Multi-source data real-time fusion processing method and system of mobile intelligent device

Abstract

The invention provides a multi-source data real-time fusion processing method and system for a mobile intelligent device, and relates to the technical field of data processing.The method comprises the steps that 1, multi-dimensional original data streams are collected in real time through a heterogeneous sensor array integrated by the mobile intelligent device, data streams of different sensors are aligned by applying a space-time synchronization mechanism, and the data streams of different sensors are obtained; generating an original data set with consistent time and space; 2, dynamic interpolation compensation operation is executed on the original data set, and a dynamic calibration framework is constructed based on the internal topological relation of the data flow to form a dynamic sensing domain; and generating an evolution sequence according to the data unit evolution behavior of the domain boundary, generating a space correction value through the evolution sequence and the offset feature of the preset reference, and generating preprocessed data fused with the space correction value in combination with real-time data correlation analysis. According to the method, dynamic adjustment is triggered through anomaly detection, the fusion parameters are updated through the sliding window, and real-time efficient fusion processing of multi-source data of the mobile intelligent device is achieved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method and system for real-time fusion processing of multi-source data of a mobile intelligent device. Background Art

[0002] In smart homes, traditional technologies have some defects when performing real-time fusion processing of multi-source data collected by mobile smart devices. Mobile smart devices in smart homes are usually equipped with multiple sensors, such as infrared sensors, cameras, temperature and humidity sensors, etc. For example, when an intelligent security patrol robot is patrolling indoors, the infrared sensor detects human heat source information, and the camera captures the image of the person at the same time. When traditional technologies fuse these two types of data, they may not be able to quickly establish an effective association due to differences in data format and feature dimensions. This compatibility issue makes it difficult for multi-source data to fully exert their synergistic effect, affecting the accuracy of scene status judgment.

[0003] In addition, traditional technologies have shortcomings in real-time data processing and dynamic adjustment. Mobile smart devices need to adjust their operating strategies in a timely manner according to real-time data. During the cleaning process of the smart sweeping robot, the ultrasonic sensor detects low furniture in front, and the environmental map drawn by the lidar shows that there is a certain amount of space in the area for detour. Traditional data fusion processing technology may not be able to efficiently fuse and calculate these two types of real-time data in a short period of time, resulting in the robot being stuck or making misjudgments when adjusting its route, either colliding with furniture or detouring unnecessary distances, reducing cleaning efficiency. This lack of real-time performance makes it difficult for mobile smart devices to flexibly adapt to the dynamically changing home environment. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a real-time fusion processing method and system for multi-source data of mobile intelligent devices, which improves the fusion processing of multi-source heterogeneous data through spatiotemporal synchronization calibration, dynamic reliability assessment, adaptive weighted fusion and edge layered processing.

[0005] In order to solve the above technical problems, the technical solutions of the present invention are as follows: In a first aspect, a method for real-time fusion processing of multi-source data of a mobile intelligent device includes: Step 1: Through the heterogeneous sensor array integrated in the mobile intelligent device, multi-dimensional raw data streams are collected in real time, and the data streams of different sensors are aligned using a spatiotemporal synchronization mechanism to generate a spatiotemporally consistent raw data set; Step 2: Perform dynamic interpolation and compensation on the original dataset, and construct a dynamic calibration framework based on the intrinsic topological relationship of the data stream to form a dynamic perception domain. Generate an evolution sequence based on the evolution behavior of the data units at the domain boundary, and generate spatial correction values ​​based on the offset characteristics of the evolution sequence and the preset benchmark. Combined with real-time data correlation analysis, generate preprocessed data that integrates the spatial correction values. Step 3: Perform multi-dimensional feature extraction on the pre-processed data, evaluate the confidence quality of each data source in real time, and generate a feature set with real-time confidence evaluation; Step 4: Based on the feature set with real-time confidence evaluation, the fusion weights of each data source are dynamically assigned through an adaptive weighting strategy. In combination with the layered processing mechanism of the edge computing architecture, real-time fusion calculations are performed to generate low-latency fusion results. Step 5: Send the low-latency fusion results to the edge node via a lightweight transmission protocol, where a real-time decision algorithm is executed to generate control instructions and feed them back to the mobile smart device. Step 6: Based on the generated full-process data, through real-time anomaly detection and device computing power adaptation, dynamically adjust the fusion strategy parameters, and allocate computing resources to key functional data processing links to achieve real-time fusion processing of multi-source data.

[0006] Secondly, a real-time fusion processing system for multi-source data of a mobile intelligent device includes: The data acquisition module is used to collect multi-dimensional raw data streams in real time through a heterogeneous sensor array integrated in a mobile intelligent device, and to align the data streams of different sensors using a spatiotemporal synchronization mechanism to generate a spatiotemporally consistent raw data set; The dynamic processing module is used to perform dynamic interpolation and compensation operations on the original data set, and build a dynamic calibration framework based on the intrinsic topological relationship of the data stream to form a dynamic perception domain. It generates an evolution sequence based on the evolution behavior of the data units at the domain boundary, generates spatial correction values ​​based on the offset characteristics of the evolution sequence and the preset benchmark, and combines real-time data correlation analysis to generate preprocessed data that integrates the spatial correction values; The feature evaluation module is used to extract multi-dimensional features from pre-processed data, evaluate the confidence quality of each data source in real time, and generate a feature set with real-time confidence evaluation; The fusion computing module is used to dynamically allocate the fusion weights of each data source through an adaptive weighting strategy based on a feature set with real-time confidence evaluation, and to perform real-time fusion calculations in combination with the layered processing mechanism of the edge computing architecture to generate low-latency fusion results; The command feedback module is used to send the low-latency fusion results to the edge node via a lightweight transmission protocol, execute the real-time decision algorithm at the node to generate control commands and feed them back to the mobile intelligent device; The resource scheduling module is used to dynamically adjust the fusion strategy parameters based on the generated full-process data through real-time anomaly detection and device computing power adaptation, and allocate computing resources to key functional data processing links to achieve efficient fusion processing of multi-source data.

[0007] According to a third aspect, a computing device includes: one or more processors; The storage device is used to store one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method.

[0008] In a fourth aspect, a computer-readable storage medium stores a program, which implements the method when executed by a processor.

[0009] The above solution of the present invention includes at least the following beneficial effects: Heterogeneous sensor data are aligned through a spatiotemporal synchronization mechanism, and data missing areas are filled in with dynamic interpolation compensation operations. Data offsets are then calibrated using spatial correction values, reducing errors and discontinuities in the original data acquisition process, making preprocessed data more relevant to actual scenarios, and achieving low latency and real-time data fusion. A layered processing mechanism based on an edge computing architecture is used, combined with an adaptive weighting strategy to dynamically allocate fusion weights, accelerating the calculation speed while ensuring fusion accuracy. The application of lightweight transmission protocols reduces data transmission time, ensuring that fusion results can be quickly converted into control instructions and fed back to the device, meeting the core requirements of mobile smart devices for real-time response. By evaluating the confidence quality of each data source in real time, dynamically changing confidence labels are assigned to different sensor data, enabling the fusion process to rely on high-quality data and reducing the impact of single sensor failures or data anomalies on the overall results. At the same time, based on the real-time anomaly detection and computing power adaptation mechanism of the entire process data, the fusion strategy parameters can be dynamically adjusted to maintain stable operation in complex environments or when the device status changes; by dynamically adjusting the fusion strategy parameters, computing resources are accurately allocated to key functional data processing links, avoiding resource waste and improving the utilization efficiency of the device computing power. The layered processing mechanism of the edge computing architecture reduces data transmission and redundant calculations, while reducing the energy consumption of the device and extending the battery life of the mobile intelligent device; processing multi-dimensional data generated by different types of sensors in complex dynamic environments, through the construction of dynamic perception domains and evolutionary sequence analysis, the device's perception of environmental changes is more delicate and accurate. The combination of real-time decision-making algorithms and control command feedback mechanisms allows the device to quickly respond to various situations in the environment, improving the reliability and practicality of autonomous navigation, intelligent inspection, dynamic environment interaction, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1It is a flowchart of a method for real-time fusion processing of multi-source data of a mobile intelligent device provided by an embodiment of the present invention.

[0011] Figure 2 Schematic diagram of a multi-source data real-time fusion processing system for a mobile intelligent device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0012] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0013] like Figure 1 As shown, an embodiment of the present invention provides a method for real-time fusion processing of multi-source data of a mobile smart device, the method comprising the following steps: Step 1: Through the heterogeneous sensor array integrated in the mobile intelligent device, multi-dimensional raw data streams are collected in real time, and the data streams of different sensors are aligned using a spatiotemporal synchronization mechanism to generate a spatiotemporally consistent raw data set; Step 2: Perform dynamic interpolation and compensation on the original dataset, and construct a dynamic calibration framework based on the intrinsic topological relationship of the data stream to form a dynamic perception domain. Generate an evolution sequence based on the evolution behavior of the data units at the domain boundary, and generate spatial correction values ​​based on the offset characteristics of the evolution sequence and the preset benchmark. Combined with real-time data correlation analysis, generate preprocessed data that integrates the spatial correction values. Step 3: Perform multi-dimensional feature extraction on the pre-processed data, evaluate the confidence quality of each data source in real time, and generate a feature set with real-time confidence evaluation; Step 4: Based on the feature set with real-time confidence evaluation, the fusion weights of each data source are dynamically assigned through an adaptive weighting strategy. In combination with the layered processing mechanism of the edge computing architecture, real-time fusion calculations are performed to generate low-latency fusion results. Step 5: Send the low-latency fusion results to the edge node via a lightweight transmission protocol, where a real-time decision algorithm is executed to generate control instructions and feed them back to the mobile smart device. Step 6: Based on the generated full-process data, through real-time anomaly detection and device computing power adaptation, dynamically adjust the fusion strategy parameters, and allocate computing resources to key functional data processing links to achieve real-time fusion processing of multi-source data.

