A multi-terminal data fusion system for trackless rubber-tyred vehicles

Abstract

The invention discloses a multi-terminal data fusion system for a trackless rubber-tyred vehicle, relates to the technical field of data fusion, and is used to solve the problem of poor multi-terminal data fusion of a trackless rubber-tyred vehicle; the system comprises a path quantification module, an equipment health classification module, an alternative optimization module, and a consistency verification module; the invention fuses the multi-terminal data of the trackless rubber-tyred vehicle, dynamically constructs a sensor dependency graph and calculates dependency weights, optimizes key paths, defines health indicators according to signal-to-noise ratio, output volatility and response delay, combines sliding windows and Mahalanobis distance to detect multidimensional anomalies, generates health scores to monitor sensor states, isolates faulty sensors, generates alternative data through redundant data sources, incorporates the data into a fusion process after verification, adjusts the dependency graph and optimizes the path, performs global consistency verification analysis, triggers different strategy management modules, optimizes data fusion rules and task allocation, and improves the data fusion reliability of the trackless rubber-tyred vehicle in a complex operating environment.

Description

Technical Field

[0001] The present invention relates to the technical field of data fusion, and more specifically to a multi-terminal data fusion system for a trackless rubber-tyred vehicle. Background Art

[0002] The rubber-tyred trackless vehicle is an important transportation equipment widely used in complex environments such as mines, construction sites, and logistics transportation. Its flexible operation mode and high load capacity make it an indispensable operating tool. With the continuous advancement of industrialization and modernization, the demand for efficient transportation equipment is increasing. Although the traditional rail transportation method has certain load-bearing advantages, its flexibility and adaptability are poor, especially in scenes with complex terrain or frequent dynamic adjustment needs. The rubber-tyred trackless vehicle has become an ideal choice for many industries because it does not rely on fixed tracks and is flexible and efficient.

[0003] Deficiencies in the existing technology: In the multi-sensor system of rubber-tyred trackless vehicles, there is a strong dependency between some sensors. For example, position sensors and speed sensors need to cooperate with each other to provide accurate vehicle motion data. When one of the sensors fails or outputs abnormally, the other sensor may have calculation deviations or failures due to its dependence on its input data, resulting in a significant decrease in the accuracy of the entire data fusion result. This phenomenon is called the data dependency destruction effect between sensors, which will make the system unable to respond in time or make correct decisions in a complex environment. Especially in harsh environments such as mines and construction sites, the incidence of sensor failures is relatively high, further aggravating the impact of this problem on system performance. It is impossible to isolate or replace faulty sensors in real time, and there is a lack of methods for dynamically compensating data in the dependency chain, which seriously limits the reliability and robustness of the system. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, there is a solution as follows to solve the problem of poor multi-terminal data fusion effect of the trackless rubber-tyred vehicle in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A multi-terminal data fusion system for a trackless rubber-tyred vehicle, comprising a path quantification module, an equipment health classification module, an alternative optimization module and a consistency verification module, wherein the modules are connected via signals;

[0007] The path quantification module is used to calculate the dependency weights by combining the impact metrics of data correlation, time synchronization and environmental correlation, quantify the dependency of sensors through graph theory methods, and identify the critical paths;

[0008] The device health classification module is used to define health indicators through signal-to-noise ratio, output volatility and response delay, use sliding windows to capture short-term changes, detect multi-dimensional anomalies based on Mahalanobis distance, generate health scores for sensors and classify them;

[0009] The alternative optimization module is used to isolate faulty sensors with substandard health scores, fuse erroneous data with redundant data sources to generate alternative data, incorporate it into the data fusion process after verifying the deviation, reconstruct the dependency graph and optimize the path, and perform global consistency verification analysis on the fusion results;

[0010] The consistency verification module is used to obtain the verification analysis information generated during the global consistency verification of the fusion results, analyze the verification analysis information and generate different signals, and perform policy management based on the generated different signals.

[0011] In a preferred embodiment, the dependency weights are calculated by combining the impact metrics of data correlation, time synchronization and environmental correlation, quantifying the dependency of sensors through graph theory methods, and identifying the critical path. The specific steps are as follows:

[0012] The sensors of rubber-tyred trackless vehicles are modeled by graph theory, each sensor is regarded as a node, and the data dependency is regarded as an edge. The weight of the dependency is determined by the data relevance, time synchronization and environmental impact measurement.

[0013] Use mutual information to measure the correlation of data, use dynamic time warping to calculate the synchronization measurement of time series data, and use distance weighting method to measure environmental correlation;

[0014] Aggregate the correlation measures, synchronization measures, and environmental correlations to form a dependency graph;

[0015] The path with the largest sum of weights in the dependency chain in the dependency graph is taken as the critical path.

[0016] In a preferred embodiment, the health index is defined by signal-to-noise ratio, output volatility and response delay, a sliding window is used to capture short-term changes, and multi-dimensional anomalies are detected according to Mahalanobis distance, and a health score is generated and classified for the sensor. The specific steps include:

[0017] Extract health indicators, including signal-to-noise ratio, data output volatility, and response delay;

[0018] The sliding window method is used to observe the short-term changes of each health indicator to determine whether there are any abnormalities;

[0019] If an abnormality occurs, a multi-dimensional anomaly detection method is used for identification. The multi-dimensional data of the sensor health status is used to form a feature vector. The detection value is determined by the Mahalanobis distance detection anomaly. When the detection value is greater than the set abnormal threshold, the sensor status is determined to be abnormal.

