Investigation data integration analysis method and system based on multi-source data fusion

Abstract

The application relates to the technical field of data processing of multi-source data fusion, and discloses a survey data integrated analysis method and system based on multi-source data fusion. The scheme deploys a multi-source sensor and collects original voltage, and normalizes in the range of each physical quantity; temperature, pressure and vibration three-dimensional observation and state are constructed based on a unified time index, and initial values are set; a constant value is kept to recursively predict; the variances of multiple measuring points are statistically calculated in a static window, and the observation noise is determined by combining the minimum quantization step to calculate the fusion gain; the covariance is linearly updated and corrected, and then the amplitude of each dimension is limited according to the abnormal upper limit; the three-dimensional results of multiple measuring points at each sampling time are spatially averaged, and if necessary, the results are rolled back; a fusion sequence and a statistical report are formed in the whole cycle. The method unifies the data scale, reduces subjective parameters and algorithm burden, models the noise adaptively according to the environment, stably suppresses noise and prevents faults, and guarantees continuous output and comparable analysis.

Description

Technical Field

[0001] This invention relates to the field of data processing technology of multi-source data fusion, specifically to a survey data integration and analysis method based on multi-source data fusion. Background Technology

[0002] With the increasing complexity of applications such as industrial equipment condition surveys, production process assessments, and environmental safety inspections, multiple types of sensors, including temperature, pressure, and vibration sensors, are often deployed simultaneously on-site to collect multi-dimensional data from the measured objects. To facilitate analysis and decision-making, the unified processing and integrated analysis of multi-source heterogeneous sensor data has become an important direction in the field of data processing technology. Currently, in engineering practice, time-series data from different sensors are typically simply summarized, or a single physical quantity is used as the primary indicator, with other measurements only serving as auxiliary references. This approach has significant limitations in terms of data utilization efficiency and analytical accuracy.

[0003] In existing technologies, a common approach is to apply threshold discrimination or moving average filtering to each sensor signal separately, then set fixed weights based on human experience, and finally perform a weighted average of the measurement results from each channel to obtain a certain "comprehensive index." While this method is simple to implement, it usually lacks a rigorous calibration and standardization process. The dimensions and ranges of different physical quantities vary significantly, and direct weighting can lead to unclear physical meanings. Furthermore, it is difficult to guarantee the consistency and comparability of the analysis results when the acquisition conditions change or the survey subjects vary considerably. In addition, fixed weights are often given empirically in the early stages of the experiment and are rarely adjusted later based on the statistical characteristics of the data, making it difficult to reflect sensor performance drift or changes in noise levels in a timely manner. Another type of existing technology uses traditional filtering or estimation methods to smooth single-channel sensor data, such as using simple recursive filtering algorithms for temperature or pressure signals, or frequency domain smoothing for vibration signals. However, this is usually limited to denoising within a single physical quantity and does not systematically model the correlation and complementarity between different physical quantities. In subsequent analysis, multiple signals are often used in isolation, failing to form a unified state description. This makes it impossible to effectively compare and comprehensively evaluate the survey data between sensors, measurement points, and time series. Furthermore, existing multi-source data processing solutions are rather crude in handling sensor calibration errors, quantization noise, and extreme measurement values. Some methods only perform one-time calibration at the initial stage of sensor installation, without subsequently combining actual acquired data for fine-grained noise level estimation; they lack quantitative analysis of the impact of the minimum step size caused by ADC quantization, easily misjudging low-amplitude changes as no change or anomalies; and the handling of outliers and extreme data often relies on simple rejection or overall amplitude limiting, lacking fine-grained labeling and branching logic that combines statistical characteristics and measurement point distribution. This can affect the reliability and representativeness of the data when the surveyed object is affected by sudden environmental changes or local faults.

[0004] Therefore, this case aims to propose a survey data integration and analysis method and system based on multi-source data fusion. First, temperature, pressure and vibration velocity sensors are deployed at equal intervals in the survey area to complete the unified sampling and standardized conversion of the original voltage signals. Then, the state is recursively estimated based on a discrete prediction model that maintains constant values. The observation noise covariance is constructed by observing the residuals and statistical analysis of multiple measurement points within the static time window. The fusion gain is calculated using the recursive least squares principle, thereby realizing the weighted fusion of multi-source data. Summary of the Invention

[0005] This invention provides a survey data integration and analysis method and system based on multi-source data fusion, which helps to solve the problems mentioned in the background art.

[0006] This invention provides the following technical solution: a survey data integration and analysis method based on multi-source data fusion, comprising: Perform the deployment of multi-source sensors for the survey subjects, the acquisition of raw voltage data, and the standardized conversion operations within the range of various physical quantities; Under a unified time index, a three-dimensional normalized observation vector consisting of normalized observations of temperature, pressure, and vibration velocity, and a corresponding three-dimensional state estimation vector are constructed, and an initial state estimation vector and an initial error covariance matrix are set. A constant-preserving discrete prediction model is used to perform time recursion on the three-dimensional state estimation vector and error covariance matrix of the previous sampling time to form the state prediction vector and error covariance prediction matrix of the current sampling time. At the current sampling time, construct the observation residual vector, perform multi-point statistical analysis on the observation data within the static time window, obtain the effective observation variance of each physical quantity and construct the observation noise covariance matrix, and calculate the fusion gain matrix accordingly. The state prediction vector is linearly updated using the fusion gain matrix to obtain an unlimited 3D fusion state estimation vector, and the error covariance matrix is ​​corrected simultaneously. Based on the observation statistics within the static time window, anomaly upper limits for each physical quantity are constructed. The unlimited three-dimensional fusion state estimation vector is then bounded dimension by dimension using the zero lower limit and anomaly upper limit values ​​to obtain the three-dimensional bounded fusion state estimation vector. At each sampling time, the three-dimensional amplitude-limited fusion state estimation vector of all measuring points is spatially averaged according to the three dimensions of temperature, pressure and vibration velocity to obtain the spatial average fusion result after removing abnormal measuring points, and the back-off spatial average fusion result is used when there are no valid samples. Throughout the entire survey period, a fusion output sequence is constructed using the spatial average fusion results at each sampling time. Full-cycle statistical analysis is then performed on each physical quantity to generate a survey data integration report that includes the fusion output sequence and full-cycle statistical indicators.

[0007] Optionally, the process of deploying multi-source sensors for the survey target, acquiring raw voltage data, and performing standardized conversion operations within each physical quantity range specifically includes: Several measuring points are set up on the surface of the survey area at a preset interval. A temperature acquisition unit, a pressure acquisition unit, and a vibration velocity acquisition unit are installed at each measuring point. The three acquisition units output voltage signals. Set a uniform sampling frequency for all acquisition units, and perform discrete sampling of the voltage signal at the uniform sampling frequency throughout the entire survey period. Record the sequence number of each sampling moment and the corresponding continuous time, and record the maximum sampling sequence number of the entire survey period. The raw voltage values ​​of the temperature acquisition unit, pressure acquisition unit, and vibration velocity acquisition unit are collected at each measuring point and at each sampling time to form the raw voltage time series of temperature, pressure, and vibration velocity. Linear calibration is performed on the temperature acquisition unit at each measuring point: Two sets of known temperature points are selected within the temperature channel range, and the raw voltage values ​​of the corresponding temperature acquisition units are collected at the two sets of temperature points respectively. The upper limit of the full-scale voltage of the temperature channel and the number of bits of the analog-to-digital converter of the temperature channel are obtained. The upper limit of the full-scale voltage is divided by the power of two to obtain the minimum quantization voltage step size of the temperature channel. The power exponent is the number of bits of the analog-to-digital converter of the temperature channel minus one. When the absolute value of the difference between the two sets of calibration voltage values ​​is less than twice the minimum quantization voltage step size, the temperature calibration point is reselected until the absolute value of the difference is not less than twice the minimum quantization voltage step size. At the calibration point that meets the conditions, the linear slope and intercept are calculated based on the two sets of temperature and voltage data, and the calculated slope and intercept are used as the linear calibration coefficients of the temperature acquisition unit at the corresponding measuring point. Linear calibration is performed on the pressure acquisition unit at each measuring point: Two sets of known pressure points are selected within the pressure channel range, and the raw voltage values ​​of the pressure acquisition unit are collected at the two sets of pressure points respectively. The upper limit of the full-scale voltage of the pressure channel and the number of bits of the analog-to-digital converter of the pressure channel are obtained. The upper limit of the full-scale voltage of the pressure channel is divided by the power of two to obtain the minimum quantization voltage step size of the pressure channel. The power value is the number of bits of the analog-to-digital converter of the pressure channel minus one. Under the condition that the difference between the two sets of calibration voltages is not less than twice the minimum quantization voltage step size, the linear slope and intercept are calculated based on the two sets of pressure and voltage data, and the calculated slope and intercept are used as the linear calibration coefficients of the pressure acquisition unit at the corresponding measuring point. Linear calibration is performed on the vibration velocity acquisition unit at each measuring point: Two sets of known vibration velocity points are selected within the vibration velocity range. The original voltage values ​​of the vibration velocity acquisition unit are collected at the two sets of vibration velocity points respectively. The upper limit of the full-scale voltage of the vibration velocity channel and the number of bits of the analog-to-digital converter of the vibration velocity channel are obtained. The upper limit of the full-scale voltage of the vibration velocity channel is divided by a power of two to obtain the minimum quantization voltage step size of the vibration velocity channel. The power of two is one less than the number of bits of the analog-to-digital converter of the vibration velocity channel. Under the condition that the difference between the two sets of calibration voltages is not less than twice the minimum quantization voltage step size, the linear slope and intercept of the vibration velocity are calculated based on the two sets of vibration velocity and voltage data. The calculated slope and intercept are used as the linear calibration coefficients of the vibration velocity acquisition unit at the corresponding measuring point. At each measuring point and at each sampling time, the original voltage of the temperature acquisition unit is multiplied by the corresponding temperature linear slope and the corresponding temperature intercept is added; the original voltage of the pressure acquisition unit is multiplied by the corresponding pressure linear slope and the corresponding pressure intercept is added; and the original voltage of the vibration velocity acquisition unit is multiplied by the corresponding vibration velocity linear slope and the corresponding vibration velocity intercept is added to obtain the true temperature value, true pressure value, and true vibration velocity value of the corresponding measuring point at the corresponding sampling time. The lower and upper limits of the range of the physical quantity of temperature, the lower and upper limits of the range of the physical quantity of pressure, and the lower and upper limits of the range of the physical quantity of vibration velocity are obtained respectively, and the value ranges of the three physical quantities are formed accordingly. At each measuring point and at each sampling time, the actual temperature, pressure, and vibration velocity values ​​are standardized: the actual value is subtracted from the corresponding lower limit of the range, and then divided by the difference between the corresponding upper and lower limits of the range to obtain a standardized value between zero and one; when the standardized value is less than zero, it is taken as zero, and when the standardized value is greater than one, it is taken as one, thus forming standardized data sequences for temperature, pressure, and vibration velocity respectively.