[0014] In an embodiment of the present invention, heterogeneous sensor data streams are aligned through a spatiotemporal synchronization mechanism, which solves the time misalignment problem caused by different sampling rates and transmission delays, as well as the spatial coordinate inconsistency problem caused by differences in sensor installation positions, thereby improving the spatiotemporal matching accuracy of multi-source data. The dynamic interpolation compensation operation can intelligently fill in data missing or abnormal areas. Combined with a dynamic calibration framework based on topological relationships, it can capture the inherent structural characteristics of the data stream in real time, generate spatial correction values ​​by analyzing the evolutionary behavior of domain boundary data units, and achieve accurate correction of data offset in dynamic environments, thereby enhancing data integrity and reliability. The multidimensional feature extraction and real-time credibility assessment mechanism can comprehensively analyze the characteristic performance of each data source in the time domain, frequency domain and statistical domain, and quantify data quality through indicators such as entropy value change rate and spectral energy distribution. This refined evaluation system can dynamically identify and reduce the impact of low-quality data, thereby improving the accuracy of overall data fusion.

[0015] The adaptive weighted strategy dynamically allocates fusion weights according to data reliability, and combines the layered processing mechanism of the edge computing architecture to reasonably distribute computing tasks between terminals and edge nodes. This collaborative processing mode reduces data transmission overhead and computing redundancy, achieves millisecond-level fusion response speed, and meets the real-time decision-making needs of mobile smart devices; the lightweight transmission protocol ensures that the fusion results are efficiently transmitted to the edge nodes, and the real-time decision-making algorithm quickly generates control instructions based on environmental characteristic parameters, and ensures reliable execution of instructions through a retransmission mechanism with conflict detection, enabling the device to respond quickly in complex environments, improving the stability of core applications such as autonomous navigation and target tracking. The real-time anomaly detection mechanism based on full-process data can promptly detect data fluctuations or system failures, and dynamically adjust fusion strategy parameters and computing power allocation.

[0016] In a preferred embodiment of the present invention, step 1, which involves collecting multi-dimensional raw data streams in real time using a heterogeneous sensor array integrated in a mobile smart device and applying a spatiotemporal synchronization mechanism to align the data streams of different sensors to generate a spatiotemporally consistent raw data set, may include: In an embodiment of the present invention, during the data collection phase, after the mobile intelligent device is started, the integrated heterogeneous sensor array is first activated, including but not limited to lidar, camera, inertial measurement unit, ultrasonic sensor, etc. Each sensor continuously collects data according to its own preset sampling frequency. For example, the lidar collects environmental point cloud data at a frequency of 10 times per second, the camera captures image data at a rate of 30 frames per second, and the inertial measurement unit records acceleration and angular velocity data at a frequency of 100 times per second. The system allocates independent cache space for each sensor data stream, stores the original data frames in real time, and attaches an acquisition timestamp accurate to the millisecond level to each data frame. The timestamp is generated based on the high-precision clock built into the device to ensure the uniformity of time recording.

[0017] During the time synchronization process, the timestamps of all data frames in each sensor's data stream are first extracted, and a comparison table containing the sensor ID, data frame sequence number, and timestamp is established. By analyzing the timestamp distribution of different sensors near the same time in the comparison table, the average time offset of each sensor is calculated—that is, the average difference between the timestamp of a sensor's data frame and the device's reference clock. For sensors with large differences in sampling frequency, a sliding window method is used for time alignment. Using the time axis of the sensor with the highest sampling frequency as the reference, the data streams of other sensors are divided into time windows. Within each window, data points are supplemented through linear interpolation so that different sensors have corresponding data records at the same time node. At the same time, time synchronization errors are monitored in real time. The deviation between the current timestamp of each sensor and the reference time is calculated every 100 milliseconds. If the deviation exceeds 5 milliseconds, the time offset is recalculated and the interpolation parameters are adjusted to ensure that the time synchronization accuracy is maintained within the millisecond level.

[0018] Spatial alignment is based on the physical installation parameters of the sensors. These parameters include the three-dimensional coordinates (X, Y, and Z axis positions) and installation attitude angles (pitch, roll, and yaw) of each sensor in the device coordinate system. A spatial conversion mechanism is constructed based on these parameters. For the raw data collected by each sensor, a coordinate translation operation is first performed to eliminate the installation position offset. The data point coordinates in the sensor's local coordinate system are added to the installation position coordinates and converted to the origin of the device coordinate system. The rotation matrix corresponding to the attitude angle is then used to correct the directional deviation caused by the difference in installation angle, so that the data directions of all sensors are uniformly pointed to the device's motion coordinate system. For visual sensors, distortion correction is also required through calibration parameters to convert the image pixel coordinates into actual physical space coordinates to ensure consistency with the spatial scale of other sensors.

[0019] The spatiotemporal alignment verification phase adopts a double verification mechanism: in the time dimension, 10 groups of multi-sensor data frames at different times are randomly selected, and the standard deviation of the timestamp difference is calculated. If the standard deviation is less than 2 milliseconds, the time synchronization is considered to be qualified; in the spatial dimension, fixed reference points around the device are selected, and the coordinate measurement results of different sensors for the same reference point are compared, and the root mean square error of the coordinate deviation is calculated. If the error is less than the preset threshold (such as 5 cm), the spatial alignment is considered to be qualified; if any dimension verification fails, the time offset calculation window size or spatial conversion matrix parameters will be adjusted retroactively, and the alignment operation will be re-executed until the verification passes; the multi-sensor data that have passed the verification will be re-sorted in timestamp order and integrated into a unified data structure. Each time node includes the original data, spatial coordinate information and data integrity mark of each sensor. At the same time, a spatiotemporal alignment report is generated, recording the time offset, spatial conversion parameters and alignment error value of each sensor, and finally forming an original data set with completely consistent spatiotemporal attributes.

[0020] In a preferred embodiment of the present invention, the above step 2, performing a dynamic interpolation compensation operation on the original data set and constructing a dynamic calibration framework based on the intrinsic topological relationship of the data stream to form a dynamic perception domain, may include: Step 220 , performing manifold learning on the original data set that is consistent in time and space, extracting spatial adjacency characteristics and temporal continuity characteristics of data units, and generating a sparse correlation matrix including quantified correlation strength; Step 221 , based on the sparse correlation matrix, calculate the gradient modulus of each row element, detect the mutation position where the gradient modulus exceeds a set threshold, and use the mutation position as the boundary point to form a spatial segmentation of the internal stable region and the external change region; Step 222 , within the external change region, performing strength-weighted interpolation calculation on the data missing region according to the corresponding connection strength values ​​in the sparse correlation matrix to generate a compensated continuous data stream; Step 223: Based on the spatial distribution characteristics of the continuous data stream, a dynamic perception domain of the structured mapping framework is formed.

[0021] In an embodiment of the present invention, all data units are selected from a time-space consistent original data set, and each data unit includes corresponding spatial coordinates, timestamps and feature parameters. In manifold learning analysis, the feature parameters of each data unit are first converted into a high-dimensional vector, and the cosine similarity between any two data unit vectors is calculated. The closer the similarity value is to 1, the closer the association between the two in the feature space is. For spatial adjacent feature extraction, the Euclidean distance between each two data units is calculated based on the three-dimensional spatial coordinates of the data unit. The distance values ​​less than the preset spatial threshold are regarded as potential spatial adjacent units. For temporal continuity features, the continuous data units in the same monitoring area are arranged in ascending order by timestamps, and the absolute difference of the feature parameters of adjacent data units is calculated. The smaller the value, the better the temporal continuity; the association strength is calculated by comprehensively considering the spatial distance and the temporal difference, and the spatial distance weight and the temporal difference weight are set. The spatial distance is normalized and multiplied by the spatial weight, and the temporal difference is normalized and multiplied by the temporal weight. The sum of the two is the initial association strength; the initial association strength is then corrected by the cosine similarity, and finally a quantitative association strength value between 0 and 1 is obtained. The larger the value, the more significant the association; when constructing the matrix, the rows and columns correspond to different data unit numbers, and the matrix cells are filled with the association strength values ​​corresponding to the two data units; for cells with association strength values ​​lower than the minimum valid threshold, they are directly assigned a value of 0, forming a sparse association matrix that only retains strong association relationships, and the proportion of non-zero elements in the matrix does not exceed the preset proportion of the total number of elements.

[0022] For each row of the sparse correlation matrix, the positions and corresponding correlation strength values ​​of all non-zero elements are extracted in column index order. The correlation strength difference of non-zero elements in adjacent columns is calculated. The squared values ​​of each difference are accumulated and then squared to obtain the gradient modulus of the elements in the row. The gradient modulus reflects the overall change in the correlation strength of the data units in the row. The gradient modulus of each row is compared with the preset mutation detection threshold. When the gradient modulus of a row is greater than the threshold, the column position with the most drastic gradient change in the row is further located. The local gradient value of the elements in the adjacent columns of the row is calculated. The local gradient value is the absolute value of the difference between the correlation strengths of the two adjacent columns. The column position with the largest local gradient value is the mutation position. The mutation positions of all rows are collected and marked in the spatial coordinate system. The adjacent mutation positions are connected to form a closed boundary line. The area inside the boundary line contains data units whose corresponding matrix row gradient moduli are all lower than the threshold and the correlation strength between data units changes steadily. This area is delineated as the internal stable area. The area outside the boundary line contains rows whose gradient modulus exceeds the threshold and the correlation strength of data units fluctuates significantly. This area is delineated as the external variation area.