[0020] Based on the anomaly detection results of monitoring, a health score is generated for each sensor, and the status is classified according to the health score;

[0021] The status classification is based on the health score setting rules. A health score greater than or equal to 0.8 indicates a good health status, a health score greater than or equal to 0.5 and less than 0.8 indicates a status degradation warning, and a health score less than 0.5 indicates a serious fault warning.

[0022] In a preferred embodiment, the faulty sensors with substandard health scores are isolated, and the erroneous data are fused with redundant data sources to generate replacement data, which are then incorporated into the data fusion process after the deviation is verified. The specific steps are as follows:

[0023] Sensors whose health scores are less than the fault identification threshold are regarded as faulty sensors, and the output data of the faulty sensors are removed from the data stream and marked as isolated;

[0024] After the faulty sensor is isolated, the redundant data sources available in the system are enabled to replace the output of the faulty sensor. The set of redundant data source nodes associated with the faulty sensor is determined through the dependency graph, and the outputs of the redundant data sources are weightedly fused to generate replacement data.

[0025] When no redundant data source is available or the redundant data is incomplete, use historical data and time series prediction algorithms to generate replacement data for the faulty sensor;

[0026] For sensor data with obvious periodicity or trend, the ARIMA model or variational autoencoder is used to determine the alternative data. For dynamically changing sensor data, the LSTM model or Transformer model is used to determine the alternative data.

[0027] Fuse the substitute data with the current operating data, and ensure the reliability of the substitute data through a verification mechanism, and evaluate the deviation of the substitute data based on the current sensor data fusion results;

[0028] When the deviation of the alternative data is less than the tolerance threshold, the alternative data is confirmed to be valid. After verification, the alternative data is included in the fusion process, and the replacement strategy is adjusted in real time according to the fusion results.

[0029] In a preferred embodiment, the dependency graph is reconstructed and the path is optimized, and a global consistency check analysis is performed on the fusion result, including the following steps:

[0030] Re-evaluate and update the weights in the dependency graph using the latest sensor health status and alternative data results;

[0031] Based on the updated dependency graph, the critical path is dynamically readjusted, and the sensor dependency chain is adjusted using the shortest path or minimum spanning tree;

[0032] After completing the dynamic adjustment of the dependency chain, the multi-terminal data fusion process is globally optimized and the fusion results are globally checked for consistency.

[0033] In a preferred embodiment, the method for obtaining the verification analysis information generated during the global consistency verification of the fusion result includes the following steps:

[0034] Obtain verification and analysis information during the scheduling optimization process, including collaborative impact information and fusion matching information;

[0035] The collaborative impact information includes the sensor collaborative index, and the fusion matching information includes the fusion deviation consistency index;

[0036] The obtained sensor synergy index and fusion deviation consistency index are combined to generate a fusion state coefficient;

[0037] The sensor synergy index and fusion bias consistency index are positively correlated with the fusion state coefficient.

[0038] In a preferred embodiment, analyzing and verifying the analysis information and generating different signals, and performing policy management according to the generated different signals, includes the following steps:

[0039] Compare the generated fusion state coefficient with the set fusion evaluation threshold;

[0040] If the fusion state coefficient is greater than or equal to the fusion evaluation threshold, a multi-terminal fusion stability signal is generated, indicating that the output data fluctuation is stable and no additional adjustment is required;

[0041] If the fusion state coefficient is less than the fusion evaluation threshold, a multi-terminal fusion fluctuation signal is generated, indicating that a deviation has occurred in the data fusion and timely optimization is required.

[0042] The technical effects and advantages of the multi-terminal data fusion system for trackless rubber-tyred vehicles of the present invention are as follows:

[0043] The present invention dynamically constructs a sensor dependency graph, integrates the metrics of data correlation, time synchronization and environmental correlation, quantifies the dependencies between sensors, and dynamically identifies key paths, thereby achieving precise management and optimization of the sensor network structure. The health index is defined by signal-to-noise ratio, output volatility and response delay, a sliding window is used to capture short-term change trends, and multi-dimensional anomalies are detected in combination with Mahalanobis distance. Health scores are generated and classified, effectively realizing real-time monitoring of the health status of sensors. For faulty sensors, anomalies are identified and quickly isolated through health scores, and substitute data is generated in combination with weighted fusion of redundant data sources or prediction algorithms to ensure data integrity and continuity. The substitute data is incorporated into the data fusion process after deviation verification, thereby reducing the impact of sensor failures on the operation of rubber-tyred trackless vehicles.

[0044] After fault isolation, the dependency graph is dynamically reconstructed, the critical path is optimized, and the fusion results are analyzed using global consistency verification to ensure the stability and accuracy of multi-terminal data fusion. At the same time, different signals are generated based on the verification and analysis results, and the signal triggers the strategy management module to achieve intelligent control of data fusion and task allocation, thereby improving the real-time adaptability and robustness of the multi-terminal data fusion process of trackless rubber-tyred vehicles, and providing support for high-demand scenarios such as mining transportation. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The present invention is a structural schematic diagram of a multi-terminal data fusion system for a trackless rubber-tyred vehicle. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] In order to achieve the above objectives, Figure 1 A structural schematic diagram of a multi-terminal data fusion system for a trackless rubber-tyred vehicle of the present invention is given, which specifically includes a path quantification module, an equipment health classification module, an alternative optimization module and a consistency verification module, and each module is connected by a signal;

[0048] The path quantification module is used to calculate the dependency weights by combining the impact metrics of data correlation, time synchronization and environmental correlation, quantify the dependency of sensors through graph theory methods, and identify the critical paths;

[0049] The device health classification module is used to define health indicators through signal-to-noise ratio, output volatility and response delay, use sliding windows to capture short-term changes, detect multi-dimensional anomalies based on Mahalanobis distance, generate health scores for sensors and classify them;

[0050] The alternative optimization module is used to isolate faulty sensors with substandard health scores, fuse erroneous data with redundant data sources to generate alternative data, incorporate it into the data fusion process after verifying the deviation, reconstruct the dependency graph and optimize the path, and perform global consistency verification analysis on the fusion results;

[0051] The consistency verification module is used to obtain the verification analysis information generated during the global consistency verification of the fusion results, analyze the verification analysis information and generate different signals, and perform policy management based on the generated different signals.