[0008] Optionally, the step of constructing a three-dimensional normalized observation vector and a corresponding three-dimensional state estimation vector composed of normalized observations of temperature, pressure, and vibration velocity under a unified time index, and setting an initial state estimation vector and an initial error covariance matrix, specifically includes: At each measuring point and at each sampling time, the standardized temperature value, standardized pressure value, and standardized vibration velocity value of the corresponding measuring point at the corresponding sampling time are arranged in a fixed order of temperature, pressure, and vibration velocity to form a three-dimensional standardized observation vector. A three-dimensional state estimation vector is constructed at each measuring point and at each sampling time. The three components are the normalized true estimate of temperature, the normalized true estimate of pressure, and the normalized true estimate of vibration velocity, respectively. At the start of the survey, the three-dimensional standardized observation vector of each measuring point at the initial sampling time is directly used as the initial state estimation vector of the same measuring point, and the continuous time corresponding to the initial sampling time is recorded. At the start of the survey, an initial error covariance matrix is ​​established for each measuring point. The initial error covariance matrix is ​​a three-row, three-column diagonal matrix, with diagonal elements equal to one and off-diagonal elements equal to zero.

[0009] Optionally, the step of using a constant-preserving discrete prediction model to perform time recursion on the three-dimensional state estimation vector and error covariance matrix of the previous sampling time to form the state prediction vector and error covariance prediction matrix of the current sampling time specifically includes: At any non-initial sampling time, for each measurement point, the three-dimensional state estimation vector updated at the previous sampling time is directly used as the state prediction vector at the current sampling time, so that the state between adjacent sampling times is kept constant and the discrete prediction model is recursively applied. At any non-initial sampling time, for each measurement point, the error covariance matrix updated at the previous sampling time is directly used as the error covariance prediction matrix at the current sampling time.

[0010] Optionally, the step of constructing an observation residual vector at the current sampling time, performing multi-point statistical analysis on the observation data within a static time window, obtaining the effective observation variance of each physical quantity, constructing an observation noise covariance matrix, and calculating the fusion gain matrix accordingly, specifically includes: At each measuring point and at each sampling time, the observation residual vector is calculated dimension by dimension by subtracting the corresponding state prediction vector from the three-dimensional standardized observation vector. Set the observation matrix as a three-row, three-column identity matrix so that each dimension of observation corresponds directly to only the same dimension of state component; A static time window is selected when the system is in a steady state. The static time window is determined by the start sampling number and the end sampling number. Within the static time window, the three-dimensional standardized observation vectors of all measuring points and all sampling times are organized: the temperature dimension forms the first-dimensional sample set, the pressure dimension forms the second-dimensional sample set, and the vibration velocity dimension forms the third-dimensional sample set. The arithmetic mean was calculated for the sample sets of temperature, pressure and vibration velocity respectively, and the global average value of the standardized observations in each dimension was obtained within the static time window. For the three dimensions of temperature, pressure, and vibration velocity, the variance of each sample value is squared and summed with the global mean of the corresponding dimension, and then divided by the total number of samples to obtain the total variance of the standardized observations in each dimension within the static time window. Based on the linear calibration slope of temperature, pressure and vibration velocity, and the minimum quantization voltage step size of the temperature channel, pressure channel and vibration velocity channel, the minimum quantization step size of the physical quantity scale is mapped to the normalization scale according to the corresponding slope. The larger of the normalized quantization step sizes corresponding to all measurement points in the three dimensions is selected as the minimum effective normalized quantization step size of temperature, pressure and vibration velocity. The total variance of the static time window is compared with the lower bound of the observation noise variance derived from the standardized quantization step size in three dimensions, and the larger value in each dimension is taken as the effective observation variance. Construct an observation noise covariance matrix based on the effective observation variance in three dimensions. The diagonal elements of the matrix are the effective observation variances in the three dimensions, and the off-diagonal elements are zero. At each measurement point and at each sampling time, the three-row, three-column fusion gain matrix is ​​calculated based on the recursive least squares fusion principle, using the observation noise covariance matrix and the corresponding measurement point error covariance prediction matrix at the current sampling time as input.

[0011] Optionally, the step of performing a linear update on the state prediction vector using the fusion gain matrix to obtain an unlimited 3D fusion state estimation vector, and simultaneously correcting the error covariance matrix, specifically includes: At each measurement point and at each sampling time, the fusion gain matrix of the corresponding measurement point at the current sampling time is multiplied by the corresponding observation residual vector to obtain a three-dimensional increment vector. The three-dimensional increment vector is then added to the state prediction vector to form an unlimited three-dimensional fusion state estimation vector. At each measurement point and at each sampling time, the error covariance prediction matrix is ​​multiplied by subtracting the fusion gain matrix from the three-row, three-column identity matrix as the left and right coefficients. At the same time, the weighted matrix terms are calculated using the fusion gain matrix and the observation noise covariance matrix as coefficients. The two results are added together to obtain the corrected error covariance matrix.

[0012] Optionally, the step of constructing anomaly upper limits for each physical quantity based on observational statistics within a static time window, and applying zero lower limit and anomaly upper limit values ​​to the unlimited three-dimensional fusion state estimation vector dimensionally to obtain a three-dimensional limited fusion state estimation vector, specifically includes: Within the static time window, a dimensional maximum search is performed on the standardized observations at all measuring points and at all sampling times in the three dimensions of temperature, pressure, and vibration velocity to obtain the global maximum observations in the three dimensions. Based on the effective observation variances of the three dimensions, the square root of the effective observation variance of each dimension is calculated to obtain the effective observation standard deviations of the three dimensions. The corresponding global maximum observation value is added to three times the effective observation standard deviation in each of the three dimensions to obtain the original anomaly upper limit value. The value is then restricted to between zero and one according to the allowed range of the standardized physical quantity to form the final anomaly upper limit value. At any measurement point, at any sampling time, and in any dimension, the range of each component of the unlimited 3D fusion state estimation vector is determined: when the component value is less than zero or greater than the upper limit of the corresponding dimension, the anomaly marker value is set to one; when the component value is between zero and the upper limit of the corresponding dimension, the anomaly marker value is set to zero. At any measurement point, at any sampling time, and in any dimension, amplitude limiting is performed on each component of the unlimited 3D fusion state estimation vector: component values ​​less than zero are replaced with zero, component values ​​greater than the corresponding dimension's abnormal upper limit value are replaced with the corresponding dimension's abnormal upper limit value, and component values ​​between zero and the abnormal upper limit value remain unchanged. At each measurement point and at each sampling time, the fused estimation components of the temperature, pressure, and vibration velocity dimensions are recombined in a fixed order to form a three-dimensional fused state estimation vector after limiting.