[0023] First, scan all data units in the external change area, and identify missing areas where data values ​​are empty or feature parameter jumps exceed the allowable range by comparing the timestamp intervals and characteristic parameter continuity of the complete data sequence. Record the start and end timestamps and spatial coordinate range of the missing area, and extract the five nearest data units around the missing area from the sparse association matrix (with the closest spatial distance and adjacent timestamps). Obtain the connection strength values ​​of these data units and the center of the missing area, normalize the connection strength values, and divide each connection strength value by the sum of all connection strength values ​​to obtain the weight coefficient of each reference data unit. The sum of the weight coefficients is 1. Collect the valid data values ​​of each reference data unit in the corresponding time period of the missing area, multiply each data value by the corresponding weight coefficient and sum them to obtain the interpolated data value of the first time point of the missing area, and calculate the interpolated data of each time point in the missing area in sequence according to the time interval. For missing areas with a large spatial range, divide them into multiple sub-areas according to the spatial grid, and perform the above weighted interpolation calculation in each sub-area separately. Fill all the interpolated data values ​​into the missing area and splice them with the original valid data to form a temporally continuous and spatially complete data stream. By calculating the smoothness error between the interpolated data and the surrounding valid data, ensure that the error value is lower than the preset standard.

[0024] The compensated continuous data stream is gridded according to spatial coordinates. Each grid cell contains all data values ​​at that spatial location within a set time window. The mean, variance, and distribution density of the data are calculated for each grid cell. The mean reflects the typical data level in the area, the variance reflects the degree of data fluctuation, and the distribution density reflects the intensity of data collection. Grid cells are clustered based on the similarity of their data means. Grid cells with mean differences less than a set threshold are grouped into the same feature region. Each feature region is labeled with its core data characteristics (such as mean range and major change trends). The adjacent relationships and data interaction frequencies between different feature regions are counted. A higher interaction frequency indicates a stronger correlation between regions. Weighted connecting lines represent the strength of the correlation between regions. Using feature regions as basic units, a structured mapping framework is constructed, which includes region boundary coordinates, core feature parameters, and inter-regional correlation relationships. Each element in the framework corresponds to a specific spatial region and its data characteristics. As new data continues to be input, the core parameters and correlation relationships of each feature region are updated in real time. When the change in regional data characteristics exceeds the adjustment threshold, the region boundaries and cluster groupings are dynamically adjusted, forming a dynamic perception domain that can adapt to data changes in real time.

[0025] Manifold learning is used to capture the spatial proximity and temporal continuity of data units. The sparse correlation matrix generated by quantifying the correlation strength filters out irrelevant data interference, retaining only meaningful correlations while clarifying the intrinsic connections between data units. Gradient modulus is used to detect mutation locations and divide regions, automatically identifying stable and changing regions of the data distribution. This avoids the limitations of a unified processing strategy for the entire data. Lightweight processing can be used to save computing power in internal stable regions, while intensive processing is used in external changing regions. This achieves precise allocation of data processing resources and improves overall processing efficiency. Intensity-weighted interpolation calculates the weights of reference data for missing data in external changing regions based on correlation strength, making the interpolation results more consistent with the actual data distribution. Compared with simple linear interpolation, this reduces the error caused by missing data. The resulting dynamic perception domain can reflect the spatial distribution changes of the data in real time. The structured mapping framework clearly presents the data characteristics and correlations of each region, providing a more comprehensive and accurate perception of complex environments. The dynamic adjustment mechanism ensures that the perception domain can adapt to data changes in a timely manner, improving its adaptability to dynamic environments.

[0026] In a preferred embodiment of the present invention, step 2, generating an evolution sequence based on the evolution behavior of data units at the domain boundary, generating a spatial correction value based on the offset characteristics of the evolution sequence and a preset benchmark, and generating preprocessed data that integrates the spatial correction value by combining real-time data correlation analysis, may include: Step 224: Acquire core perception area data units and collect state transition vectors in a continuous time window in real time; Step 225 , based on the state transition vectors and sorted by timestamps, a multi-dimensional time series describing the evolution law of the data units is generated; Step 226 , dynamically time warping the multidimensional time series and the preset reference sequence, and calculating the deviation of the covariance feature of each dimension; Step 227, based on the covariance characteristic deviation, generates a spatial correction coefficient through a preset conversion rule, injects the step spatial correction coefficient into the continuous data stream, and dynamically adjusts the correction amplitude based on the Pearson correlation characteristics of the multi-source sensor data to generate pre-processed data of the fused spatial correction value.

[0027] In an embodiment of the present invention, the core perception area is comprehensively screened from the dynamic perception domain based on the data distribution density, association strength and stability index. First, the ratio of the number of data points of each grid unit to the regional area is calculated to obtain the data distribution density; then the sum of the association strengths of the unit with the adjacent units is counted; finally, the fluctuation range of the data values ​​in the unit in the past 10 time periods is analyzed. The smaller the fluctuation range, the higher the stability. The continuous area that simultaneously meets the conditions of density 1.5 times higher than the average value, the top 20% of the total association strength and the top 30% of the stability index is selected as the core perception area.

[0028] The length of the continuous time window is set to 60 seconds, and the time accuracy is 100 milliseconds, that is, the window contains 600 sampling moments. For each sampling moment, the multidimensional feature parameters of each data unit are extracted from the core perception area, including the value size, change direction, change rate, spatial position and distribution density. The difference in the feature parameters of the same data unit at two adjacent sampling moments is calculated. For example, the value at the current moment is subtracted from the value at the previous moment to obtain the value change, and the position coordinates at the current moment are subtracted from the coordinates at the previous moment to obtain the position offset. These differences are arranged into vector form according to the feature dimensions, such as [value change, direction change angle, rate change value, X-axis offset, Y-axis offset, Z-axis offset, density change rate] to form a state transition vector.

[0029] The validity of all generated state transition vectors is verified, and the modulus of each vector (the square root of the sum of the squares of the values ​​of each dimension) is calculated. If the modulus exceeds the normal range (three times the historical average modulus of this type of data unit), the characteristic parameters of the data unit corresponding to the vector at adjacent moments are checked to see if there is a mutation. If there is a mutation and there is no reasonable explanation for the environmental change, the vector is judged as an outlier and is eliminated; if the mutation is related to environmental changes (such as suddenly entering a high-temperature area), the vector is retained and marked as a special event.

[0030] Collect all state transition vectors that have been verified for validity, extract the precise timestamp (including millisecond information) when each vector is generated, build a time index table, arrange the timestamps in ascending order to form a continuous time axis, and for each state transition vector, find the corresponding position on the time axis according to its timestamp. Split the values ​​of each feature dimension in the vector and fill them into the corresponding positions of the multidimensional time series. For example, the state transition vector [5, 15°, 2.3, 0.2, -0.1, 0, 0.05] corresponds to time point t. Split it into the value of the numerical change dimension at time t being 5, the value of the direction change dimension at time t being 15°, and so on. For missing The time points with missing data (i.e., there is no valid state transition vector at that moment) are filled by the linear interpolation method of the adjacent time points before and after. The difference between the corresponding dimension values ​​of the previous valid time point and the next valid time point is calculated, and the difference is allocated to the missing points according to the time interval ratio; the time series of all feature dimensions are combined together to form a multidimensional time series matrix. Each row of the matrix represents the time change sequence of a feature dimension, and each column represents the multidimensional state of a time point. The matrix is ​​smoothed by using the moving average method. For each dimension value at each time point, the average value of the five time points before and after is calculated as the new value of the point to eliminate the influence of accidental noise.

[0031] Recall the preset benchmark sequence in the system that matches the current scenario. The benchmark sequence is a multidimensional time series of the same type of data units collected under an ideal environment, containing the same feature dimensions and similar length (±10% length difference is allowed). Dynamic time warping (DTW) is performed on the multidimensional time series and the preset benchmark sequence. First, a distance matrix is ​​constructed. Each element of the matrix represents the distance between the feature vectors of a time point in the multidimensional time series and a time point in the benchmark sequence. When calculating the distance, the difference in each dimension is weighted and summed. The weight is pre-set according to the importance of the dimension to the overall data (for example, the weight of the numerical change dimension is 0.3, the weight of the position offset dimension is 0.2, etc.).

[0032] The final path is found in the distance matrix through the dynamic programming algorithm. The starting point of the path is the upper left corner of the matrix and the end point is the lower right corner. The path movement direction can only be right, downward, or right-lower. The cumulative distance of each possible path is calculated, and the path with the smallest cumulative distance is selected as the final alignment path. Along the final path, the time points of the multidimensional time series and the benchmark series are matched one by one to achieve sequence alignment. After alignment, the covariance features are calculated according to the feature dimensions. For each dimension, the average value of the multidimensional time series at all time points in that dimension is first calculated, and then the average value of the benchmark series at all time points in that dimension is calculated. Then, for each time point, the difference between the multidimensional time series value and its own average value, as well as the difference between the benchmark series value and its own average value, are calculated. These two differences are multiplied and the average value of all time points is calculated to obtain the covariance value of the dimension. The standard covariance value of the benchmark series is subtracted from the covariance value of the multidimensional time series to obtain the covariance feature deviation of each dimension.

[0033] For the covariance feature deviation of each feature dimension, a preset piecewise linear transformation rule is applied to generate a spatial correction coefficient. Multiple deviation threshold intervals are set, such as [-∞, -5), [-5, -2), [-2, 2), [2, 5), [5, +∞). Each interval corresponds to a different correction coefficient generation formula. When the deviation falls in the interval [-∞, -5), the correction coefficient = deviation × (-0.3); when it falls in the interval [-5, -2), the correction coefficient = deviation × (-0.2); and so on. The larger the deviation, the larger the absolute value of the correction coefficient, and the correction direction is opposite to the deviation direction. The generated spatial correction coefficient is associated with the corresponding data unit in the continuous data stream according to the feature dimension. For each data point in the data stream, the corresponding correction coefficient is found according to its timestamp and spatial position. The correction coefficient is superimposed on the original value of the data point. For example, if the original value is V and the corresponding correction coefficient is K, the corrected value is V+K. For multi-dimensional data points, correction operations are performed on each dimension separately.

