mean square error, and the error compensation adopts the formula: wherein In order to compensate for the final fusion result after compensation, In order to not compensate for the fusion result, A system error value predicted for the LSTM model.
7. The method for real-time fusion and error compensation of multi-sensor physical experiment data according to claim 1, wherein the real-time iterative optimization and data output stage comprises the following steps: wherein In order to compensate for the post-fusion result, Setting an error threshold Eth less than or equal to 0.5% of sensor range, adjusting LSTM model parameters and sensor reliability calculation coefficients by adopting a gradient descent method when E > Eth, triggering abnormal early warning when the sensor reliability lasts for 30 frames <0.6, outputting data in two modes including real-time display and database storage, wherein the real-time display output delay is less than or equal to 50ms, and the database adopts a MySQL+MongoDB mixed architecture to support time stamp and sensor ID retrieval.

Description

Real-time fusion and error compensation method for multi-sensor physical experiment data Technical Field The invention relates to the technical field of sensor error compensation, in particular to a method for real-time fusion and error compensation of multi-sensor physical experiment data. Background The multi-sensor physics refers to a technical application mode for integrating sensors with different types and functions in a physical experiment scene, synchronously collecting multidimensional physical quantity data such as temperature, pressure, displacement and the like, providing multi-source data support for experimental analysis and result verification through data cooperative processing, and has the core that the measurement limitation of a single sensor is made up by the complementarity of the multi-sensor, so that experimental data has more comprehensive and reference value; in the existing multi-sensor physical application, a fixed weight distribution mode is adopted for data processing, the contribution degree cannot be dynamically adjusted according to the data quality of the sensor, the preprocessing means is single, abnormal data caused by environmental interference are difficult to effectively influence, and the dimension conflict problem of heterogeneous sensor data cannot be properly solved. Disclosure of Invention The invention aims to provide a real-time fusion and error compensation method for multi-sensor physical experimental data, which aims to solve the problems that in the prior art, in the prior multi-sensor physical application, a fixed weight distribution mode is adopted for data processing, the contribution degree cannot be dynamically adjusted according to the quality of sensor data, the preprocessing means is single, abnormal data caused by environmental interference is difficult to effectively influence, and the dimension conflict problem of heterogeneous sensor data cannot be properly solved. In order to achieve the above purpose, the present invention adopts the following technical scheme: A real-time fusion and error compensation method of multi-sensor physical experiment data comprises the following steps: S1, in the system initialization and sensor reference calibration stage, loading multi-sensor basic parameters, completing sensor zero point and measuring range calibration by adopting a standard calibration source, realizing multi-sensor time synchronization, and setting three reference thresholds of data effective range, abnormal fluctuation and sensor reliability; S2, in a multi-source data synchronous acquisition stage, the multi-channel synchronous acquisition unit sets acquisition frequency according to the characteristics of the sensor, synchronously acquires multi-dimensional physical experiment data, associates exclusive labels for each frame of data and stores the exclusive labels in the annular cache; s3, in the data preprocessing and abnormal purifying stage, abnormal data are removed by adopting a3 sigma rule, missing values are complemented, unified data dimension is normalized through a Z-score, and the real-time credibility of the sensor is calculated; S4, in a dynamic weight self-adaptive fusion stage, extracting time domain characteristics of sensor data, calculating dynamic weights based on an information entropy theory, and completing multi-source data weighted fusion by combining the dynamic weights, standardized data and sensor credibility; S5, in a model driving error compensation stage, an LSTM error prediction model is constructed to predict a system error, and error compensation of the fusion data is realized through reverse offset; S6, in the stage of real-time iterative optimization and data output, the absolute error is calculated by comparing the compensated result with an experimental standard value, and the model parameters are iteratively optimized according to the error threshold value, and data is output in real time through a designated interface or stored in a database. The technical scheme is further improved, the system initialization and sensor reference calibration stage comprises the following steps that loaded multi-sensor basic parameters comprise sensor types, rated measurement ranges, precision grades and response time, a sensor parameter library is built, a standard calibration source provides high-precision zero point standards and measuring range standards, zero point standard errors are less than or equal to 0.001% FS, the measuring range standards cover the full range of a sensor, the multi-sensor time synchronization is achieved by adopting a PTP protocol, synchronization errors are less than or equal to 1ms, a reference threshold comprises a data effective range threshold, an abnormal fluctuation threshold (single data jump is less than or equal to 5% measuring range), and a sensor reliability minimum threshold (more than or equal to 0.6). The multi-source data synchronous acquisition stage comprises the following ste