[0013] Optionally, the step of spatially averaging the three-dimensional fused state estimation vectors of all measurement points at each sampling time according to the three dimensions of temperature, pressure, and vibration velocity to obtain the spatially averaged fusion result after removing outlier measurement points, and using the backtracked spatially averaged fusion result when there are no valid samples, specifically includes: At each sampling time, the three-dimensional fused state estimation vectors of all measuring points at the current sampling time are arranged in order of measuring point number to form a state matrix with measuring points as rows and temperature, pressure and vibration velocity as columns. At each sampling time, the number of measurement points with anomaly marker values ​​of zero is counted for each of the three dimensions to obtain the number of valid samples in the three dimensions at the current sampling time. When the number of valid samples in a certain dimension is greater than zero, the arithmetic mean of the components of the three-dimensional amplitude-limited fusion state estimation vector of all measurement points with anomaly label values ​​of zero is taken in the same dimension to obtain the spatial average fusion result in the same dimension at the current sampling time. When the number of effective samples in a certain dimension is zero, the arithmetic mean of the components of the three-dimensional fused state estimation vector of all measurement points in the same dimension is calculated, and the calculated average value is used as the back-off space average fusion result in the same dimension at the current sampling time.

[0014] Optionally, the step of constructing a fusion output sequence based on the spatial average fusion results at each sampling time throughout the entire survey period, and performing full-cycle statistical analysis on each physical quantity to generate a survey data integration report containing the fusion output sequence and full-cycle statistical indicators, specifically includes: At each sampling moment, the spatial average fusion results of temperature, pressure and vibration velocity dimensions are combined in a fixed order to form a three-dimensional multi-source fusion output vector for the entire survey object; From the initial sampling time up to the sampling time corresponding to the maximum sampling number, the three-dimensional multi-source fusion output vectors of all sampling times are collected throughout the entire survey period, forming a fusion output sequence arranged in chronological order; Throughout the entire research period, extreme values ​​and mean values ​​were calculated for the corresponding spatial average fusion result sequences for each of the three dimensions, resulting in the maximum, minimum, and average values ​​for each dimension, forming full-cycle statistical indicators for the three dimensions. Based on the fused output sequence and the full-cycle statistical indicators of three dimensions, a survey data integration report is generated. The report includes at least the fused output data in time series form and the maximum, minimum and average values ​​of each physical quantity.

[0015] A system for implementing the survey data integration and analysis method based on multi-source data fusion, comprising: The multi-source data acquisition module performs the deployment of temperature acquisition units, pressure acquisition units, and vibration velocity acquisition units at the measurement points in the survey area, voltage signal sampling, linear calibration, and range standardization processing, and outputs standardized observation data of temperature, pressure, and vibration velocity. The fusion estimation module performs three-dimensional normalized observation vector construction, state prediction, observation residual vector construction, observation noise covariance matrix calculation, fusion gain matrix calculation, state recursive update and error covariance correction, and outputs unlimited and limited three-dimensional fusion state estimation vectors and anomaly labeling results. The multi-point fusion integration module performs the following tasks: constructing the state matrix of the fusion state estimation vector after multi-point three-dimensional amplitude limiting, calculating the number of effective samples in each dimension, calculating the spatial average fusion result, and calculating the back-off spatial average fusion result, and outputting the spatial average fusion result in three dimensions. The results output module receives the spatial average fusion results, constructs the fusion output sequence, performs full-cycle statistical index calculations, and generates and stores a survey data integration report containing the fusion output sequence and full-cycle statistical indicators.

[0016] The present invention has the following beneficial effects: 1. This solution achieves unified acquisition of voltage signals for three different physical quantities—temperature, pressure, and vibration velocity—during sensor deployment and raw data acquisition, and performs linear calibration and range standardization conversion within the measurement range. Compared to the traditional approach that requires separate calibration of each channel and subsequent alignment at the back end, this solution completes two-point calibration, acquisition of the lower and upper limits of the measurement range, automatic calculation of the minimum quantization step size, and dynamic recalibration judgment on-site, forming a closed-loop, fully automated calibration process. This effectively avoids measurement deviations caused by insufficient calibration accuracy or acquisition errors. By dynamically judging the relationship between the difference between two sets of calibration voltages and the minimum quantization step size during the calibration process, the solution can automatically skip invalid calibration points and ensure that the selected points meet the quantization resolution requirements.

[0017] 2. At the start of the complete survey period, this scheme uses the standardized observation vectors collected at the first moment of each measuring point as the initial state estimate, and establishes an initial unit diagonal error covariance matrix for each measuring point. Unlike existing technologies that rely heavily on experience or pre-assumed covariance matrices, this scheme relies entirely on actual observation data for the initial state and covariance matrix settings, avoiding the influence of empirical parameters on the fusion results. By directly using actual observations as the initial state, the dependence on historical data is eliminated. Simultaneously, the initial uncertainty is expressed in the form of an identity matrix, providing a controllable initial error range for the prediction model, enabling the subsequent prediction and fusion processes to have self-correcting capabilities. This "data-driven" initialization method enhances the stability of the model in the initial sample stage, avoids long-term cumulative errors caused by initial parameter deviations, and improves the overall reliability and accuracy of data fusion.

[0018] 3. The proposed solution employs a constant-value-preserving discrete prediction model to recursively estimate the state at the previous time step, maintaining model simplicity while meeting real-time requirements. Unlike complex Kalman or particle filters, which require high computational costs and model parameter adjustments, this method uses only the previous time step's state as the current prediction and recursively estimates the covariance matrix in a statically preserved manner, eliminating the need to introduce any noise or derive system matrices, thus reducing computational overhead and on-site hardware requirements. This lightweight prediction strategy ensures continuous real-time prediction even at high sensor sampling frequencies, while leaving ample room for subsequent observation residual correction and gain calculation, making it suitable for resource-constrained scenarios or those requiring large-scale deployment.

[0019] 4. Before calculating the fusion gain, the scheme performs global statistics on the observation residuals of all measurement points based on a selected static time window, calculates the overall variance of the three physical quantity dimensions, and derives the minimum effective quantization step size by combining the quantization step size, thus dynamically adjusting the lower bound of the observation noise. Simultaneously considering both statistical variance and quantization noise, the larger one is dynamically selected as the effective observation variance, avoiding the underestimation of noise in low-noise environments due to simple variance estimation or the over-conservatism caused by quantization noise in high-noise environments. Unlike existing technologies that fix the noise covariance or require manual parameter tuning, this method achieves true online self-calibration, adapting to drastic changes in sensor noise levels in the field environment. This provides a more realistic and balanced noise model for the fusion gain and effectively improves the robustness and accuracy of the fusion results.

[0020] 6. Utilizing the aforementioned adaptive fusion gain, this scheme performs linear updates to the state vector at each time step, while simultaneously employing a correction rule to perform matrix operations on the covariance matrix. By combining the interaction between the unit diagonal matrix form and the fusion gain matrix, the integrity of the covariance correction is preserved while maintaining computational efficiency. Unlike traditional methods that use the complete Kalman filter formula, resulting in high computational complexity and frequent matrix inversions, this scheme breaks down the correction formula into simple matrix multiplication and addition operations, making it suitable for embedded platforms or high-speed real-time systems. This optimization not only reduces the online computational burden but also ensures that the error metric after covariance correction accurately reflects the uncertainty of the latest fusion result, achieving real-time and reliable error estimation and dynamic feedback.

[0021] 7. To address anomalous peaks or noise abrupt changes in the fused state, this scheme first calculates the original anomaly upper limit for each component after prediction fusion and multi-standard deviation expansion, then prunes it to the standardized interval to achieve double-sided amplitude limiting; simultaneously, anomaly markers are generated for subsequent removal. Compared to the general method of using a fixed threshold and only single-sided amplitude limiting, this strategy combines global statistical extremes and standard deviation expansion, making it more adaptable to changes in noise distribution under dynamic environments; by using multi-standard deviation expansion to identify a small number of true outliers, and then limiting the amplitude value to a reasonable range, it not only prevents excessive truncation of normal fluctuations but also effectively suppresses the interference of outliers on subsequent multi-point statistics, improving the reliability of spatial fusion and reporting indicators.

[0022] This scheme performs an arithmetic average of the amplitude-limited results from the outliers at each sampling time. If the number of valid samples is zero, it degenerates into a direct average of all sampling results. An automatic backoff strategy for zero samples is introduced to address the problem of not being able to generate fusion results in extreme environments or when there is widespread sensor failure. Unlike fixed fusion weights or backoff heavily reliant on experience, this mechanism ensures that the system can output fusion results at any sampling time, avoiding system failure or output interruption. This design enhances robustness and availability in high-reliability scenarios such as disaster monitoring and industrial safety, maximizing the usable time window of the fusion results and effectively meeting the reliability requirements of continuous monitoring and reporting.