[0034] The Pearson correlation characteristics of multi-source sensor data are calculated simultaneously. For the same feature dimension, the Pearson correlation coefficient between data collected from different sensors is calculated. First, the average value of the two sensor data series is calculated. Then, for each time point, the difference between the two series values ​​and their respective average values ​​is calculated. The two differences are multiplied and summed, and then divided by the product of the standard deviations of the two series to obtain the correlation coefficient. The closer the correlation coefficient is to 1, the stronger the linear correlation between the two sensor data; the closer it is to 0, the weaker the correlation; the correction amplitude is dynamically adjusted according to the correlation coefficient, and the correlation coefficient threshold is set, for example, 0.7. When the correlation coefficient is greater than 0.7, the correction coefficient remains unchanged at the original value; when the correlation coefficient is between 0.4-0.7, the correction coefficient is multiplied by the correlation coefficient; when the correlation coefficient is less than 0.4, the correction coefficient is multiplied by 0.4. In this way, sufficient correction is given to sensor data with strong correlation; for sensor data with weak correlation, the correction amplitude is appropriately reduced to avoid excessive correction.

[0035] After the correction coefficients have been injected and the amplitude has been adjusted, the data stream is integrated and the continuity between adjacent data points is checked. If a sudden change occurs (the difference between adjacent data points exceeds 5 times the normal fluctuation range), the sudden change point is smoothed and the weighted average of the three data points before and after the point is calculated as the new value, with the weight decreasing with distance from the point (for example, the weight of the first three points is 0.1, the weight of the first two points is 0.2, and so on). The final result is preprocessed data that is continuous in time, consistent in space, and incorporates spatial correction values.

[0036] By focusing on the core perception area and collecting state transition vectors, the data change characteristics of key areas can be accurately captured. The setting of continuous time windows ensures the temporal integrity of data changes and avoids interference from data in irrelevant areas. The state transition vectors are sorted by time to generate a multidimensional time series, which clearly presents the evolution trajectory of different feature dimensions over time, and changes the change pattern of data units from disorder to order, facilitating comparative analysis with the benchmark sequence. Dynamic time alignment solves the problem of differences in time rhythm between the actual sequence and the benchmark sequence. The calculation of the covariance characteristic deviation quantifies the degree of deviation of the sequence fluctuation characteristics, which can accurately identify anomalies in data stability and provide a clear quantitative basis for correction. The spatial correction coefficient is generated by preset rules and the amplitude is dynamically adjusted to achieve accurate correction of data deviation. It not only ensures that the correction direction is consistent with the deviation direction, but also adjusts the amplitude through the Pearson correlation characteristic to avoid over-correction or under-correction. The preprocessed data that integrates the spatial correction value reduces the data offset caused by environmental interference and sensor error, and improves the accuracy and reliability of the overall data processing.

[0037] In a preferred embodiment of the present invention, the above step 3, performing multi-dimensional feature extraction on the pre-processed data, evaluating the confidence quality of each data source in real time, and generating a feature set with real-time confidence evaluation, may include: Step 330 , based on the pre-processed data stream, separate the time domain, frequency domain and statistical domain feature dimensions through parallel feature extraction channels to generate a multi-dimensional feature set; Step 331: Based on the multidimensional feature set, the entropy change rate is independently calculated for each feature dimension, and based on the entropy stability within the sliding window, a preliminary confidence index is generated for each feature dimension. The spectral energy distribution of each feature dimension is extracted, and the spectral confidence coefficient of each feature dimension is calculated based on the main frequency band energy concentration. Step 332: Based on the preliminary confidence index and the spectral confidence coefficient, a weighted geometric mean is used to fuse them to generate a comprehensive confidence weight for each feature dimension; Step 333: Based on the comprehensive confidence weight, dynamically attenuate the comprehensive confidence weight of the feature dimension associated with the corresponding data source to generate an attenuated comprehensive confidence weight. The comprehensive confidence weight is used as metadata and annotated on the corresponding feature vector to form a feature set with weight annotation. Step 334 , based on the weighted annotated feature set, analyze the mutual exclusivity between feature dimensions. When conflicting feature dimensions are detected, dynamically rebalance the annotated confidence weights to generate a feature set with real-time confidence evaluation.

[0038] In an embodiment of the present invention, the preprocessed data stream is classified according to sensor type and data acquisition frequency and assigned to three independent parallel feature extraction channels. Each channel processes data simultaneously without interfering with each other to ensure feature extraction efficiency. The continuous preprocessed data stream is first divided into multiple data segments at fixed time intervals (such as 1 second). Each data segment contains all data points within the time period. The mean is calculated for each data segment. That is, all data values ​​in the segment are added and then divided by the number of data points to obtain a mean feature reflecting the average level of the data in the time period. The maximum value in the data segment is then found as the peak feature, and the minimum value as the valley feature. The difference between the two is the fluctuation range of the data in the segment. When calculating the variance, the difference between each data value and the mean is first calculated, the squared differences are added, and then divided by the number of data points. The larger the variance, the more severe the data fluctuation. At the same time, the numerical difference between two adjacent data points is calculated and divided by the corresponding time interval to obtain a change rate feature used to reflect how fast the data changes over time.

[0039] The frequency domain feature extraction channel first divides the preprocessed data stream into windows of 5 seconds. The number of data points in each window is determined by the sampling frequency (for example, if the sampling frequency is 10 Hz, each window has 50 data points). The data in each window is periodically analyzed, and the pattern of repeated statistical values ​​is observed to determine the main fluctuation frequency. The amplitude of data fluctuation at each frequency is calculated, and the energy value of the frequency is obtained by squaring the amplitude. The total energy of all frequencies is counted, and the frequency interval with an energy proportion exceeding 50% of the total energy is found as the main frequency band. The ratio of the total energy in the main frequency band to the total energy is calculated to obtain the energy proportion of the main frequency band. At the same time, the standard deviation of the energy value of each frequency in the main frequency band is calculated, and then divided by the mean energy of the main frequency band to obtain the uniformity of energy distribution. The lower the uniformity, the more concentrated the energy is on a few frequencies.

[0040] The statistical domain feature extraction channel performs an overall distribution analysis on the preprocessed data stream. After sorting all data points by size, the data value in the middle position is taken as the median to reflect the medium level of the data. The upper quartile value (the data value at the 75% position after sorting) and the lower quartile value (the data value at the 25% position after sorting) are calculated. The difference between the two is the interquartile range, which is used to describe the degree of dispersion of the data. When calculating skewness, the cube of the difference between each data value and the mean is first calculated, and the average of these cubes is obtained to obtain the third-order central moment, which is then divided by the cube of the standard deviation. A positive skewness indicates that the data distribution is right-skewed, and a negative one indicates left-skewed. The kurtosis is calculated by dividing the fourth-order central moment (the average of the fourth power of the difference between the data value and the mean) by the fourth power of the standard deviation. A kurtosis greater than 3 indicates a steeper data distribution, and a kurtosis less than 3 indicates a flatter distribution. Finally, all features extracted by the three channels are classified and organized according to the time domain, frequency domain, and statistical domain to form a multidimensional feature set containing multiple feature parameters.

[0041] For each feature dimension in the multidimensional feature set, the eigenvalues ​​of all data points under this dimension are collected, the number of occurrences of each eigenvalue is counted, and the probability of occurrence of each eigenvalue is obtained by dividing the number of occurrences by the total number of data points, and the entropy value of this feature dimension is obtained. The higher the entropy value, the more dispersed the eigenvalue distribution is. A sliding window with a duration of 10 seconds is set, and the window moves for 2 seconds each time. Each window contains the entropy values ​​of 5 data segments, and the standard deviation of the entropy values ​​in the window is calculated. The smaller the standard deviation, the smaller the change in entropy value in the window, that is, the better the stability of the entropy value. The entropy change rate is calculated by subtracting the average entropy value of the previous window from the average entropy value of the current window, and then dividing it by the time interval between the two windows (2 seconds) to obtain the entropy change per unit time.

[0042] When generating the preliminary confidence index, the following standards are used: when the standard deviation of the entropy stability in the sliding window is less than 0.1 and the absolute value of the entropy change rate is less than 0.05, the preliminary confidence index is 0.9-1.0; when the stability standard deviation is between 0.1-0.3, or the absolute value of the entropy change rate is between 0.05-0.1, the index is 0.6-0.8; when the stability standard deviation is greater than 0.3, or the absolute value of the entropy change rate is greater than 0.1, the index is 0.3-0.5. For the newly enabled feature dimension, the preliminary confidence index in the first 30 seconds is temporarily set at 0.7. After accumulating enough data, it is calculated according to the above standards. When extracting the spectral energy distribution, for For the frequency domain feature dimension, the energy value of each frequency point is recorded, and the total energy of the main frequency band (the frequency interval whose energy accounts for more than 50% of the total energy) is calculated. The main frequency band energy concentration is the ratio of this sum to the total energy of all frequencies. When calculating the spectrum confidence coefficient, if the concentration is greater than 0.7, the coefficient is 0.8-1.0; if the concentration is between 0.5-0.7, the coefficient is 0.5-0.7; if the concentration is less than 0.5, the coefficient is 0.2-0.4. For feature dimensions without obvious main frequency bands, the energy fluctuation amplitude (the difference between the maximum energy value and the minimum energy value) is calculated. If the fluctuation amplitude is less than 10% of the total energy, the coefficient is 0.6-0.8, otherwise it is 0.2-0.5.

[0043] The weight distribution ratio of the preliminary confidence index and the spectrum confidence coefficient is determined according to the type of feature dimension. For the time domain feature dimension, since the entropy value change rate can better reflect its temporal stability, the weight of the preliminary confidence index is set to 0.6 and the weight of the spectrum confidence coefficient is set to 0.4; in the frequency domain feature dimension, the spectrum energy distribution is the core feature, so the weight of the two is 0.5 each; the statistical domain feature dimension is greatly affected by the spectrum characteristics, so the weight of the preliminary confidence index is set to 0.4 and the weight of the spectrum confidence coefficient is set to 0.6. When performing weighted geometric mean fusion calculations, first raise the preliminary confidence index to the power of the corresponding weight. For example, if the preliminary confidence index of the time domain feature is 0.8 and the weight is 0.6, then 0.8 raised to the power of 0.6 is approximately 0.85. Then, raise the spectral confidence coefficient to the power of the corresponding weight. For example, if the spectral confidence coefficient is 0.7 and the weight is 0.4, then 0.7 raised to the power of 0.4 is approximately 0.91. Then, multiply the two results together, 0.85×0.91≈0.77, to obtain the original value of the comprehensive confidence weight of the feature dimension.