[0052] Step 1: Model the sensor dependency relationship, build a dynamic dependency diagram of sensors in the multi-terminal data fusion system of rubber-tyred trackless vehicles, clarify the dependency path and strength between sensors, and provide complete decision support data for subsequent health monitoring, fault isolation and adjustment. The specific steps are as follows:

[0053] The sensors of the trackless rubber-tyred vehicle are modeled by graph theory. Each sensor is regarded as a node V, and its data dependency is regarded as an edge E. For any two sensors i and j, the weight of the dependency is The weight is determined by the impact measures of data relevance, time synchronization, and environmental relevance. The weight calculation formula is: , where the weight coefficient , , Dynamically adjust through reinforcement learning algorithms to adapt to changes in different operating scenarios;

[0054] The correlation calculation is performed, and the data correlation between sensors i and j is calculated by the mutual information Metrics reflect the correlation between data, that is, the degree of nonlinear correlation between sensor outputs: , where p(x, y) represents the joint distribution of sensors i and j, p(x) and p(y) are the marginal distributions of sensors i and j, X is the data set or distribution of sensor i, and Y is the data set or distribution of sensor j. If the output of sensor i is a certain sequence of values, then X is the set of these values, that is, it represents the value range of the data, not a specific set of values. Similarly, Y is the data set of sensor j;

[0055] To perform synchronization calculation, dynamic time warping (DTW) is used to measure the synchronization of time series data, that is, the synchronization measurement based on dynamic time warping reflects the degree of time series matching. The calculation formula is: , where P is the alignment path of the time series, is the difference between the points of sensors i and j at path k;

[0056] Perform environmental correlation calculations. Based on environmental sensor data (such as slope, temperature and humidity), measure the dependency strength of sensors i and j in the operating scenario, that is, measure the impact of the external environment on the sensor dependency relationship. The distance-weighted method is used for calculation. The formula is as follows: ,in, and are the characteristic vectors of the environment where sensors i and j are located (such as temperature, humidity, slope, etc.), To influence the range parameter, we control the sensitivity of the differences between environmental feature vectors, which is used to determine the degree to which environmental differences will affect the correlation between sensors;

[0057] There are multi-dimensional dependencies between the sensors of the rubber-tyred trackless vehicle (such as speed, position, load, and power state). For example, the accuracy of the speed sensor may depend on the data of the position sensor, and the power state sensor may be indirectly affected by the load sensor. The dependency weights between the sensors are quantified by the above method to form a dynamic dependency graph G(V, E, W), thereby providing an accurate dependency chain description for the system operation status.

[0058] In the actual operation of rubber-tyred trackless vehicles, the critical path may include the dependency chain of position sensors, speed sensors and load sensors. After marking the critical paths, the system can prioritize monitoring and protecting these paths to ensure the stability of key data flows, thereby reducing misjudgments and failures during operation.

[0059] Identify the critical path from the dependency graph, that is, the path with the largest sum of weights in the dependency chain, which is used to indicate the sensor path that has the most important impact on system operation. The calculation formula for the critical path is: , where P represents the set of paths from the starting point to the end point, that is, the path from the starting point to the end point, each edge in the path consists of a pair of sensors, G is the dependency graph, is the weight of the dependency relationship of edge (i, j) in the path, indicating the degree of dependence or influence of sensor i on sensor j;

[0060] Dynamic weight update: The weight of the dependency relationship will fluctuate during operation due to environmental changes or sensor state changes, so dynamic update is performed. The weight update adopts the sliding window weighted average method. For example, the weight of the dependency relationship at time t+1 is determined by the following calculation formula: ,in, To update the coefficient, control the balance between historical weight and current weight; is the weight change;

[0061] The dynamically updated dependency graph and its critical path are stored in the graph database, and the critical path data is preferentially transmitted to the monitoring-related modules to ensure that system resources are concentrated on the most important dependency chain;

[0062] In summary, this step 1 is to ensure the system’s priority attention to important dependency chains by building a dependency graph between sensors, integrating data correlation, time synchronization, and environmental correlation, and marking critical paths, and dynamically update weights to adapt to real-time environmental changes and maintain the accuracy of the dependency graph. Finally, the dynamic dependency graph and critical path data are transmitted and monitored for subsequent fault isolation and system adjustment.

[0063] Step 2: Monitor the health status of the sensor. After completing multi-layer data collection and cleaning, monitor the health status of the real-time output data of the sensor. Evaluate the operating status of the sensor in multiple dimensions to detect possible anomalies, failures, or performance degradation. The specific steps are as follows:

[0064] Define and extract health indicators. In the rubber-tyred trackless vehicle scenario, the calculation of sensor health indicators needs to be combined with the dynamic requirements of the vehicle's actual operation. For example, for speed sensors, low signal-to-noise ratio may lead to incorrect recognition of driving status, which in turn affects path planning. For load sensors, noise interference may cause power distribution errors, resulting in overload or increased energy consumption.

[0065] Data output fluctuation: During vehicle driving, stable sensor output is particularly critical. If the position sensor fluctuation is abnormal, it may cause vehicle positioning deviation or path planning error.