[0023] 8. This solution sequentially combines the fusion output vectors from the entire sampling period into a time series, and calculates the extreme values ​​and average values ​​for each of the three dimensions to form a complete set of full-cycle statistical indicators, generating a survey report containing both time series data and statistical indicators. Unlike traditional monitoring methods that only output single fusion results or lack statistical analysis, this solution provides comprehensive analysis from instantaneous details to periodic fluctuations, facilitating subsequent decision support and trend assessment. Combined with a modular system design—from data acquisition, fusion estimation, spatial integration to modularized layering of report output—it meets the needs of large-scale deployment and platform integration, providing a one-stop solution for applications such as engineering monitoring, environmental surveys, and intelligent detection. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Example, refer to Figure 1 The survey data integration and analysis method based on multi-source data fusion includes: Perform the deployment of multi-source sensors for the survey subjects, the acquisition of raw voltage data, and the standardized conversion operations within the range of various physical quantities; Under a unified time index, a three-dimensional normalized observation vector consisting of normalized observations of temperature, pressure, and vibration velocity, and a corresponding three-dimensional state estimation vector are constructed, and an initial state estimation vector and an initial error covariance matrix are set. A constant-preserving discrete prediction model is used to perform time recursion on the three-dimensional state estimation vector and error covariance matrix of the previous sampling time to form the state prediction vector and error covariance prediction matrix of the current sampling time. At the current sampling time, construct the observation residual vector, perform multi-point statistical analysis on the observation data within the static time window, obtain the effective observation variance of each physical quantity and construct the observation noise covariance matrix, and calculate the fusion gain matrix accordingly. The state prediction vector is linearly updated using the fusion gain matrix to obtain an unlimited 3D fusion state estimation vector, and the error covariance matrix is ​​corrected simultaneously. Based on the observation statistics within the static time window, anomaly upper limits for each physical quantity are constructed. The unlimited three-dimensional fusion state estimation vector is then bounded dimension by dimension using the zero lower limit and anomaly upper limit values ​​to obtain the three-dimensional bounded fusion state estimation vector. At each sampling time, the three-dimensional amplitude-limited fusion state estimation vector of all measuring points is spatially averaged according to the three dimensions of temperature, pressure and vibration velocity to obtain the spatial average fusion result after removing abnormal measuring points, and the back-off spatial average fusion result is used when there are no valid samples. Throughout the entire survey period, a fusion output sequence is constructed using the spatial average fusion results at each sampling time. Full-cycle statistical analysis is then performed on each physical quantity to generate a survey data integration report that includes the fusion output sequence and full-cycle statistical indicators.

[0027] First, the deployment and standardized conversion of multi-source sensors unifies the data scale at the source, avoiding weighting imbalances caused by differences in range and quantization accuracy between different physical quantities or channels. Existing technologies often require manual calibration and post-processing normalization, which is prone to subjective bias. Second, by constructing three-dimensional observation and state vectors under a unified time index, and using actual observations as the initial state and the unit diagonal matrix as the initial covariance, the reliance on empirical parameters is completely eliminated. This ensures that the fusion process is consistent with real data from the start, improving the reliability of the algorithm's startup phase. Third, the use of a constant-value-preserving discrete prediction model simplifies the process and addresses the issues of insufficient real-time environmental processing capabilities or limited computing resources on embedded platforms. Compared to traditional Kalman filtering or particle filtering, the computational load is reduced while accuracy is still guaranteed through subsequent gain correction. Finally, multidimensional variance statistics are performed on the residuals of all measurement points within a static time window, and the noise lower bound is dynamically selected based on the minimum quantization step size, forming an adaptive observation noise covariance matrix. This effectively suppresses the risk of a fixed noise model distorting the fusion results when the environment changes. Next, the fusion gain is used in state updates and covariance corrections, achieving real-time and efficient computation through matrix multiplication and addition, avoiding complex matrix inversion. More importantly, the anomaly limiting and spatial averaging framework based on statistical indicators not only eliminates the impact of noise abrupt changes or single-point failures, but also automatically backs up the fusion results when a few measurement points fail, ensuring uninterrupted output throughout the entire cycle. Finally, the fusion output sequence is constructed and a full-cycle statistical report is provided, achieving a seamless transition from instantaneous time-domain estimation to periodic analysis.

[0028] The process of deploying multi-source sensors, acquiring raw voltage data, and performing standardized conversion operations within various physical quantity ranges for the survey targets specifically includes: Several measuring points are set up on the surface of the survey area at a preset interval. A temperature acquisition unit, a pressure acquisition unit, and a vibration velocity acquisition unit are installed at each measuring point. The three acquisition units output voltage signals. Set a uniform sampling frequency for all acquisition units, and perform discrete sampling of the voltage signal at the uniform sampling frequency throughout the entire survey period. Record the sequence number of each sampling moment and the corresponding continuous time, and record the maximum sampling sequence number of the entire survey period. The raw voltage values ​​of the temperature acquisition unit, pressure acquisition unit, and vibration velocity acquisition unit are collected at each measuring point and at each sampling time to form the raw voltage time series of temperature, pressure, and vibration velocity. Linear calibration is performed on the temperature acquisition unit at each measuring point: Two sets of known temperature points are selected within the temperature channel range, and the raw voltage values ​​of the corresponding temperature acquisition units are collected at the two sets of temperature points respectively. The upper limit of the full-scale voltage of the temperature channel and the number of bits of the analog-to-digital converter of the temperature channel are obtained. The upper limit of the full-scale voltage is divided by the power of two to obtain the minimum quantization voltage step size of the temperature channel. The power exponent is the number of bits of the analog-to-digital converter of the temperature channel minus one. When the absolute value of the difference between the two sets of calibration voltage values ​​is less than twice the minimum quantization voltage step size, the temperature calibration point is reselected until the absolute value of the difference is not less than twice the minimum quantization voltage step size. At the calibration point that meets the conditions, the linear slope and intercept are calculated based on the two sets of temperature and voltage data, and the calculated slope and intercept are used as the linear calibration coefficients of the temperature acquisition unit at the corresponding measuring point. Linear calibration is performed on the pressure acquisition unit at each measuring point: Two sets of known pressure points are selected within the pressure channel range, and the raw voltage values ​​of the pressure acquisition unit are collected at the two sets of pressure points respectively. The upper limit of the full-scale voltage of the pressure channel and the number of bits of the analog-to-digital converter of the pressure channel are obtained. The upper limit of the full-scale voltage of the pressure channel is divided by the power of two to obtain the minimum quantization voltage step size of the pressure channel. The power value is the number of bits of the analog-to-digital converter of the pressure channel minus one. Under the condition that the difference between the two sets of calibration voltages is not less than twice the minimum quantization voltage step size, the linear slope and intercept are calculated based on the two sets of pressure and voltage data, and the calculated slope and intercept are used as the linear calibration coefficients of the pressure acquisition unit at the corresponding measuring point. Linear calibration is performed on the vibration velocity acquisition unit at each measuring point: Two sets of known vibration velocity points are selected within the vibration velocity range. The original voltage values ​​of the vibration velocity acquisition unit are collected at the two sets of vibration velocity points respectively. The upper limit of the full-scale voltage of the vibration velocity channel and the number of bits of the analog-to-digital converter of the vibration velocity channel are obtained. The upper limit of the full-scale voltage of the vibration velocity channel is divided by a power of two to obtain the minimum quantization voltage step size of the vibration velocity channel. The power of two is one less than the number of bits of the analog-to-digital converter of the vibration velocity channel. Under the condition that the difference between the two sets of calibration voltages is not less than twice the minimum quantization voltage step size, the linear slope and intercept of the vibration velocity are calculated based on the two sets of vibration velocity and voltage data. The calculated slope and intercept are used as the linear calibration coefficients of the vibration velocity acquisition unit at the corresponding measuring point. At each measuring point and at each sampling time, the original voltage of the temperature acquisition unit is multiplied by the corresponding temperature linear slope and the corresponding temperature intercept is added; the original voltage of the pressure acquisition unit is multiplied by the corresponding pressure linear slope and the corresponding pressure intercept is added; and the original voltage of the vibration velocity acquisition unit is multiplied by the corresponding vibration velocity linear slope and the corresponding vibration velocity intercept is added to obtain the true temperature value, true pressure value, and true vibration velocity value of the corresponding measuring point at the corresponding sampling time. The lower and upper limits of the range of the physical quantity of temperature, the lower and upper limits of the range of the physical quantity of pressure, and the lower and upper limits of the range of the physical quantity of vibration velocity are obtained respectively, and the value ranges of the three physical quantities are formed accordingly. At each measuring point and at each sampling time, the actual temperature, pressure, and vibration velocity values ​​are standardized: the actual value is subtracted from the corresponding lower limit of the range, and then divided by the difference between the corresponding upper and lower limits of the range to obtain a standardized value between zero and one; when the standardized value is less than zero, it is taken as zero, and when the standardized value is greater than one, it is taken as one, thus forming standardized data sequences for temperature, pressure, and vibration velocity respectively.