[0044] Normalize the original values ​​of all feature dimensions. First, calculate the sum of all original values, and then divide each original value by the sum so that the sum of the normalized comprehensive confidence weights is 1. For example, the original values ​​of the three feature dimensions are 0.77, 0.5, and 0.3, respectively, and the sum is 1.57. After normalization, they are approximately 0.49, 0.32, and 0.19, respectively. The higher the comprehensive confidence weight value, the more reliable the data of the feature dimension is in the current environment.

[0045] Query the abnormal records of the data source corresponding to each feature dimension in the past 5 minutes. Abnormal records include feature values ​​outside the normal range, data transmission interruption, and large deviation from other data sources. Count the number of abnormalities and calculate the abnormal frequency (number of abnormalities ÷ 5 minutes). When the abnormal frequency is 1-3 times per hour (i.e. 0.08-0.25 times within 5 minutes), the attenuation coefficient is set to 0.9; when the abnormal frequency is 3-5 times per hour (0.25-0.42 times within 5 minutes), the attenuation coefficient is set to 0.8; when the abnormal frequency exceeds 5 times per hour (more than 0.42 times within 5 minutes), the attenuation coefficient is set to 0.7; apply attenuation to the comprehensive confidence weight of all feature dimensions associated with the corresponding data source. The attenuated weight = normalized comprehensive confidence weight × attenuation coefficient; for example, if the normalized weight of a feature dimension is 0.49, the corresponding data source abnormal frequency is 4 times per hour, and the attenuation coefficient is 0.8, then the attenuated weight is 0.49 × 0.8 = 0.39. If the data source has no abnormal records in the past 5 minutes, or the abnormal frequency is less than 1 per hour, the attenuation coefficient is 1.0 and the weight remains unchanged.

[0046] The attenuated comprehensive confidence weight is used as metadata and annotated on the corresponding feature vector in the form of a key-value pair. For example, if the feature vector is {"mean": 25, "peak": 30, "variance": 5}, the annotated value is {"mean": 25, "peak": 30, "variance": 5, "confidence_weight": 0.39}. All feature vectors with weight annotations are classified according to data source type and feature dimension. Feature vectors from the same data source are placed in the same subset to form a clearly structured feature set with weight annotations. Each feature vector in the set is clearly accompanied by weight information reflecting its own reliability.

[0047] Calculate the mutual exclusivity index between feature dimensions. For different feature dimensions of the same data source, collect the feature value change trends of the two in 10 consecutive time windows, and use the correlation coefficient of the change trend to measure mutual exclusivity. If the correlation coefficient is less than -0.7, it indicates that the change trends of the two are completely opposite, and they are determined to be mutually exclusive feature dimensions. For similar feature dimensions of different data sources (such as the distance features of two lidars), calculate the feature value difference of 5 consecutive time points. If the average value of the difference exceeds 3 times the normal difference average value and lasts for more than 3 time points, it is determined to be a conflicting feature dimension. When a conflicting feature dimension is detected, Perform dynamic rebalancing adjustments. First, calculate the weight difference between the conflicting parties. For example, if feature A has a weight of 0.39 and feature B has a weight of 0.32, with a difference of 0.07. Reduce the weight of the dimension with the lower weight by 20% of the difference, that is, feature B is reduced by 0.07×20%=0.014, and after adjustment, it is 0.32-0.014=0.306; increase the weight of the dimension with the higher weight by 10% of the difference, that is, feature A is increased by 0.07×10%=0.007, and after adjustment, it is 0.39+0.007=0.397. After the adjustment, the sum of the weights of the two remains unchanged (0.39 + 0.32 = 0.397 + 0.306≈0.70).

[0048] If the conflicting feature dimensions come from different data sources, compare the historical reliability scores of the data sources (number of records without abnormalities divided by the total number of records), prioritize the feature weights of data sources with high reliability scores, and reduce the feature weights of data sources with low scores by 10%. After the adjustment is completed, integrate all feature vectors and the corresponding real-time confidence assessment weights to form a feature set with real-time confidence assessment, ensuring that each feature has a clear reliability identifier.

[0049] Multi-domain features are extracted through parallel channel separation. Each channel focuses on feature mining of a specific dimension, avoiding mutual interference when extracting different types of features. It can more accurately capture the changing trend of data in the time domain, the energy distribution in the frequency domain, and the distribution characteristics in the statistical domain. The comprehensive extraction of multi-domain features has a more three-dimensional understanding of the data and makes up for the limitations of a single feature dimension. The entropy value change rate and spectrum energy concentration are calculated independently for each feature dimension, realizing a refined assessment of data quality. The setting of the sliding window can track the stability changes of the data in real time. The entropy value change rate reflects the fluctuation of the data over time, and the spectrum energy concentration reflects the effectiveness of the frequency domain features. The hierarchical generation of preliminary confidence indicators and spectral confidence coefficients transforms the abstract concept of confidence into quantifiable indicators, providing a unified standard for reliability comparison of different feature dimensions. The weighted geometric mean fusion method fully considers the characteristic differences of different feature types and dynamically adjusts the weight ratio to make the fusion results more consistent with the essential attributes of various features. Time domain features focus on temporal stability, frequency domain features focus on spectral concentration, and statistical domain features take into account multiple aspects of performance. This differentiated fusion strategy improves the accuracy of the comprehensive confidence weights. Normalization ensures the comparability of weights, allowing high-reliability feature dimensions to receive higher attention during the fusion process.

[0050] In a preferred embodiment of the present invention, step 4 above, based on the feature set with real-time confidence evaluation, dynamically assigns the fusion weights of each data source through an adaptive weighting strategy, and performs real-time fusion calculation in combination with the layered processing mechanism of the edge computing architecture to generate a low-latency fusion result, may include: Step 440: sort the data sources according to the comprehensive confidence weights corresponding to the feature sets to generate a data source sorting sequence; Step 441: Dynamically assign weight values ​​corresponding to each data source based on the data source sorting sequence, and receive the weight values ​​and feature sets of each data source at the perception layer of the edge computing architecture, perform feature-level weighted fusion calculations, and generate preliminary fusion results. Step 442 , delivering the preliminary fusion result to the decision layer of the edge computing architecture, where the decision layer performs result-level calculations based on the context information and the fusion result to generate a fusion result; Step 443 , monitoring the calculation delay of the fusion result in real time. When the calculation delay exceeds a preset delay threshold, dynamically adjusting the weight value distribution method to obtain a fusion result that meets the preset delay threshold.

[0051] In an embodiment of the present invention, the comprehensive confidence weight value of each feature vector is extracted from the feature set with real-time confidence assessment, and a correspondence table between feature dimensions and data sources is established to clarify which data source each feature dimension belongs to. For the same data source, the comprehensive confidence weight values ​​of all feature dimensions under it are collected, and the arithmetic mean of these weight values ​​is calculated, that is, all weight values ​​are added together and divided by the number of feature dimensions to obtain the overall reliability score of the data source. For example, data source A contains three feature dimensions: distance, speed, and direction, with weight values ​​of 0.82, 0.78, and 0.85, respectively. The overall reliability score is (0.82 + 0.78 + 0.85) ÷ 3 = 0.82.

[0052] The overall reliability scores of all data sources are arranged in descending order to form a preliminary data source ranking sequence. If two or more data sources have the same overall reliability score (the difference is within 0.02), their stability indicators are further compared. The stability index is calculated by counting the number of data source anomalies in the past 10 minutes. The fewer the number of anomalies, the higher the stability index. The calculation formula is: stability index = 1-(number of anomalies ÷ total number of data transmissions). The closer the index value is to 1, the better the stability. The data source with a higher stability index is ranked first. If the stability index is still the same, the data transmission delay of the data source is compared. The transmission delay is the average time interval from sending data from the data source to receiving data from the edge computing architecture. It is obtained by taking the average of 5 consecutive measurements. The data source with a smaller delay value is ranked higher in the ranking. After the above multi-layer comparison, a data source ranking sequence arranged from high to low according to comprehensive reliability is finally generated. Each position in the sequence corresponds to a data source and its related information.

[0053] According to the sorting sequence of data sources, the exponential decay method is used to dynamically assign weight values. The basic weight value of the data source ranked first is set to 0.4. For each subsequent position in the ranking, the weight value is multiplied by a decay coefficient of 0.8. For example, the weight of the data source ranked first is 0.4, the weight of the second is 0.4×0.8=0.32, the weight of the third is 0.32×0.8=0.256, and the weight of the fourth is 0.256×0.8=0.2048. After assigning weights to all data sources, the sum of all weight values ​​is calculated. If the sum is not equal to 1, it is normalized and each weight value is divided by the sum so that the sum of the normalized weight values ​​is 1 to ensure the rationality of weight distribution.

[0054] The perception layer of the edge computing architecture deploys multiple independent receiving ports, each corresponding to a data source. The perception layer receives feature sets and corresponding weights from each data source in real time. The perception layer multiplies the feature values ​​of different data sources by the weights for the same feature dimension to obtain the weighted contribution of each data source for that feature dimension. For example, for a given feature dimension, if data source A has a feature value of 50 and a weight of 0.4, and data source B has a feature value of 55 and a weight of 0.32, their weighted contributions are 50 × 0.4 = 20 and 55 × 0.32 = 17.6, respectively. The weighted contributions of all data sources for the same feature dimension are summed to obtain the fused value for that dimension. In the example above, the fused value is 20 + 17.6 = 37.6. The above calculation process is repeated for all feature dimensions to generate a feature vector containing the fused values ​​of all feature dimensions. The feature vector is then validated by calculating the deviation of each fused value from the standard value for the same period and conditions. If the deviation exceeds the allowable range (±20% of the standard value), the fused value is marked as suspicious and linearly interpolated using the fused values ​​of two adjacent time windows to ultimately generate a stable and reliable preliminary fusion result.