[0066] Response delay: If the response delay of the power state sensor is too long, it may cause the vehicle to lag in emergency obstacle avoidance, increasing the risk of accidents.

[0067] The core of health status monitoring is to define and extract the health index (HI) of the sensor. The health index comprehensively considers factors such as the stability of sensor data, signal-to-noise ratio, and output characteristic changes. The health index of sensor i is defined as follows: ,in, The signal-to-noise ratio (SNR) measures the noise interference of the output signal of sensor i. is the data output fluctuation of sensor i, measured by the output change rate at consecutive moments, is the response delay, which is used to reflect the response speed of sensor i to external changes. , , For each weight coefficient, dynamically adjust it through historical data or reinforcement learning methods;

[0068] Based on the definition of health indicators, the health status of the sensor is monitored in real time and trend analyzed. The sliding window method is used to observe the short-term changes of each health indicator. The health indicator change rate formula of sensor i at time t is: , where T is the time range of the sliding window, which is used to capture short-term changes in health indicators. is the health indicator of sensor i at time t;

[0069] By monitoring the short-term rate of change If it continues to decline, it can provide an early warning of possible performance degradation or failure trends;

[0070] Through the sliding window method, the changing trend of health indicators is captured in real time to determine whether the sensor is in a stable state. For the trackless rubber-tyred vehicle scenario, the selection of the sliding window length T needs to be combined with the vehicle's running speed and data update frequency. For example, in the mining transportation scenario, the data update frequency of the speed sensor may be higher, and the window length should be appropriately shortened. The short-term change rate A continuous decrease in may mean that the sensor is disturbed by the environment, such as the vibration of a slippery road surface affecting the stability of the speed sensor;

[0071] For the anomalies that may occur in health status monitoring, a multidimensional anomaly detection method is used to identify them. The multidimensional data of the health status of sensor i constitutes the health status feature vector , the detection value is determined by detecting anomalies through Mahalanobis distance, that is, is the health status feature vector of sensor i Mahalanobis distance from historical normal data: ,in, is the mean vector of health status characteristics (calculated by historical normal data), when the detection value When it is greater than the set abnormal threshold, the sensor status is judged to be abnormal. is the inverse matrix of the covariance matrix of the sensor health status eigenvector;

[0072] It should be noted that for speed and position sensors, the abnormality detection of the two needs to consider the synchronization of their outputs. The Mahalanobis distance can quantify the overall abnormality of the two. If the response delay of the power state sensor and the fluctuation of the speed sensor are abnormal at the same time, it may mean that the power system hardware is damaged and further diagnosis is required.

[0073] Generate a health score for each sensor based on real-time monitoring and anomaly detection results , and classify the status, the health score calculation formula is: , the score range is [0, 1], where Indicates good health. Indicates approaching failure;

[0074] The status classification is based on the health score and the following rules are set: Indicates good health. Indicates a warning of status degradation. Indicates a serious fault warning;

[0075] During the transportation process of rubber-tyred trackless vehicles, the stored health status data can also be used to analyze the long-term impact of different environmental conditions on sensor performance. For example, through data backtracking, the accelerated aging effect of humid environments on load sensors can be found, which can provide support for subsequent improvements in sensor design.

[0076] Step 3: After completing the sensor health status monitoring, quickly isolate the detected faulty sensor to prevent the erroneous data from affecting the overall operation of the system, and replace the output of the faulty sensor with redundant data or prediction algorithms. The specific steps are as follows:

[0077] Based on the health score and anomaly detection results generated in step 2, identify faulty sensors whose health status scores are below the threshold and isolate their outputs in real time to prevent erroneous data from entering the fusion system. Set the fault identification threshold for each sensor's health score. ;

[0078] when , the sensor is determined to be a faulty sensor, the output of the faulty sensor is removed from the data stream, and marked as isolated, and the isolation time is recorded: ,in, is the isolation status of sensor i, 1 is normal and 0 is isolated;

[0079] After the faulty sensor is isolated, the redundant data source available in the system is immediately enabled to replace the output of the faulty sensor to ensure the continuity of data fusion. The set of redundant data source nodes associated with the faulty sensor is determined through the dependency graph G(V, E, W). , weighted fusion of the outputs of redundant data sources is performed to generate alternative data: ,in, is the alternative output of the faulty sensor i, is the output of sensor j, is a redundant sensor set, is the weight of redundant sensor j to faulty sensor i in the dependency graph;

[0080] Through weighted fusion of redundant data sources, accurate replacement of faulty sensor outputs can be achieved, reducing the impact of data loss on system operation.

[0081] Use prediction algorithms to generate replacement data. When there is no redundant data source available in the system or the redundant data is incomplete, use historical data and time series prediction algorithms to generate replacement data for the faulty sensor.

[0082] For sensor data with obvious periodicity or trend (such as load sensors), ARIMA models or variational autoencoders (VAE) are preferred to capture long-term trends and determine alternative data. For sensor data with large dynamic changes (such as speed sensors), LSTM models or Transformer models are used to deal with nonlinear fluctuations and determine alternative data.