[0029] Further specific implementation steps include: The survey area was laid out at equal intervals. There are 1 measuring point, numbered as follows: Each measuring point is equipped with three different types of physical quantity acquisition sensors: a temperature acquisition unit, a pressure acquisition unit, and a vibration acquisition unit; among them, The total number of measuring points deployed on the surface of the survey area; The index number of the measurement point; Each sensor outputs a voltage value, and the sampling frequency is set to [value missing]. Any sampling time within the sampling period is ;in, To provide a unified sampling frequency for multiple source sensors; This refers to the discrete-time index number; For discrete-time indexing Corresponding consecutive moments; The maximum sampling index value during the entire survey period is denoted as . ; Then the first At each measuring point, the three types of sensors are at time... The output is: Temperature and original voltage Pressure original voltage Vibration original voltage ;in, , , The first Each measuring point temperature acquisition unit, pressure acquisition unit, and vibration acquisition unit continuously... The original voltage output value; The temperature sensor is linearly calibrated as follows: Select two known temperature points respectively , ;in, , These are the temperature point values ​​for the first and second groups of temperature calibration, respectively. , The first The original voltage values ​​measured by the temperature sensors at each measuring point under the first and second sets of calibration temperatures; Obtain the upper limit of the full-scale voltage range of the temperature channel, denoted as . ; Acquire temperature channel voltage sampling Bit width, denoted as ; The minimum quantization voltage step size for constructing the temperature channel is ; like Then reselect the calibration point until... Finish; A temperature calibration coefficient is established within the linear interval, specifically as follows: , ;in, , The first The temperature and voltage linear slope and linear calibration intercept of a temperature sensor; The pressure sensor is linearly calibrated as follows: Select two known pressure points respectively , ;in, , These are the pressure point values ​​for the first and second groups of temperature calibration, respectively. , The first The original voltage values ​​measured by the pressure sensors at each measuring point under the first and second sets of calibration pressures; Obtain the upper limit of the full-scale voltage range of the pressure channel, denoted as . ; Acquire pressure channel voltage sampling Bit width, denoted as ; The minimum quantization voltage step size for constructing the pressure channel is ; like Then reselect the calibration point until... Finish; Establish pressure calibration coefficients within the linear interval, specifically as follows: , ;in, , The first The pressure and voltage linear slope and linear calibration intercept of a pressure sensor; The vibration sensor is linearly calibrated as follows: Select two known vibration velocity points respectively , ;in, , These are the pressure point values ​​for the first and second groups, respectively, for vibration velocity calibration. , The first The original voltage values ​​measured by the vibration sensors at each measuring point under the first and second sets of calibration vibrations; Obtain the upper limit of the full-scale voltage range of the vibration channel, denoted as . ; Acquire vibration channel voltage sampling Bit width, denoted as ; The minimum quantization voltage step size for constructing the vibration channel is ; like Then reselect the calibration point until... Finish; Vibration calibration coefficients are established within the linear interval, specifically as follows: , ;in, , The first Vibration velocity of a vibration sensor versus voltage linear slope and linear calibration intercept; The converted physical values ​​are as follows: , , ;in, , , The first Each measuring point at consecutive times The actual temperature value, actual pressure value, and actual vibration velocity value; Obtain the minimum value of the temperature range respectively. With the maximum value The temperature range of composition is ; Obtain the minimum value of the pressure range respectively. With the maximum value The pressure range is ; Obtain the minimum value of the vibration velocity range respectively. With the maximum value The range of vibration velocities is ; For each type of physical quantity, standardization and explicit amplitude limiting are performed within the measurement range, specifically as follows: , , ;in, , , The first Each measuring point at time [time] Standardized temperature values, standardized pressure values, and standardized vibration values.

[0030] The process of constructing a three-dimensional normalized observation vector and a corresponding three-dimensional state estimation vector, composed of normalized observations of temperature, pressure, and vibration velocity under a unified time index, and setting an initial state estimation vector and an initial error covariance matrix, specifically includes: At each measuring point and at each sampling time, the standardized temperature value, standardized pressure value, and standardized vibration velocity value of the corresponding measuring point at the corresponding sampling time are arranged in a fixed order of temperature, pressure, and vibration velocity to form a three-dimensional standardized observation vector. A three-dimensional state estimation vector is constructed at each measuring point and at each sampling time. The three components are the normalized true estimate of temperature, the normalized true estimate of pressure, and the normalized true estimate of vibration velocity, respectively. At the start of the survey, the three-dimensional standardized observation vector of each measuring point at the initial sampling time is directly used as the initial state estimation vector of the same measuring point, and the continuous time corresponding to the initial sampling time is recorded. At the start of the survey, an initial error covariance matrix is ​​established for each measuring point. The initial error covariance matrix is ​​a three-row, three-column diagonal matrix, with diagonal elements equal to one and off-diagonal elements equal to zero.

[0031] Further specific implementation steps include: The standardized observation vector is constructed as follows: ;in, For the first Each measuring point at time [time] The three-dimensional normalized observation vector formed by three types of sensors has a dimension of ; It is a three-dimensional real number space; The fused state vector is constructed as follows: ;in, For the first Each measuring point at time [time] The three-dimensional true state estimation vector, with dimension . ; , , The first Each measuring point at time [time] The standardized true temperature estimate, standardized true pressure estimate, and standardized true vibration estimate; The initial estimate is set to the data collected at the first time step: ;in, For the first Each measuring point at the initial time State estimates; The continuous initial moments when the survey system begins collecting data; The initial state error covariance matrix is: ;in, For the first Each measuring point at the initial time The state estimation error covariance matrix has dimensions of .

[0032] The method employs a constant-preserving discrete prediction model to perform time-based recursion on the three-dimensional state estimation vector and error covariance matrix of the previous sampling time, forming the state prediction vector and error covariance prediction matrix of the current sampling time. Specifically, this includes: At any non-initial sampling time, for each measurement point, the three-dimensional state estimation vector updated at the previous sampling time is directly used as the state prediction vector at the current sampling time, so that the state between adjacent sampling times is kept constant and the discrete prediction model is recursively applied. At any non-initial sampling time, for each measurement point, the error covariance matrix updated at the previous sampling time is directly used as the error covariance prediction matrix at the current sampling time.

[0033] Further specific implementation steps include: The prediction model is sampled in a constant-preservation form, specifically as follows: ;in, For a known time Given the state estimation results, for time... The predicted state value; For the first Each measuring point at time [time] State estimation results; Covariance propagation uses static preservation, specifically: ,in, For a known time Under the premise of covariance, for time Error covariance prediction; For the first Each measuring point at time [time] The error covariance matrix.

[0034] The process of constructing an observation residual vector at the current sampling time, performing multi-point statistical analysis of the observation data within a static time window, obtaining the effective observation variance of each physical quantity, constructing an observation noise covariance matrix, and calculating the fusion gain matrix accordingly includes: At each measuring point and at each sampling time, the observation residual vector is calculated dimension by dimension by subtracting the corresponding state prediction vector from the three-dimensional standardized observation vector. Set the observation matrix as a three-row, three-column identity matrix so that each dimension of observation corresponds directly to only the same dimension of state component; A static time window is selected when the system is in a steady state. The static time window is determined by the start sampling number and the end sampling number. Within the static time window, the three-dimensional standardized observation vectors of all measuring points and all sampling times are organized: the temperature dimension forms the first-dimensional sample set, the pressure dimension forms the second-dimensional sample set, and the vibration velocity dimension forms the third-dimensional sample set. The arithmetic mean was calculated for the sample sets of temperature, pressure and vibration velocity respectively, and the global average value of the standardized observations in each dimension was obtained within the static time window. For the three dimensions of temperature, pressure, and vibration velocity, the variance of each sample value is squared and summed with the global mean of the corresponding dimension, and then divided by the total number of samples to obtain the total variance of the standardized observations in each dimension within the static time window. Based on the linear calibration slope of temperature, pressure and vibration velocity, and the minimum quantization voltage step size of the temperature channel, pressure channel and vibration velocity channel, the minimum quantization step size of the physical quantity scale is mapped to the normalization scale according to the corresponding slope. The larger of the normalized quantization step sizes corresponding to all measurement points in the three dimensions is selected as the minimum effective normalized quantization step size of temperature, pressure and vibration velocity. The total variance of the static time window is compared with the lower bound of the observation noise variance derived from the standardized quantization step size in three dimensions, and the larger value in each dimension is taken as the effective observation variance. Construct an observation noise covariance matrix based on the effective observation variance in three dimensions. The diagonal elements of the matrix are the effective observation variances in the three dimensions, and the off-diagonal elements are zero. At each measurement point and at each sampling time, the three-row, three-column fusion gain matrix is ​​calculated based on the recursive least squares fusion principle, using the observation noise covariance matrix and the corresponding measurement point error covariance prediction matrix at the current sampling time as input.

[0035] Further specific implementation steps include: The observation error residuals are constructed as follows: ;in, For the first Each measuring point at time [time] The three-dimensional observation residuals; Set the observation matrix to the identity matrix: ;in, The observation matrix; for Unit array; Select a static time window in which the system is in steady state, denoted as . ;in, The discrete sampling index at the start of the static time window; The discrete sampling index at the end of the static time window; From the start time By the end time A continuous time interval; Will The Each component is denoted as ;in, For the dimension index in the observation, Time represents the temperature dimension. Time represents the stress dimension. Time represents the vibration dimension; For the first Each measuring point at time [time] At that time, the first Dimensionally standardized observations; Within this window, statistics are performed on all measurement points and all time samples, specifically: remember ;in, Within a static window, the dimension is recorded using an index. Each measurement point is in the discrete index Corresponding time The Dimensionally standardized observations; Calculate the total number of samples consisting of all measurement points and all time indices within the static window, specifically: ; Calculate the first Global average observations of dimension: ;in, For the first static window The global average of the standardized observations; The corresponding global observation variance is: , ;in, For the first static window The population variance of standardized observations; The quantization step size of the standardized domain is derived from the calibration coefficients and measurement range, specifically as follows: , , ;in, , , These are the minimum effective quantization steps for temperature, pressure, and vibration in the standardized domain after unit conversion; Based on this, the effective lower bound of the observation variance is modified as follows: , ;in, The first bound after combining statistical variance and quantization noise. Dimensional effective observation variance; The global observation covariance matrix is ​​constructed as follows: ;in, Let be the covariance matrix of the three-dimensional observation noise, with dimension . ; The input set of scalars is placed on the main diagonal of the matrix to form a diagonal matrix; , , These correspond to the effective variances of the standardized observation errors for temperature, pressure, and vibration, respectively. Construct the fusion gain matrix: ;in, For the first Each observation point at time The three-dimensional recursive least squares fusion gain matrix, with dimensions of .