[0055] The preliminary fusion results are transmitted from the perception layer to the decision layer via a high-speed data bus within the edge computing architecture. Data is compressed and verified during transmission. Compression uses a lightweight algorithm to reduce the amount of data, and verification ensures the integrity of data transmission by adding a checksum. After receiving the preliminary fusion results, the decision layer retrieves the current context information from the system database, including environmental parameters (such as light intensity, temperature, and humidity), task information (such as the type, priority, and progress of the currently executed task), and device status (such as power, computing power utilization, and sensor operating status). The weights of feature dimensions are adjusted based on the environmental parameters in the context information. For example, when the light intensity is above the threshold, the weight of the visual feature dimension is reduced by 15%, and the weight of the infrared feature dimension is increased by 20%. When the temperature is abnormal, the weight of the temperature-related feature dimension is increased by 25%. Based on the task priority, the weight of feature dimensions related to the core objective of the task is increased. For example, in navigation tasks, the weight of the position and direction feature dimensions is increased by 30%, and in obstacle avoidance tasks, the weight of the distance and speed feature dimensions is increased by 35%.

[0056] The adjusted weights are multiplied by the feature dimension fusion values ​​in the preliminary fusion results to obtain weighted fusion values. All weighted fusion values ​​are then normalized so that their sum is 1. The decision layer analyzes the normalized weighted fusion values ​​according to preset decision rules (such as if-else logic and threshold judgment). For example, when the weighted fusion value of a target feature dimension exceeds 0.6, it is determined that the target exists; the target location is determined by combining the fusion value of the location feature dimension, and finally a fusion result containing target information, environmental assessment, status recommendations, etc. is generated.

[0057] A high-precision timer is set at the decision layer to record the time interval from receiving the preliminary fusion results to outputting the final fusion results, accurate to the millisecond level. This time interval is the current calculation delay. The calculation delay is compared with a preset delay threshold (set according to the application scenario, such as 50 milliseconds for real-time navigation scenarios and 100 milliseconds for environmental monitoring scenarios). If the calculation delay is less than or equal to the threshold, the current weight distribution method is maintained. If it exceeds the threshold, the weight adjustment mechanism is triggered to analyze the processing time of each data source. By recording the time from data reception to feature-level fusion completion for each data source at the perception layer, the average processing time for each data source is calculated. Data sources with longer processing times (more than twice the total average processing time) are marked and their priority in the data source ranking sequence is lowered. For example, the top 20% of data sources with the longest processing times will be moved down 2-3 places in the ranking sequence.

[0058] The weight values ​​are reallocated based on the adjusted sorting sequence, increasing the weights of low-cost data sources with a higher ranking and reducing the weights of high-cost data sources with a lower ranking. For example, the weight of the data source originally ranked first is increased from 0.4 to 0.5, the weight of the data source originally ranked second is increased from 0.32 to 0.35, and the weight of the data source originally ranked third is reduced from 0.256 to 0.15, ensuring that the total weight remains 1. The new weight distribution method is sent to the perception layer, which re-performs the feature-level fusion calculation according to the new weights, generates a new preliminary fusion result, and passes it to the decision layer. The decision layer uses the new preliminary fusion result for result-level calculation and continues to monitor the calculation delay. If the delay still exceeds the threshold after adjustment, the above sorting adjustment and weight distribution process is repeated, further increasing the weight of the low-cost data source (by 5%-10% each time) and reducing the weight of the high-cost data source (by 10%-15% each time) until the calculation delay meets the preset threshold requirement, and a fusion result that meets real-time requirements is generated.

[0059] By ranking data sources based on comprehensive confidence weights, the system achieves quantitative assessment and prioritization of data source quality, ensuring that highly reliable data sources receive higher weights in the fusion. Multi-layer comparisons of stability and transmission delay ensure that the ranking results more comprehensively reflect the comprehensive performance of data sources, avoiding the limitations of single-metric ranking. A dynamic weight allocation strategy uses exponential decay to highlight the role of high-priority data sources while also taking into account the contributions of other data sources, achieving reasonable collaboration among data sources. Parallel processing and feature-level weighted fusion at the edge computing perception layer reduce data transmission and processing redundancy, improving the efficiency of fusion computation. A validity verification and correction mechanism further ensures the accuracy of preliminary fusion results, providing high-quality input data for decision-making layer computation. Result-level computation, combined with contextual information, enables fusion results to dynamically adapt to environmental changes and task requirements, enhancing the system's scenario adaptability. Dynamic adjustment of feature weights based on environmental parameters and task priorities ensures the dominance of key features in fusion and improves the match between fusion results and actual application requirements. The application of decision rules transforms complex feature data into intuitive decision information, providing clear guidance for the real-time operation of mobile smart devices and improving the device's intelligence.

[0060] In a preferred embodiment of the present invention, the above step 5, sending the low-latency fusion result to the edge node via a lightweight transmission protocol, executing the real-time decision algorithm at the node to generate control instructions and feeding back to the mobile smart device, may include: Step 550: Based on the environmental feature parameters in the low-latency fusion result, query the preset instruction generation rule library to match the state transition strategy adapted to the current scenario; Step 551 , inputting the state transition strategy into the state transition equation calculation process, and combining it with the current device state parameters to generate the final action sequence of the mobile intelligent device; Step 552: Analyze the execution timestamps of each action in the final action sequence and calculate the urgency level of each action; compress the control instruction data set using a coding algorithm based on timestamp difference based on the difference between the urgency level and the action timestamp; In step 553 , the compressed control instruction data packet is sent via wireless communication, and a retransmission mechanism with conflict detection is adopted at the transport layer to ensure reliable feedback of the instruction to the execution end of the mobile intelligent device.

[0061] In an embodiment of the present invention, environmental feature parameters are extracted from the low-latency fusion results. These parameters include, but are not limited to, the distance, number, and type of obstacles around the mobile intelligent device, the terrain slope and road surface smoothness of the area, and the movement direction and speed of dynamic targets in the environment (such as pedestrians and other mobile devices). These extracted environmental feature parameters are then compared one by one with index conditions in a preset instruction generation rule library. The rule library stores state transition strategy templates corresponding to different combinations of environmental features. For example, when the obstacle distance is less than a preset safety threshold and the number is single, a "slow down and avoid" state transition strategy is applied; when the terrain slope is greater than a certain value, a "reduce driving speed and adjust power output" state transition strategy is applied. During the comparison process, the rule entry with the highest match to the key parameters (such as obstacle distance and terrain slope) among the extracted environmental feature parameters is first selected. The additional parameters under this entry (such as obstacle type and dynamic target speed) are then secondary verified to ensure that all parameters meet the rule entry requirements. Ultimately, a state transition strategy adapted to the current scenario is determined.

[0062] First, the current state parameters of the mobile intelligent device are collected, including the device's real-time driving speed, remaining power, current location coordinates, steering angle, power system output power and other data. Then, the state transition strategy determined in step 550 is decomposed into specific constraints and target parameters. For example, the "slow down and avoid" strategy is decomposed into constraints such as "speed reduced to 50% of the original speed" and "steering angle adjusted to deviate from the obstacle direction by 15 degrees". After that, these constraints and target parameters are input into the state transition equation calculation process and substituted into the current device state parameters. During the calculation process, the speed constraint in the state transition strategy is first combined with the current state parameters. The system calculates the speed difference that needs to be reduced or increased based on the previous driving speed, and determines the speed change within each time unit. It then calculates the steering angle adjustment range and the time required based on the steering constraint and the current steering angle. For the power system output power, it calculates the power adjustment value required to maintain the target speed based on the speed change and terrain conditions (combined with the terrain slope in the environmental characteristic parameters). By calculating parameters in multiple dimensions such as speed, steering, and power, it arranges the various action instructions in chronological order to generate a final action sequence that includes content such as "first decelerate to X kilometers per hour for Y seconds, then turn to Z degrees, and at the same time adjust the power output to A watts."

[0063] The execution timestamp corresponding to each action in the final action sequence is extracted with millisecond accuracy. The specific time point when each action plan starts to be executed is recorded. Then, the urgency level of each action is calculated by comparing the execution timestamp of the action with the current system time to obtain the time difference. The smaller the time difference, the faster the action needs to be executed and the higher the urgency level. At the same time, combined with the importance of the action, such as actions involving obstacle avoidance and emergency parking, their urgency level is one level higher than that of ordinary steering and acceleration actions at the same time difference. For example, if the difference between the execution timestamp of an action and the current system time is 1 second and it is an obstacle avoidance action, the urgency level is set to the highest level; if the time difference of another action is 3 seconds and it is an ordinary acceleration action, the urgency level is set to medium.

[0064] Next, the difference between the action timestamps is calculated, that is, the difference between the execution timestamps of two adjacent actions, to obtain the time interval data. When the encoding algorithm based on timestamp difference is used for compression processing, the complete timestamp of the first action is retained first. For subsequent actions, the complete timestamp is no longer stored, but the difference between it and the previous action timestamp is stored. At the same time, the difference data is classified according to the urgency level. The timestamp difference corresponding to the action with a high urgency level is recorded with a higher precision to ensure the accuracy of the time information; for actions with a low urgency level, the precision of the difference record can be appropriately reduced to reduce the amount of data. In addition, the repeated action instructions in the control instruction data set are identified, such as multiple consecutive identical steering angle adjustment instructions. Only the first instruction and the number of repetitions are recorded to further compress the data volume.