[0083] Build a historical data model of faulty sensors, select a suitable prediction algorithm (such as LSTM, variational autoencoder VAE), obtain the sensor i at the past N time points (i.e., from arrive ) historical data, which will be used as the input features of the model and use the data of the most recent N time points , predict the output at the current moment , can be based on the LSTM model: ,in, is the predicted surrogate output of sensor i at time t, is the trained LSTM model;

[0084] For example, in mining transportation, when a load sensor fails temporarily due to vibration or overload, the LSTM-based prediction algorithm can generate current alternative load data based on historical load change patterns to ensure the continuity of power system distribution;

[0085] After the alternative data is generated, its reliability needs to be evaluated through a verification mechanism to avoid the system decision being affected by excessive errors in the alternative data. The verification of the operation of rubber-tyred trackless vehicles needs to consider the overall deviation after multi-terminal data fusion, rather than the accuracy of a single data source;

[0086] The alternative data is fused with the current operating data of the system, and the reliability of the alternative data is ensured through a verification mechanism. Assessing bias in surrogate data : ;

[0087] When the bias of the alternative data Less than tolerance threshold When the replacement data is verified, it is confirmed to be valid. After the replacement data is verified, it is gradually incorporated into the fusion process, and the replacement strategy is adjusted in real time according to the fusion results: ;

[0088] Detailed records of fault isolation and replacement processes are stored to provide support for subsequent analysis and system optimization. Records include fault time, replacement strategy, verification results, etc. For long-term data storage, the isolation frequency and time of sensors are recorded, and the life decay law is analyzed in combination with the health score trend to provide data support for equipment maintenance, annotate abnormal data, and annotate sensor data during isolation for subsequent analysis of the cause of the fault or improvement of hardware design.

[0089] Step 4: Dynamically adjust the dependency chain. After the faulty sensor is isolated and replaced, the weights and dependencies between sensors are redistributed through dynamic adjustment of the dependency chain to ensure that the data fusion system can adapt to changes in the current sensor status and operating environment. The dynamic adjustment of the dependency chain aims to optimize the overall reliability and robustness of data fusion and avoid the spread of fault impacts on the dependency chain. The specific steps are as follows:

[0090] Re-evaluate the weights in the dependency graph G(V, E, W) using the latest sensor health status and alternative data results The adjustment of weights needs to comprehensively consider the sensor's health score, data accuracy, and external environmental impact. The weight update formula is: ,in, is the health score of sensor i, which is used to measure its current operating status. is the sensor data accuracy, determined by the surrogate data bias and validation results, is the environmental impact factor of the sensor pair, reflecting the dynamic effect of environmental changes on the dependency relationship. , , is the weight coefficient, which is also adjusted through historical data and reinforcement learning, and uses external sensor data (such as temperature and humidity) to To update: , and are the feature vectors of the environments where sensors i and j are located, is the impact range parameter;

[0091] Based on the updated dependency graph, the critical path is dynamically adjusted again. The critical path is the set of dependency chains that have the greatest impact on the data fusion results during the current system operation. The critical path identification formula is: , where P represents the set of paths from the starting point to the end point, that is, the path from the starting point to the end point, each edge in the path consists of a pair of sensors, G is the dependency graph, is the weight of the dependency relationship of edge (i, j) in the path, indicating the degree of dependence or influence of sensor i on sensor j;

[0092] After updating the critical path, more resources (such as higher data transmission frequency and fault tolerance) are allocated to high-priority sensors on the path;

[0093] Dynamic dependency chain reconstruction ensures that after fault isolation and replacement are completed, the system can re-establish efficient dependency relationships, and then use graph optimization algorithms (such as the shortest path or minimum spanning tree) to reorganize the sensor dependency chain to avoid redundant and inefficient paths: ,For sensors with low health scores and their connected edges, their weights are reduced to avoid the system from being overly dependent on potential faulty nodes;

[0094] Dependency chain reconstruction requires comprehensive optimization of the connection paths between sensors to avoid degradation of data fusion performance due to inefficient or faulty chains. Dependency chain reconstruction also needs to support optimization of specific scenarios. For example, when running on a steep slope in a mine, the path of the power state and load sensor should be closer.

[0095] For example, when a speed sensor fails, dependency chain reconstruction needs to generate alternative speed data through the GPS position sensor combined with the time difference method, and at the same time reduce the association weight of the speed sensor. In addition, for slippery road environments, the power state sensor needs to temporarily improve the association with the environmental sensor and adjust the power output strategy in real time;

[0096] It should be noted that for sensors with low health scores and data accuracy, their associated paths should be weakened first, and in multi-vehicle collaboration scenarios, redundant sensor paths across vehicles (such as shared GPS data) should be enabled;

[0097] The process and results of dependency chain adjustment are recorded for subsequent fault analysis, system optimization, and task allocation strategy improvement. The feedback mechanism transmits the adjustment results to the modules used for data fusion and upper-level scheduling to improve global optimization capabilities.

[0098] After completing the dynamic adjustment of the dependency chain, the multi-terminal data fusion process is globally optimized to improve the accuracy, stability and overall system operation efficiency of the fusion results. By optimizing the data source, fusion rules and system structure, the system is made more robust and real-time in a complex operating environment.

[0099] Perform global consistency check on the fusion results to ensure the logical consistency between different sensor data and identify potential contradictions or errors in a timely manner. The global consistency check directly measures the logical consistency and synergy between multiple sensor data in the system, which is crucial to the reliability and accuracy of the fusion results. Obtain the verification analysis information generated during the global consistency check of the fusion results, which includes the synergy impact information and fusion matching information;

[0100] The collaborative impact information includes the sensor collaborative index and is calibrated as SCI, and the fusion matching information includes the fusion bias consistency index and is calibrated as FBCI;

[0101] The sensor interoperability index is used to measure the interoperability of data between different sensors in a multi-sensor system. Specifically, it reflects the consistency and correlation of the output data of multiple sensors at the logical and statistical levels, and aims to evaluate whether the sensors can support, supplement or verify each other in the data fusion process.