[0036] The process of using the fusion gain matrix to perform a linear update on the state prediction vector to obtain an unlimited 3D fusion state estimation vector, and simultaneously correcting the error covariance matrix, specifically includes: At each measurement point and at each sampling time, the fusion gain matrix of the corresponding measurement point at the current sampling time is multiplied by the corresponding observation residual vector to obtain a three-dimensional increment vector. The three-dimensional increment vector is then added to the state prediction vector to form an unlimited three-dimensional fusion state estimation vector. At each measurement point and at each sampling time, the error covariance prediction matrix is ​​multiplied by subtracting the fusion gain matrix from the three-row, three-column identity matrix as the left and right coefficients. At the same time, the weighted matrix terms are calculated using the fusion gain matrix and the observation noise covariance matrix as coefficients. The two results are added together to obtain the corrected error covariance matrix.

[0037] Further specific implementation steps include: Perform a status update: ;in, For the first Each observation point at time The unconstrained fusion state estimation has not yet undergone anomalous amplitude limiting processing; Covariance correction is performed as follows: ;in, For the first Each observation point at time The result of error covariance correction.

[0038] The process involves constructing anomaly upper limits for each physical quantity based on observational statistics within a static time window, and then applying zero lower bound and anomaly upper bound values ​​to the unconstrained 3D fusion state estimation vector dimension-by-dimensional constraint to obtain the 3D constrained fusion state estimation vector. Specifically, this includes: Within the static time window, a dimensional maximum search is performed on the standardized observations at all measuring points and at all sampling times in the three dimensions of temperature, pressure, and vibration velocity to obtain the global maximum observations in the three dimensions. Based on the effective observation variances of the three dimensions, the square root of the effective observation variance of each dimension is calculated to obtain the effective observation standard deviations of the three dimensions. The corresponding global maximum observation value is added to three times the effective observation standard deviation in each of the three dimensions to obtain the original anomaly upper limit value. The value is then restricted to between zero and one according to the allowed range of the standardized physical quantity to form the final anomaly upper limit value. At any measurement point, at any sampling time, and in any dimension, the range of each component of the unlimited 3D fusion state estimation vector is determined: when the component value is less than zero or greater than the upper limit of the corresponding dimension, the anomaly marker value is set to one; when the component value is between zero and the upper limit of the corresponding dimension, the anomaly marker value is set to zero. At any measurement point, at any sampling time, and in any dimension, amplitude limiting is performed on each component of the unlimited 3D fusion state estimation vector: component values ​​less than zero are replaced with zero, component values ​​greater than the corresponding dimension's abnormal upper limit value are replaced with the corresponding dimension's abnormal upper limit value, and component values ​​between zero and the abnormal upper limit value remain unchanged. At each measurement point and at each sampling time, the fused estimation components of the temperature, pressure, and vibration velocity dimensions are recombined in a fixed order to form a three-dimensional fused state estimation vector after limiting.

[0039] Further specific implementation steps include: In static window Calculate the first number within this window. Maximum value of global observations Specifically: , ; Utilizing effective observation variance Constructing the standard deviation ;in, For the first 1 / 3 effective observation standard deviation; First, construct the initial anomaly limit, specifically as follows: , ;in, This is the original upper limit obtained by expanding based on the static maximum value and 3 times the standard deviation; Then, by limiting it to within 1, the upper bound allowed by the standardized physical quantity, we obtain the final anomaly upper limit: , ;in, For the final use of the limiter Upper limit of anomalies; For any measuring point At any time and any dimension First, based on unconstrained estimation and interval The relationship constructs the exception marking function, specifically: ; in, For the first Each measuring point at time [time] The unconstrained fusion state estimation in the th Dimensional component values; For the first Each measuring point at time [time] The This is a flag indicating whether a dimension is considered abnormal; a value of 1 indicates an abnormality, and a value of 0 indicates normality. Then, a two-sided clipping operation is performed on each component to obtain the clipped estimate: , ;in, For the first Each measuring point at time [time] The Dimensionally limited fusion estimate; Finally, the three-dimensional clipped components are recombined into a clipped fused state vector: ;in, For the first Each measuring point at time [time] The three-dimensional fused state vector after amplitude limiting.

[0040] The process involves spatially averaging the three-dimensional amplitude-limited fusion state estimation vectors of all measurement points at each sampling time along three dimensions: temperature, pressure, and vibration velocity. This yields a spatially averaged fusion result after removing outlier measurement points. When no valid samples are available, a backtracking spatially averaged fusion result is used. Specifically, this includes: At each sampling time, the three-dimensional fused state estimation vectors of all measuring points at the current sampling time are arranged in order of measuring point number to form a state matrix with measuring points as rows and temperature, pressure and vibration velocity as columns. At each sampling time, the number of measurement points with anomaly marker values ​​of zero is counted for each of the three dimensions to obtain the number of valid samples in the three dimensions at the current sampling time. When the number of valid samples in a certain dimension is greater than zero, the arithmetic mean of the components of the three-dimensional amplitude-limited fusion state estimation vector of all measurement points with anomaly label values ​​of zero is taken in the same dimension to obtain the spatial average fusion result in the same dimension at the current sampling time. When the number of effective samples in a certain dimension is zero, the arithmetic mean of the components of the three-dimensional fused state estimation vector of all measurement points in the same dimension is calculated, and the calculated average value is used as the back-off space average fusion result in the same dimension at the current sampling time.

[0041] Further specific implementation steps include: The fused states of all sensor points are summarized into a state matrix: ;in, For at any time all A three-dimensional fusion state set of measurement points, with dimension . ; It is the space of real numbers; First, for each dimension Calculation time Valid sample size: , ;in, For at any time No. The number of valid measurement points involved in the average calculation; when When an anomaly is removed, the arithmetic mean is used for calculation: ;in, For at any time No. Spatial average fusion result in 3D; when In this case, a simple average without removing outliers is used as a rollback strategy: .

[0042] The process involves constructing a fusion output sequence based on the spatial average fusion results at each sampling time throughout the entire survey period, performing full-cycle statistical analysis on each physical quantity, and generating a survey data integration report that includes the fusion output sequence and full-cycle statistical indicators. Specifically, this includes: At each sampling moment, the spatial average fusion results of temperature, pressure and vibration velocity dimensions are combined in a fixed order to form a three-dimensional multi-source fusion output vector for the entire survey object; From the initial sampling time up to the sampling time corresponding to the maximum sampling number, the three-dimensional multi-source fusion output vectors of all sampling times are collected throughout the entire survey period, forming a fusion output sequence arranged in chronological order; Throughout the entire research period, extreme values ​​and mean values ​​were calculated for the corresponding spatial average fusion result sequences for each of the three dimensions, resulting in the maximum, minimum, and average values ​​for each dimension, forming full-cycle statistical indicators for the three dimensions. Based on the fused output sequence and the full-cycle statistical indicators of three dimensions, a survey data integration report is generated. The report includes at least the fused output data in time series form and the maximum, minimum and average values ​​of each physical quantity.

[0043] Further specific implementation steps include: Construct the fused output state vector: ;in, For at any time The overall three-dimensional multi-source fusion output of the surveyed object shows that the first component corresponds to temperature, the second component corresponds to pressure, and the third component corresponds to vibration. Let the entire research period start from time 1. Continue until The corresponding sampling index is from arrive ,common Each sampling time; Within this interval, the extreme values ​​and means of the three types of indicators are calculated separately, as follows: , , , ;in, , , The respective times within the full sampling period The maximum, minimum, and average values ​​of the dimensional space average fusion results; The final survey data integration report will be generated, and the report content will include: S801, Full-Time Fusion Output Sequence ; S802, Set of full-cycle statistical indicators for three types of physical quantities .