[0065] The compressed control command data packet is encapsulated according to the format requirements of the wireless communication protocol, and the data packet header information is added, including the sender address, receiver address, data packet sequence number, data length, etc. After encapsulation, the data packet is sent out through a wireless communication module (such as Wi-Fi, Bluetooth, 5G, etc.). During the transmission process, the transport layer monitors the transmission status of the data packet in real time. After sending each data packet, the sender starts a timer and waits for the receiver's confirmation reply signal. If the confirmation signal is received from the receiver within a preset time threshold, it means that the data packet has been successfully received, the timer is reset, and the next data packet is sent. If the confirmation signal is not received, it is determined that data conflict or loss may have occurred. At this time, the sender first checks the occupancy of the current wireless channel. If the channel is idle, the data packet is immediately resent. If the channel is busy, it waits for a random time interval and tries to send again. During the retransmission process, the number of retransmissions is counted. When the number of retransmissions reaches a preset upper limit, the data packet is marked as a transmission anomaly and the transmission status is fed back to the system. At the same time, subsequent data packets are sent to ensure that normal instructions are transmitted first, ultimately ensuring that the control instructions are reliably fed back to the execution end of the mobile intelligent device.

[0066] By accurately extracting environmental characteristic parameters and performing multi-dimensional matching with a preset rule base, a state transition strategy adapted to the current scenario can be quickly and accurately determined, avoiding decision-making biases caused by environmental misjudgment. This rule-based matching method reduces decision-making time in complex scenarios and improves the pertinence and reliability of the state transition strategy. The current device state parameters are combined with the state transition strategy to calculate and generate an action sequence, ensuring that the action sequence not only meets the requirements of the environmental scenario but also fully considers the actual situation of the device itself. Analysis of the action urgency level and encoding compression based on timestamp differential processing can ensure the efficient transmission of emergency action instructions. By compressing the control instruction data set, the data transmission volume is reduced, the wireless communication bandwidth utilization rate is reduced, and the transmission time is shortened. The retransmission mechanism with conflict detection solves the problems of data collision and loss during wireless transmission, improving the reliability of control instruction transmission. Real-time monitoring and timely retransmission avoid interruption or erroneous execution of the device execution end due to instruction loss. At the same time, the limit on the number of retransmissions and the abnormal feedback mechanism balance transmission reliability and transmission efficiency, ensuring that control instructions can still be stably and promptly delivered to the execution end in complex wireless communication environments, ensuring the normal operation of mobile intelligent devices.

[0067] In a preferred embodiment of the present invention, step 6 above, based on the generated full-process data, dynamically adjusts the fusion strategy parameters through real-time anomaly detection and device computing power adaptation, and allocates computing resources to key functional data processing links to achieve real-time fusion processing of multi-source data, which may include: Step 660: construct an anomaly detection matrix for the entire process data, and generate a fusion strategy parameter adjustment instruction when it is detected that the singular value of the matrix exceeds a preset threshold; Step 661: Based on the fusion strategy parameter adjustment instruction, the real-time computing power level of the mobile intelligent device is obtained, and according to the confidence quality coefficient ratio of each processing link, the core resources of the processor are dynamically reallocated to the feature extraction, fusion calculation and decision generation links to obtain the core resource reallocation result; Step 662 , based on the core resource reallocation result, uses the sliding window mechanism to update the sensor compensation coefficient, multi-source data weights, and state transition decision rules to achieve real-time fusion processing of multi-source data.

[0068] In an embodiment of the present invention, when constructing an anomaly detection matrix for full-process data, various types of data in the full-process of multi-source data are first collected, including raw data collected by sensors, intermediate data in the feature extraction link, result data after fusion calculation, and output data generated by decision-making, etc., and then these data are filled into the rows and columns of the matrix in sequence according to the time sequence and data type of the data generation to form a two-dimensional matrix containing the full-process data, namely the anomaly detection matrix; when detecting the singular values ​​of the matrix, the constructed anomaly detection matrix is ​​first subjected to data normalization processing to eliminate the influence of different data types and magnitudes, and then all singular values ​​are extracted from the matrix through the relevant calculation process of singular value decomposition, and each extracted singular value is compared one by one with a preset threshold. When it is found that any singular value exceeds the preset threshold, the system automatically triggers the instruction generation mechanism to generate an instruction for adjusting the fusion strategy parameters.

[0069] When the real-time computing power level of the mobile intelligent device is obtained by adjusting the instruction based on the fusion strategy parameters, the instruction triggers the built-in computing power monitoring mechanism of the device, which will count the number of instructions completed by the processor per unit time, the number of processor cores currently active, and the processor utilization rate in real time. By comprehensively analyzing these data, the current real-time computing power level of the mobile intelligent device is obtained, including the total number of available processor cores, the processing capacity of each core and other information; when allocating processor core resources according to the confidence quality coefficient ratio of each processing link, the performance data of the three processing links of feature extraction, fusion calculation and decision generation in the data processing process are first collected, including the accuracy of data processing. Rate, stability, error rate, etc., based on these data, the confidence quality coefficient of each link is calculated. The higher the confidence quality coefficient, the higher the reliability and importance of the link in data processing. Then, the proportional relationship between the confidence quality coefficients of the three links is calculated, and then combined with the total number of available processor cores in the obtained real-time computing power level, the total core resources are allocated to the three links according to the above ratio; for example, if the total number of available cores is 10, and the confidence quality coefficient ratio of the three links is 3:5:2, then 3 cores are allocated to the feature extraction link, 5 cores are allocated to the fusion calculation link, and 2 cores are allocated to the decision generation link, thereby obtaining the core resource reallocation result.

[0070] When updating the sensor compensation coefficient using the sliding window mechanism based on the core resource reallocation results, the size of the sliding window is first determined based on the resource proportion of each link after the core resource reallocation. The window corresponding to the link with a high resource proportion can be appropriately increased to more comprehensively capture data changes. The sliding window slides on the data sequence at a fixed time interval or data quantity interval. After each sliding of the window, the collected data of the sensor in the window and the actual real data (or calibrated data) are collected. By comparing the difference between the sensor collected data and the real data in the window, the average error of each sensor in the window is calculated, and the original sensor compensation coefficient is adjusted according to the size of the average error. The larger the error, the larger the adjustment range of the compensation coefficient, thereby completing the update of the sensor compensation coefficient.

[0071] When updating the weights of multi-source data, the sliding window mechanism is also relied upon. After the window slides, the contribution of data from different data sources in the window to the fusion processing is collected, including indicators such as data accuracy, completeness, and timeliness. The performance of each data source is evaluated and scored based on these indicators. The scoring results are combined with the resource proportion of each link in the core resource reallocation, and the original multi-source data weights are adjusted. For data sources that perform well in the window and have a higher proportion of resources in the corresponding processing link, their weights are appropriately increased; otherwise, their weights are reduced. When updating the state transition decision rules, the sliding window collects the state transition data and the corresponding decision results in the window, analyzes the laws of state transition in these data and the effectiveness of the decision results, and combines the changes in the processing capabilities of each link after the core resource reallocation to revise the original state transition decision rules. For example, when the resources of the fusion computing link increase and the processing capability is improved, more considerations of the fusion computing details can be added to the decision rules to make the state transition decision more accurate, thereby realizing real-time fusion processing of multi-source data.

[0072] By constructing an anomaly detection matrix for the entire data process and monitoring singular values, anomalies in the multi-source data process can be detected in a timely manner. When the singular value exceeds the preset threshold, an adjustment instruction is generated to prevent the abnormal data from interfering with the fusion process, providing a preliminary guarantee for the accuracy of multi-source data fusion and ensuring that the fusion process is carried out under the premise of a reliable data foundation. Based on the adjustment instructions and the real-time computing power level, the processor core resources are dynamically reallocated according to the confidence quality coefficient ratio, which can make more rational use of resources and allocate more resources to key links with high confidence quality coefficients, thereby improving the processing efficiency and quality of these links, avoiding resource waste, and enabling mobile intelligent devices to exert their ultimate processing performance under limited computing power conditions. The sliding window mechanism is used to update the sensor compensation coefficient, multi-source data weights, and state transition decision rules, so that the fusion strategy parameters can adapt to data changes and resource allocation in real time. The dynamic characteristics of the sliding window ensure the timeliness of parameter updates. The update of the sensor compensation coefficient can reduce the impact of sensor errors. The adjustment of the multi-source data weights can highlight the role of high-quality data. The state transition decision rules can improve decision accuracy, ultimately achieving efficient and accurate real-time fusion processing of multi-source data.

[0073] like Figure 2 As shown, an embodiment of the present invention also provides a multi-source data real-time fusion processing system for a mobile smart device, comprising: The data acquisition module is used to collect multi-dimensional raw data streams in real time through a heterogeneous sensor array integrated in a mobile intelligent device, and to align the data streams of different sensors using a spatiotemporal synchronization mechanism to generate a spatiotemporally consistent raw data set; The dynamic processing module is used to perform dynamic interpolation and compensation operations on the original data set, and build a dynamic calibration framework based on the intrinsic topological relationship of the data stream to form a dynamic perception domain. It generates an evolution sequence based on the evolution behavior of the data units at the domain boundary, generates spatial correction values ​​based on the offset characteristics of the evolution sequence and the preset benchmark, and combines real-time data correlation analysis to generate preprocessed data that integrates the spatial correction values; The feature evaluation module is used to extract multi-dimensional features from pre-processed data, evaluate the confidence quality of each data source in real time, and generate a feature set with real-time confidence evaluation; The fusion computing module is used to dynamically allocate the fusion weights of each data source through an adaptive weighting strategy based on a feature set with real-time confidence evaluation, and to perform real-time fusion calculations in combination with the layered processing mechanism of the edge computing architecture to generate low-latency fusion results; The command feedback module is used to send the low-latency fusion results to the edge node via a lightweight transmission protocol, execute the real-time decision algorithm at the node to generate control commands and feed them back to the mobile intelligent device; The resource scheduling module is used to dynamically adjust the fusion strategy parameters based on the generated full-process data through real-time anomaly detection and device computing power adaptation, and allocate computing resources to key functional data processing links to achieve efficient fusion processing of multi-source data.