[0102] The larger the sensor synergy index is, the more highly the data between sensors are coordinated, and the outputs between sensors have high consistency and correlation, indicating that the data fusion result is reliable, the correlation between data is strong, and a stable dependency chain can be formed; the smaller the sensor synergy index is, the poorer the data synergy between sensors is, and there may be inconsistencies, conflicts or failures, indicating that the data fusion quality may be impaired and the dependency relationship between sensors is destroyed (such as the sudden change of position sensor data is inconsistent with the logic of speed sensor);

[0103] In critical tasks (such as path planning, obstacle avoidance, or power distribution), a system with a high sensor cooperation index can quickly form high-quality fusion results, thereby improving the efficiency of task execution; a low sensor cooperation index may lead to system decision delays or misjudgments;

[0104] The logic for obtaining the sensor synergy index is as follows:

[0105] Get the fusion output data of sensor i , get the actual observation value of sensor i , calculate the fusion deviation: , get the maximum allowable value of the deviation , calculate the data fusion accuracy of sensor i: Similarly, the data fusion accuracy of sensor j is calculated ;

[0106] Get the data of sensor i and sensor j , , calculate the variance of sensor i: , where is the output value of sensor i at time point k, is the mean of the sensor i data, N is the total number of data points, and the variance of sensor j is obtained in the same way , calculate the covariance, the calculation expression is: , where is the output value of sensor j at time point k, is the mean value of sensor j data;

[0107] The correlation value is obtained by calculating the Pearson correlation coefficient, and the calculation expression is: , calculate the sensor synergy index: .

[0108] It should be noted that the fused output data is generated by combining multi-sensor outputs and weights. For example, the fused output of the speed sensor may be generated by combining GPS data and wheel speed sensor data. The actual observation data is provided by a known reference system (such as high-precision GPS or external calibration equipment). The maximum allowable value of the deviation is a threshold set by the system based on mission requirements and operating environment, which is used to define the acceptable range of fusion deviation.

[0109] The fusion bias consistency index (FBCI) is used to measure whether the deviation distribution of different sensors in the multi-sensor fusion data of the rubber-tyred trackless vehicle is uniform and consistent, reflecting the balance of the overall data fusion of the system. It evaluates the relative consistency of the deviation of the fusion output of each sensor (the degree of deviation between the actual observation value and the fusion result);

[0110] A high fusion bias consistency index reflects that the fusion bias is evenly distributed and the system runs more stably. A low fusion bias consistency index indicates that there is a risk that the system has bias concentrated in individual sensors, which may cause the system to be overly dependent on a certain type of sensor data. The higher the fusion bias consistency index, the more balanced the fusion result, indicating that the system is more comprehensive in data utilization. If the fusion bias consistency index is low, the fusion process may rely too much on a single data source.

[0111] Multi-sensor fusion in mine transportation route planning for rubber-tyred trackless vehicles. In the scenario of high fusion deviation consistency index, the deviations of speed sensor, GPS position sensor and load sensor are all within the allowable range, the deviation distribution is uniform, and the route planning is based on the comprehensive fusion results of multi-sensor data, with high decision reliability.

[0112] In the low fusion deviation consistency index scenario, the speed sensor is subject to greater interference on a slippery slope, and the deviation is significantly higher than that of the GPS and load sensor. The weight of the speed sensor may be too high, causing the path planning result to deviate from the actual demand;

[0113] In tasks such as path planning and power allocation, a system with a high fusion deviation consistency index can allocate resources and execute decisions more reliably, while a low fusion deviation consistency index may require adjusting task priorities or optimizing fusion rules.

[0114] The logic for obtaining the fusion deviation consistency index is as follows:

[0115] Get the fusion output data of sensor k , the actual observation data of sensor k , obtain the absolute difference between the fusion output data of sensor k and the actual observation data, and the calculation expression is: , the mean fusion deviation of all sensors is calculated as: , where M is the total number of sensors in the system, The fusion bias of each sensor is used to calculate the discrete value of the fusion bias of all sensors: , the fusion deviation consistency index is calculated based on the discrete degree value, fusion deviation mean, and absolute difference. The calculation expression is: .

[0116] The fusion state coefficient is generated by combining the collaborative influence information and the fusion matching information. The fusion state coefficient is generated by combining the obtained sensor synergy index and fusion deviation consistency index. The expression is: , where , is the preset proportional coefficient of the sensor synergy index and the fusion bias consistency index, and , Both are greater than 0.

[0117] The specific method of jointly generating the fusion state coefficient may involve a variety of algorithms and models, which depends on the actual situation and application requirements. In this embodiment, the sensor synergy index and the fusion deviation consistency index can be combined by weighted summation to generate a comprehensive fusion state coefficient. This fusion state coefficient can be used as an input for evaluating the multi-terminal data fusion effect of the rubber-tyred trackless vehicle, and is used to determine the multi-terminal data fusion optimization situation.

[0118] It should be noted that the size of the preset proportional coefficient is to quantify each parameter to obtain a specific value. In order to facilitate subsequent comparison, the size of the coefficient depends on the amount of sample data and the preliminary setting of the corresponding preset proportional coefficient for each group of sample data by technical personnel in this field. It is not unique, as long as it does not affect the proportional relationship between the parameter and the quantized value. For example, the sensor synergy index and the fusion bias consistency index are positively correlated with the fusion state coefficient. The sensor synergy index and the fusion bias consistency index are normalized to have the same dimension and range. This can be achieved by subtracting the mean from the original data and dividing it by the standard deviation, or mapping the data to the range of [0, 1].