[0044] This embodiment also provides a system for integrated analysis of survey data based on multi-source data fusion, including: The multi-source data acquisition module performs the deployment of temperature acquisition units, pressure acquisition units, and vibration velocity acquisition units at the measurement points in the survey area, voltage signal sampling, linear calibration, and range standardization processing, and outputs standardized observation data of temperature, pressure, and vibration velocity. The fusion estimation module performs three-dimensional normalized observation vector construction, state prediction, observation residual vector construction, observation noise covariance matrix calculation, fusion gain matrix calculation, state recursive update and error covariance correction, and outputs unlimited and limited three-dimensional fusion state estimation vectors and anomaly labeling results. The multi-point fusion integration module performs the following tasks: constructing the state matrix of the fusion state estimation vector after multi-point three-dimensional amplitude limiting, calculating the number of effective samples in each dimension, calculating the spatial average fusion result, and calculating the back-off spatial average fusion result, and outputting the spatial average fusion result in three dimensions. The results output module receives the spatial average fusion results, constructs the fusion output sequence, performs full-cycle statistical index calculations, and generates and stores a survey data integration report containing the fusion output sequence and full-cycle statistical indicators.

[0045] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0046] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A survey data integration and analysis method based on multi-source data fusion, characterized in that, include: Perform the deployment of multi-source sensors for the survey subjects, the acquisition of raw voltage data, and the standardized conversion operations within the range of various physical quantities; Under a unified time index, a three-dimensional normalized observation vector consisting of normalized observations of temperature, pressure, and vibration velocity, and a corresponding three-dimensional state estimation vector are constructed, and an initial state estimation vector and an initial error covariance matrix are set. A constant-preserving discrete prediction model is used to perform time recursion on the three-dimensional state estimation vector and error covariance matrix of the previous sampling time to form the state prediction vector and error covariance prediction matrix of the current sampling time. At the current sampling time, construct the observation residual vector, perform multi-point statistical analysis on the observation data within the static time window, obtain the effective observation variance of each physical quantity and construct the observation noise covariance matrix, and calculate the fusion gain matrix accordingly. The state prediction vector is linearly updated using the fusion gain matrix to obtain an unlimited 3D fusion state estimation vector, and the error covariance matrix is ​​corrected simultaneously. Based on the observation statistics within the static time window, anomaly upper limits for each physical quantity are constructed. The unlimited three-dimensional fusion state estimation vector is then bounded dimension by dimension using the zero lower limit and anomaly upper limit values ​​to obtain the three-dimensional bounded fusion state estimation vector. At each sampling time, the three-dimensional amplitude-limited fusion state estimation vector of all measuring points is spatially averaged according to the three dimensions of temperature, pressure and vibration velocity to obtain the spatial average fusion result after removing abnormal measuring points, and the back-off spatial average fusion result is used when there are no valid samples. Throughout the entire survey period, a fusion output sequence is constructed using the spatial average fusion results at each sampling time. Full-cycle statistical analysis is then performed on each physical quantity to generate a survey data integration report that includes the fusion output sequence and full-cycle statistical indicators.

2. The survey data integration and analysis method based on multi-source data fusion according to claim 1, characterized in that, The process of deploying multi-source sensors, acquiring raw voltage data, and performing standardized conversion operations within various physical quantity ranges for the survey targets specifically includes: Several measuring points are set up on the surface of the survey area at a preset interval. A temperature acquisition unit, a pressure acquisition unit, and a vibration velocity acquisition unit are installed at each measuring point. The three acquisition units output voltage signals. Set a uniform sampling frequency for all acquisition units, and perform discrete sampling of the voltage signal at the uniform sampling frequency throughout the entire survey period. Record the sequence number of each sampling moment and the corresponding continuous time, and record the maximum sampling sequence number of the entire survey period. The raw voltage values ​​of the temperature acquisition unit, pressure acquisition unit, and vibration velocity acquisition unit are collected at each measuring point and at each sampling time to form the raw voltage time series of temperature, pressure, and vibration velocity. Linear calibration is performed on the temperature acquisition unit at each measuring point: Two sets of known temperature points are selected within the temperature channel range, and the raw voltage values ​​of the corresponding temperature acquisition units are collected at the two sets of temperature points respectively. The upper limit of the full-scale voltage of the temperature channel and the number of bits of the analog-to-digital converter of the temperature channel are obtained. The upper limit of the full-scale voltage is divided by the power of two to obtain the minimum quantization voltage step size of the temperature channel. The power exponent is the number of bits of the analog-to-digital converter of the temperature channel minus one. When the absolute value of the difference between the two sets of calibration voltage values ​​is less than twice the minimum quantization voltage step size, the temperature calibration point is reselected until the absolute value of the difference is not less than twice the minimum quantization voltage step size. At the calibration point that meets the conditions, the linear slope and intercept are calculated based on the two sets of temperature and voltage data, and the calculated slope and intercept are used as the linear calibration coefficients of the temperature acquisition unit at the corresponding measuring point. Linear calibration is performed on the pressure acquisition unit at each measuring point: Two sets of known pressure points are selected within the pressure channel range, and the raw voltage values ​​of the pressure acquisition unit are collected at the two sets of pressure points respectively. The upper limit of the full-scale voltage of the pressure channel and the number of bits of the analog-to-digital converter of the pressure channel are obtained. The upper limit of the full-scale voltage of the pressure channel is divided by the power of two to obtain the minimum quantization voltage step size of the pressure channel. The power value is the number of bits of the analog-to-digital converter of the pressure channel minus one. Under the condition that the difference between the two sets of calibration voltages is not less than twice the minimum quantization voltage step size, the linear slope and intercept are calculated based on the two sets of pressure and voltage data, and the calculated slope and intercept are used as the linear calibration coefficients of the pressure acquisition unit at the corresponding measuring point. Linear calibration is performed on the vibration velocity acquisition unit at each measuring point: Two sets of known vibration velocity points are selected within the vibration velocity range. The original voltage values ​​of the vibration velocity acquisition unit are collected at the two sets of vibration velocity points respectively. The upper limit of the full-scale voltage of the vibration velocity channel and the number of bits of the analog-to-digital converter of the vibration velocity channel are obtained. The upper limit of the full-scale voltage of the vibration velocity channel is divided by a power of two to obtain the minimum quantization voltage step size of the vibration velocity channel. The power of two is one less than the number of bits of the analog-to-digital converter of the vibration velocity channel. Under the condition that the difference between the two sets of calibration voltages is not less than twice the minimum quantization voltage step size, the linear slope and intercept of the vibration velocity are calculated based on the two sets of vibration velocity and voltage data. The calculated slope and intercept are used as the linear calibration coefficients of the vibration velocity acquisition unit at the corresponding measuring point. At each measuring point and at each sampling time, the original voltage of the temperature acquisition unit is multiplied by the corresponding temperature linear slope and the corresponding temperature intercept is added; the original voltage of the pressure acquisition unit is multiplied by the corresponding pressure linear slope and the corresponding pressure intercept is added; and the original voltage of the vibration velocity acquisition unit is multiplied by the corresponding vibration velocity linear slope and the corresponding vibration velocity intercept is added to obtain the true temperature value, true pressure value, and true vibration velocity value of the corresponding measuring point at the corresponding sampling time. The lower and upper limits of the range of the physical quantity of temperature, the lower and upper limits of the range of the physical quantity of pressure, and the lower and upper limits of the range of the physical quantity of vibration velocity are obtained respectively, and the value ranges of the three physical quantities are formed accordingly. At each measuring point and at each sampling time, the actual temperature, pressure, and vibration velocity values ​​are standardized: the actual value is subtracted from the corresponding lower limit of the range, and then divided by the difference between the corresponding upper and lower limits of the range to obtain a standardized value between zero and one; when the standardized value is less than zero, it is taken as zero, and when the standardized value is greater than one, it is taken as one, thus forming standardized data sequences for temperature, pressure, and vibration velocity respectively.

3. The survey data integration and analysis method based on multi-source data fusion according to claim 2, characterized in that, The process of constructing a three-dimensional normalized observation vector and a corresponding three-dimensional state estimation vector composed of normalized observations of temperature, pressure, and vibration velocity under a unified time index, and setting an initial state estimation vector and an initial error covariance matrix, specifically includes: At each measuring point and at each sampling time, the standardized temperature value, standardized pressure value, and standardized vibration velocity value of the corresponding measuring point at the corresponding sampling time are arranged in a fixed order of temperature, pressure, and vibration velocity to form a three-dimensional standardized observation vector. A three-dimensional state estimation vector is constructed at each measuring point and at each sampling time. The three components are the normalized true estimate of temperature, the normalized true estimate of pressure, and the normalized true estimate of vibration velocity, respectively. At the start of the survey, the three-dimensional standardized observation vector of each measuring point at the initial sampling time is directly used as the initial state estimation vector of the same measuring point, and the continuous time corresponding to the initial sampling time is recorded. At the start of the survey, an initial error covariance matrix is ​​established for each measuring point. The initial error covariance matrix is ​​a three-row, three-column diagonal matrix, with diagonal elements equal to one and off-diagonal elements equal to zero.

4. The survey data integration and analysis method based on multi-source data fusion according to claim 3, characterized in that, The method employs a constant-preserving discrete prediction model to perform time-based recursion on the three-dimensional state estimation vector and error covariance matrix of the previous sampling time, forming the state prediction vector and error covariance prediction matrix of the current sampling time. Specifically, this includes: At any non-initial sampling time, for each measurement point, the three-dimensional state estimation vector updated at the previous sampling time is directly used as the state prediction vector at the current sampling time, so that the state between adjacent sampling times is kept constant and the discrete prediction model is recursively applied. At any non-initial sampling time, for each measurement point, the error covariance matrix updated at the previous sampling time is directly used as the error covariance prediction matrix at the current sampling time.