[0074] It should be noted that this system is a system corresponding to the above method, and all implementation methods in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0075] An embodiment of the present invention further provides a computing device comprising: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the above-described method. All implementations in the above-described method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0076] The embodiment of the present invention further provides a computer-readable storage medium storing instructions, which, when executed on a computer, causes the computer to execute the above-described method. All implementations in the above-described method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0077] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for real-time fusion processing of multi-source data of a mobile intelligent device, characterized in that: The method comprises: Step 1: Through the heterogeneous sensor array integrated in the mobile intelligent device, multi-dimensional raw data streams are collected in real time, and the data streams of different sensors are aligned using a spatiotemporal synchronization mechanism to generate a spatiotemporally consistent raw data set; Step 2: Perform dynamic interpolation and compensation on the original dataset, and construct a dynamic calibration framework based on the intrinsic topological relationship of the data stream to form a dynamic perception domain. Generate an evolution sequence based on the evolution behavior of the data units at the domain boundary, and generate spatial correction values ​​based on the offset characteristics of the evolution sequence and the preset benchmark. Combined with real-time data correlation analysis, generate preprocessed data that integrates the spatial correction values. Step 3: Perform multi-dimensional feature extraction on the pre-processed data, evaluate the confidence quality of each data source in real time, and generate a feature set with real-time confidence evaluation; Step 4: Based on the feature set with real-time confidence evaluation, the fusion weights of each data source are dynamically assigned through an adaptive weighting strategy. In combination with the layered processing mechanism of the edge computing architecture, real-time fusion calculations are performed to generate low-latency fusion results. Step 5: Send the low-latency fusion results to the edge node via a lightweight transmission protocol, where a real-time decision algorithm is executed to generate control instructions and feed them back to the mobile smart device. Step 6: Based on the generated full-process data, through real-time anomaly detection and device computing power adaptation, dynamically adjust the fusion strategy parameters, and allocate computing resources to key functional data processing links to achieve real-time fusion processing of multi-source data.

2. The method for real-time fusion processing of multi-source data for a mobile intelligent device according to claim 1, characterized in that: Perform dynamic interpolation and compensation operations on the original data set, and build a dynamic calibration framework based on the intrinsic topological relationship of the data stream to form a dynamic perception domain, including: Manifold learning is performed on the original data set that is consistent in time and space to extract the spatial proximity characteristics and temporal continuity features of the data units and generate a sparse correlation matrix including the quantitative correlation strength; Based on the sparse correlation matrix, the gradient modulus of each row element is calculated, and the mutation position where the gradient modulus exceeds the set threshold is detected. The mutation position is used as the boundary point to form a spatial segmentation between the internal stable area and the external change area; In the external change area, according to the corresponding connection strength values ​​in the sparse correlation matrix, the data missing area is interpolated based on the intensity weighting to generate a compensated continuous data stream; Based on the spatial distribution characteristics of continuous data streams, a dynamic perception domain of the structured mapping framework is formed.

3. The method for real-time fusion processing of multi-source data for a mobile intelligent device according to claim 2, characterized in that: Generate an evolution sequence based on the evolution behavior of data units at the domain boundary. Generate a spatial correction value based on the offset characteristics of the evolution sequence and the preset benchmark. Combined with real-time data correlation analysis, generate pre-processed data that integrates the spatial correction value, including: Acquire the data unit of the core perception area and collect the state transition vector in the continuous time window in real time; Based on the state transition vector, sorted by timestamp, a multidimensional time series describing the evolution of data units is generated; Perform dynamic time warping on the multidimensional time series and the preset benchmark sequence, and calculate the deviation of the covariance characteristics of each dimension; According to the covariance characteristic deviation, the spatial correction coefficient is generated through the preset conversion rules, and the step spatial correction coefficient is injected into the continuous data stream. At the same time, the correction amplitude is dynamically adjusted based on the Pearson correlation characteristics of the multi-source sensor data to generate preprocessed data of the fused spatial correction value.

4. The method for real-time fusion processing of multi-source data for a mobile intelligent device according to claim 3, characterized in that: Perform multi-dimensional feature extraction on pre-processed data, evaluate the confidence quality of each data source in real time, and generate a feature set with real-time confidence evaluation, including: Based on the preprocessed data stream, the time domain, frequency domain and statistical domain feature dimensions are separated through parallel feature extraction channels to generate a multidimensional feature set; Based on the multidimensional feature set, the entropy change rate is calculated independently for each feature dimension, and based on the entropy stability within the sliding window, a preliminary confidence index for each feature dimension is generated. The spectral energy distribution of each feature dimension is extracted, and the spectral confidence coefficient of each feature dimension is calculated based on the main frequency band energy concentration. Based on the preliminary confidence index and spectral confidence coefficient, the weighted geometric average is used to fuse them and generate the comprehensive confidence weight of each feature dimension; Based on the comprehensive confidence weight, dynamic attenuation is applied to the comprehensive confidence weight of the feature dimension associated with the corresponding data source to generate the attenuated comprehensive confidence weight. The comprehensive confidence weight is used as metadata and annotated on the corresponding feature vector to form a feature set with weight annotation; Based on the feature set with weighted annotations, the mutual exclusivity between feature dimensions is analyzed. When conflicting feature dimensions are detected, the labeled confidence weights are dynamically rebalanced to generate a feature set with real-time confidence evaluation.

5. The method for real-time fusion processing of multi-source data for a mobile intelligent device according to claim 4, characterized in that: Based on a feature set with real-time confidence evaluation, the fusion weights of each data source are dynamically assigned through an adaptive weighting strategy. Combined with the layered processing mechanism of the edge computing architecture, real-time fusion calculations are performed to generate low-latency fusion results, including: Sort the data sources according to the comprehensive confidence weights corresponding to the feature sets and generate a data source sorting sequence; Based on the data source sorting sequence, the weight value corresponding to each data source is dynamically assigned. At the perception layer of the edge computing architecture, the weight value and feature set of each data source are received, and feature-level weighted fusion calculation is performed to generate preliminary fusion results. The preliminary fusion results are passed to the decision layer of the edge computing architecture, where result-level calculations based on context information and fusion results are performed to generate fusion results. The calculation delay of the fusion result is monitored in real time. When the calculation delay exceeds the preset delay threshold, the weight value distribution method is dynamically adjusted to obtain a fusion result that meets the preset delay threshold.

6. The method for real-time fusion processing of multi-source data for a mobile intelligent device according to claim 5, characterized in that: The low-latency fusion results are sent to edge nodes via a lightweight transmission protocol. The nodes then execute real-time decision-making algorithms to generate control instructions and feed them back to mobile smart devices, including: Based on the environmental feature parameters in the low-latency fusion results, the preset instruction generation rule library is queried to match the state transition strategy adapted to the current scenario; The state transition strategy is input into the state transition equation calculation process, and combined with the current device state parameters, the final action sequence of the mobile intelligent device is generated; The execution timestamps of each action in the final action sequence are analyzed to calculate the urgency level of each action; based on the difference between the urgency level and the action timestamp, a coding algorithm based on timestamp difference is used to compress the control instruction data set; The compressed control instruction data packet is sent via wireless communication, and a retransmission mechanism with conflict detection is adopted at the transport layer to ensure that the instruction is reliably fed back to the execution end of the mobile intelligent device.

7. The method for real-time fusion processing of multi-source data for a mobile intelligent device according to claim 6, characterized in that: Based on the generated full-process data, through real-time anomaly detection and device computing power adaptation, fusion strategy parameters are dynamically adjusted, and computing resources are allocated to key functional data processing links to achieve real-time fusion processing of multi-source data, including: Construct an anomaly detection matrix for the entire process data. When the singular value of the matrix exceeds the preset threshold, generate a fusion strategy parameter adjustment instruction. Based on the fusion strategy parameter adjustment instructions, the real-time computing power level of the mobile intelligent device is obtained, and according to the confidence quality coefficient ratio of each processing link, the core resources of the processor are dynamically reallocated to the feature extraction, fusion calculation and decision generation links to obtain the core resource reallocation result; Based on the results of core resource reallocation, the sliding window mechanism is used to update the sensor compensation coefficient, multi-source data weights and state transition decision rules to achieve real-time fusion processing of multi-source data.

8. A multi-source data real-time fusion processing system for a mobile intelligent device, the system implementing the method according to any one of claims 1 to 7, characterized in that: include: The data acquisition module is used to collect multi-dimensional raw data streams in real time through a heterogeneous sensor array integrated in a mobile intelligent device, and to align the data streams of different sensors using a spatiotemporal synchronization mechanism to generate a spatiotemporally consistent raw data set; The dynamic processing module is used to perform dynamic interpolation and compensation operations on the original data set, and build a dynamic calibration framework based on the intrinsic topological relationship of the data stream to form a dynamic perception domain. It generates an evolution sequence based on the evolution behavior of the data units at the domain boundary, generates spatial correction values ​​based on the offset characteristics of the evolution sequence and the preset benchmark, and combines real-time data correlation analysis to generate preprocessed data that integrates the spatial correction values; The feature evaluation module is used to extract multi-dimensional features from pre-processed data, evaluate the confidence quality of each data source in real time, and generate a feature set with real-time confidence evaluation; The fusion computing module is used to dynamically allocate the fusion weights of each data source through an adaptive weighting strategy based on a feature set with real-time confidence evaluation, and to perform real-time fusion calculations in combination with the layered processing mechanism of the edge computing architecture to generate low-latency fusion results; The command feedback module is used to send the low-latency fusion results to the edge node via a lightweight transmission protocol, execute the real-time decision algorithm at the node to generate control commands and feed them back to the mobile intelligent device; The resource scheduling module is used to dynamically adjust the fusion strategy parameters based on the generated full-process data through real-time anomaly detection and device computing power adaptation, and allocate computing resources to key functional data processing links to achieve efficient fusion processing of multi-source data.

9. A computing device, characterized in that include: one or more processors; A storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images