[0119] The larger the sensor synergy index and the fusion deviation consistency index, the larger the fusion state coefficient generated by the joint system, indicating that the multi-sensor fusion system is more stable in the current operating environment, the utilization efficiency of sensor data is higher, and high-quality fusion results can be quickly formed. The system can more efficiently complete core tasks such as path planning, obstacle avoidance, and power distribution;

[0120] The smaller the sensor synergy index and the fusion deviation consistency index are, the smaller the fusion state coefficient generated by the joint system is, indicating that the overall efficiency and quality of the system data fusion are insufficient, which directly affects the efficiency of task execution. Some sensors may produce extreme values, affecting the stability of the overall data fusion, which is likely to increase the task execution time and may miss the best decision-making opportunity, resulting in a significant decrease in system performance and an inability to complete high-precision tasks.

[0121] Compare the generated fusion state coefficient with the preset fusion evaluation threshold to generate a multi-terminal fusion stable signal and a multi-terminal fusion fluctuation signal;

[0122] After obtaining the fusion state coefficient, the fusion state coefficient is compared with the fusion evaluation threshold;

[0123] If the fusion state coefficient is greater than or equal to the fusion evaluation threshold, a multi-terminal fusion stable signal is generated, indicating that the coordination degree of sensor data and the balance of fusion deviation distribution meet the system requirements, the internal sensor dependency chain structure is stable, there is no performance fluctuation due to sensor failure or data anomaly, the output data of each sensor fluctuates little and meets expectations, and changes in the system operating environment have little impact on the fusion result;

[0124] If the fusion state coefficient is less than the fusion evaluation threshold, a multi-terminal fusion fluctuation signal is generated, indicating that the collaborative working ability between the sensors is poor, or the fusion deviation is unevenly distributed, resulting in the data fusion quality failing to meet the task requirements. The system may generate erroneous or unstable fusion results, leading to deviations in subsequent decisions, making the fusion data of the speed and position sensors of the rubber-tyred trackless vehicle inconsistent, resulting in a decrease in the credibility of the path planning. The rubber-tyred trackless vehicle may deviate from the preset path, increasing the risk of collision or getting stuck, and the power output is unstable, which may lead to increased fuel consumption or reduced vehicle driving efficiency.

[0125] When a multi-terminal fusion fluctuation signal is generated, it indicates that the multi-terminal data fusion system of the rubber-wheeled vehicle needs to be optimized in time to ensure efficient operation in harsh environments. When a multi-terminal fusion fluctuation signal is generated, the following optimization measures need to be taken:

[0126] Reduce the weight of high-deviation sensors (such as load or speed sensors), increase the data share of reliable sensors, identify and isolate abnormal sensors, reduce their negative impact on data fusion results, re-establish broken sensor dependency chains, and improve coordination;

[0127] For example, a rubber-tyred vehicle runs on a steep slope in a mine, where the path is complex and the environment is slippery. The vehicle relies on the fusion data of speed, position and load sensors to complete path planning and power distribution. The slippery road surface causes the speed sensor output data to fluctuate greatly, which is inconsistent with the logic of the position sensor. The load sensor has a large deviation due to transportation bumps, and the fusion deviation is unevenly distributed.

[0128] Specific adjustment measures can be achieved by isolating faults, reducing the weight of speed sensors, using redundant path support from GPS data, repairing dependency chains, and adjusting the data fusion weight of load sensors, thereby reducing the impact of their deviations on the performance of rubber-tyred trackless vehicles.

[0129] It should be noted that the threshold information in this embodiment is pre-set by professionals and will not be explained in detail here. Some parameter English letters in the embodiments have the same situation, but different meanings are explained when used, which will not be explained one by one here.

[0130] The present invention dynamically constructs a sensor dependency graph, integrates the metrics of data correlation, time synchronization and environmental correlation, quantifies the dependencies between sensors, and dynamically identifies key paths, thereby achieving precise management and optimization of the sensor network structure. The health index is defined by signal-to-noise ratio, output volatility and response delay, a sliding window is used to capture short-term change trends, and multi-dimensional anomalies are detected in combination with Mahalanobis distance. Health scores are generated and classified, effectively realizing real-time monitoring of the health status of sensors. For faulty sensors, anomalies are identified and quickly isolated through health scores, and substitute data is generated in combination with weighted fusion of redundant data sources or prediction algorithms to ensure data integrity and continuity. The substitute data is incorporated into the data fusion process after deviation verification, thereby reducing the impact of sensor failures on the operation of rubber-tyred trackless vehicles.

[0131] After fault isolation, the dependency graph is dynamically reconstructed, the critical path is optimized, and the fusion results are analyzed using global consistency verification to ensure the stability and accuracy of multi-terminal data fusion. At the same time, different signals are generated based on the verification and analysis results, and the signal triggers the strategy management module to achieve intelligent control of data fusion and task allocation, thereby improving the real-time adaptability and robustness of the multi-terminal data fusion process of trackless rubber-tyred vehicles, and providing support for high-demand scenarios such as mining transportation.