5. The survey data integration and analysis method based on multi-source data fusion according to claim 4, characterized in that, The process of constructing an observation residual vector at the current sampling time, performing multi-point statistical analysis on the observation data within a static time window, obtaining the effective observation variance of each physical quantity, constructing an observation noise covariance matrix, and calculating the fusion gain matrix accordingly includes: At each measuring point and at each sampling time, the observation residual vector is calculated dimension by dimension by subtracting the corresponding state prediction vector from the three-dimensional standardized observation vector. Set the observation matrix as a three-row, three-column identity matrix so that each dimension of observation corresponds directly to only the same dimension of state component; A static time window is selected when the system is in a steady state. The static time window is determined by the start sampling number and the end sampling number. Within the static time window, the three-dimensional standardized observation vectors of all measuring points and all sampling times are organized: the temperature dimension forms the first-dimensional sample set, the pressure dimension forms the second-dimensional sample set, and the vibration velocity dimension forms the third-dimensional sample set. The arithmetic mean was calculated for the sample sets of temperature, pressure and vibration velocity respectively, and the global average value of the standardized observations in each dimension was obtained within the static time window. For the three dimensions of temperature, pressure, and vibration velocity, the variance of each sample value is squared and summed with the global mean of the corresponding dimension, and then divided by the total number of samples to obtain the total variance of the standardized observations in each dimension within the static time window. Based on the linear calibration slope of temperature, pressure and vibration velocity, and the minimum quantization voltage step size of the temperature channel, pressure channel and vibration velocity channel, the minimum quantization step size of the physical quantity scale is mapped to the normalization scale according to the corresponding slope. The larger of the normalized quantization step sizes corresponding to all measurement points in the three dimensions is selected as the minimum effective normalized quantization step size of temperature, pressure and vibration velocity. The total variance of the static time window is compared with the lower bound of the observation noise variance derived from the standardized quantization step size in three dimensions, and the larger value in each dimension is taken as the effective observation variance. Construct an observation noise covariance matrix based on the effective observation variance in three dimensions. The diagonal elements of the matrix are the effective observation variances in the three dimensions, and the off-diagonal elements are zero. At each measurement point and at each sampling time, the three-row, three-column fusion gain matrix is ​​calculated based on the recursive least squares fusion principle, using the observation noise covariance matrix and the corresponding measurement point error covariance prediction matrix at the current sampling time as input.

6. The survey data integration and analysis method based on multi-source data fusion according to claim 5, characterized in that, The process of using the fusion gain matrix to perform a linear update on the state prediction vector to obtain an unlimited 3D fusion state estimation vector, and simultaneously correcting the error covariance matrix, specifically includes: At each measurement point and at each sampling time, the fusion gain matrix of the corresponding measurement point at the current sampling time is multiplied by the corresponding observation residual vector to obtain a three-dimensional increment vector. The three-dimensional increment vector is then added to the state prediction vector to form an unlimited three-dimensional fusion state estimation vector. At each measurement point and at each sampling time, the error covariance prediction matrix is ​​multiplied by subtracting the fusion gain matrix from the three-row, three-column identity matrix as the left and right coefficients. At the same time, the weighted matrix terms are calculated using the fusion gain matrix and the observation noise covariance matrix as coefficients. The two results are added together to obtain the corrected error covariance matrix.

7. The survey data integration and analysis method based on multi-source data fusion according to claim 6, characterized in that, The process involves constructing anomaly upper limits for each physical quantity based on observational statistics within a static time window, and then applying zero lower bound and anomaly upper bound values ​​to the unconstrained 3D fusion state estimation vector dimension-by-dimensional constraint to obtain the 3D constrained fusion state estimation vector. Specifically, this includes: Within the static time window, a dimensional maximum search is performed on the standardized observations at all measuring points and at all sampling times in the three dimensions of temperature, pressure, and vibration velocity to obtain the global maximum observations in the three dimensions. Based on the effective observation variances of the three dimensions, the square root of the effective observation variance of each dimension is calculated to obtain the effective observation standard deviations of the three dimensions. The corresponding global maximum observation value is added to three times the effective observation standard deviation in each of the three dimensions to obtain the original anomaly upper limit value. The value is then restricted to between zero and one according to the allowed range of the standardized physical quantity to form the final anomaly upper limit value. At any measurement point, at any sampling time, and in any dimension, the range of each component of the unlimited 3D fusion state estimation vector is determined: when the component value is less than zero or greater than the upper limit of the corresponding dimension, the anomaly marker value is set to one; when the component value is between zero and the upper limit of the corresponding dimension, the anomaly marker value is set to zero. At any measurement point, at any sampling time, and in any dimension, amplitude limiting is performed on each component of the unlimited 3D fusion state estimation vector: component values ​​less than zero are replaced with zero, component values ​​greater than the corresponding dimension's abnormal upper limit value are replaced with the corresponding dimension's abnormal upper limit value, and component values ​​between zero and the abnormal upper limit value remain unchanged. At each measurement point and at each sampling time, the fused estimation components of the temperature, pressure, and vibration velocity dimensions are recombined in a fixed order to form a three-dimensional fused state estimation vector after limiting.

8. The survey data integration and analysis method based on multi-source data fusion according to claim 7, characterized in that, The process involves spatially averaging the three-dimensional amplitude-limited fusion state estimation vectors of all measurement points at each sampling time along three dimensions: temperature, pressure, and vibration velocity. This yields a spatially averaged fusion result after removing outlier measurement points. When no valid samples are available, a backtracking spatially averaged fusion result is used. Specifically, this includes: At each sampling time, the three-dimensional fused state estimation vectors of all measuring points at the current sampling time are arranged in order of measuring point number to form a state matrix with measuring points as rows and temperature, pressure and vibration velocity as columns. At each sampling time, the number of measurement points with anomaly marker values ​​of zero is counted for each of the three dimensions to obtain the number of valid samples in the three dimensions at the current sampling time. When the number of valid samples in a certain dimension is greater than zero, the arithmetic mean of the components of the three-dimensional amplitude-limited fusion state estimation vector of all measurement points with anomaly label values ​​of zero is taken in the same dimension to obtain the spatial average fusion result in the same dimension at the current sampling time. When the number of effective samples in a certain dimension is zero, the arithmetic mean of the components of the three-dimensional fused state estimation vector of all measurement points in the same dimension is calculated, and the calculated average value is used as the back-off space average fusion result in the same dimension at the current sampling time.

9. The survey data integration and analysis method based on multi-source data fusion according to claim 8, characterized in that, The process involves constructing a fusion output sequence based on the spatial average fusion results at each sampling time throughout the entire survey period, performing full-cycle statistical analysis on each physical quantity, and generating a survey data integration report that includes the fusion output sequence and full-cycle statistical indicators. Specifically, this includes: At each sampling moment, the spatial average fusion results of temperature, pressure and vibration velocity dimensions are combined in a fixed order to form a three-dimensional multi-source fusion output vector for the entire survey object; From the initial sampling time up to the sampling time corresponding to the maximum sampling number, the three-dimensional multi-source fusion output vectors of all sampling times are collected throughout the entire survey period, forming a fusion output sequence arranged in chronological order; Throughout the entire research period, extreme values ​​and mean values ​​were calculated for the corresponding spatial average fusion result sequences for each of the three dimensions, resulting in the maximum, minimum, and average values ​​for each dimension, forming full-cycle statistical indicators for the three dimensions. Based on the fused output sequence and the full-cycle statistical indicators of three dimensions, a survey data integration report is generated. The report includes at least the fused output data in time series form and the maximum, minimum and average values ​​of each physical quantity.

10. A system employing the survey data integration and analysis method based on multi-source data fusion as described in claim 9, characterized in that, include: The multi-source data acquisition module performs the deployment of temperature acquisition units, pressure acquisition units, and vibration velocity acquisition units at the measurement points in the survey area, voltage signal sampling, linear calibration, and range standardization processing, and outputs standardized observation data of temperature, pressure, and vibration velocity. The fusion estimation module performs three-dimensional normalized observation vector construction, state prediction, observation residual vector construction, observation noise covariance matrix calculation, fusion gain matrix calculation, state recursive update and error covariance correction, and outputs unlimited and limited three-dimensional fusion state estimation vectors and anomaly labeling results. The multi-point fusion integration module performs the following tasks: constructing the state matrix of the fusion state estimation vector after multi-point three-dimensional amplitude limiting, calculating the number of effective samples in each dimension, calculating the spatial average fusion result, and calculating the back-off spatial average fusion result, and outputting the spatial average fusion result in three dimensions. The results output module receives the spatial average fusion results, constructs the fusion output sequence, performs full-cycle statistical index calculations, and generates and stores a survey data integration report containing the fusion output sequence and full-cycle statistical indicators.

Images