[0132] The above formulas are all dimensionless and numerical calculations. The formula is a formula that is closest to the actual situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0133] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0134] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0135] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0136] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0137] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A multi-terminal data fusion system for rubber-tyred trackless vehicles, characterized by: It includes a path quantification module, a device health classification module, an alternative optimization module, and a consistency verification module, and each module is connected through signals; The path quantification module is used to calculate the dependency weights by combining the impact metrics of data correlation, time synchronization and environmental correlation, quantify the dependency of sensors through graph theory methods, and identify the critical paths; The device health classification module is used to define health indicators through signal-to-noise ratio, output volatility and response delay, use sliding windows to capture short-term changes, detect multi-dimensional anomalies based on Mahalanobis distance, generate health scores for sensors and classify them; The alternative optimization module is used to isolate faulty sensors with substandard health scores, fuse erroneous data with redundant data sources to generate alternative data, incorporate it into the data fusion process after verifying the deviation, reconstruct the dependency graph and optimize the path, and perform global consistency verification analysis on the fusion results; The consistency verification module is used to obtain the verification analysis information generated in the process of global consistency verification of the fusion results, analyze the verification analysis information and generate different signals, and perform policy management according to the generated different signals; It is used to isolate faulty sensors whose health scores do not meet the standards, and fuse the erroneous data with redundant data sources to generate alternative data, which is then incorporated into the data fusion process after verifying the deviation. The specific steps are as follows: Sensors whose health scores are less than the fault identification threshold are regarded as faulty sensors, and the output data of the faulty sensors are removed from the data stream and marked as isolated; After the faulty sensor is isolated, the redundant data sources available in the system are enabled to replace the output of the faulty sensor. The set of redundant data source nodes associated with the faulty sensor is determined through the dependency graph, and the outputs of the redundant data sources are weightedly fused to generate replacement data. When no redundant data source is available or the redundant data is incomplete, use historical data and time series prediction algorithms to generate replacement data for the faulty sensor; For sensor data with obvious periodicity or trend, the ARIMA model or variational autoencoder is used to determine the alternative data. For dynamically changing sensor data, the LSTM model or Transformer model is used to determine the alternative data. Fuse the substitute data with the current operating data, and ensure the reliability of the substitute data through a verification mechanism, and evaluate the deviation of the substitute data based on the current sensor data fusion results; When the deviation of the substitute data is less than the tolerance threshold, the substitute data is confirmed to be valid. After verification, the substitute data is included in the fusion process, and the substitution strategy is adjusted in real time according to the fusion result; Reconstruct the dependency graph and optimize the path, and perform a global consistency check analysis on the fusion results, including the following steps: Re-evaluate and update the weights in the dependency graph using the latest sensor health status and alternative data results; Based on the updated dependency graph, the critical path is dynamically readjusted, and the sensor dependency chain is adjusted using the shortest path or minimum spanning tree; After completing the dynamic adjustment of the dependency chain, the multi-terminal data fusion process is globally optimized and the fusion results are globally checked for consistency.

2. A multi-terminal data fusion system for rubber-tyred trackless vehicles according to claim 1, characterized in that: It is used to calculate the dependency weight by combining the impact metrics of data correlation, time synchronization and environmental correlation, quantify the dependency relationship of sensors through graph theory methods, and identify the critical path. The specific steps are as follows: The sensors of rubber-tyred trackless vehicles are modeled by graph theory, each sensor is regarded as a node, and the data dependency is regarded as an edge. The weight of the dependency is determined by the data relevance, time synchronization and environmental relevance. Use mutual information to measure the correlation of data, use dynamic time warping to calculate the synchronization measurement of time series data, and use distance weighting method to measure environmental correlation; Aggregate the correlation measure, synchronization measure, and environmental correlation measure to form a dependency graph; The path with the largest sum of weights in the dependency chain in the dependency graph is taken as the critical path.

3. A multi-terminal data fusion system for rubber-tyred trackless vehicles according to claim 2, characterized in that: It is used to define health indicators through signal-to-noise ratio, output fluctuation and response delay, use sliding window to capture short-term changes, detect multi-dimensional anomalies based on Mahalanobis distance, generate health scores for sensors and classify them. The specific steps include: Extract health indicators, including signal-to-noise ratio, data output volatility, and response delay; The sliding window method is used to observe the short-term changes of each health indicator to determine whether there are any abnormalities; If an abnormality occurs, a multi-dimensional anomaly detection method is used for identification. The multi-dimensional data of the sensor health status is used to form a feature vector. The detection value is determined by the Mahalanobis distance detection anomaly. When the detection value is greater than the set abnormal threshold, the sensor status is determined to be abnormal. Based on the anomaly detection results of monitoring, a health score is generated for each sensor, and the status is classified according to the health score; The status classification is based on the health score setting rules. A health score greater than or equal to 0.8 indicates a good health status, a health score greater than or equal to 0.5 and less than 0.8 indicates a status degradation warning, and a health score less than 0.5 indicates a serious fault warning.

4. A multi-terminal data fusion system for rubber-tyred trackless vehicles according to claim 3, characterized in that: Used to obtain the verification analysis information generated during the global consistency verification of the fusion results, including the following steps: Obtain verification and analysis information during the scheduling optimization process, including collaborative impact information and fusion matching information; The collaborative impact information includes the sensor collaborative index, and the fusion matching information includes the fusion deviation consistency index; The obtained sensor synergy index and fusion deviation consistency index are combined to generate a fusion state coefficient; The sensor synergy index and fusion bias consistency index are positively correlated with the fusion state coefficient.

5. A multi-terminal data fusion system for rubber-tyred trackless vehicles according to claim 4, characterized in that: Analyze and verify the analysis information and generate different signals, and perform strategy management based on the generated different signals, including the following steps: Compare the generated fusion state coefficient with the set fusion evaluation threshold; If the fusion state coefficient is greater than or equal to the fusion evaluation threshold, a multi-terminal fusion stability signal is generated, indicating that the output data fluctuation is stable and no additional adjustment is required; If the fusion state coefficient is less than the fusion evaluation threshold, a multi-terminal fusion fluctuation signal is generated, indicating that a deviation has occurred in the data fusion and timely optimization is required.

Images