Flight confidence calculation method and system based on multi-dimensional sensing data

Abstract

The application discloses a flight confidence calculation method and system based on multidimensional sensing data, and the method comprises the following steps: data acquisition, multidimensional feature extraction, establishment of safety value mapping, multidimensional confidence fusion calculation and decision feedback. The application belongs to the technical field of data processing, and specifically refers to a flight confidence calculation method and system based on multidimensional sensing data. According to the scheme, dynamic resource allocation is adopted to guarantee the quality of core data acquisition, dynamic Kalman filtering is adopted to reduce noise, high-value historical data is screened, and the double mechanisms of multi-sensor cross calibration and Bayesian inference verification are adopted to output an optimal spatial parameter estimation set and a corresponding confidence interval, so that the accuracy of confidence evaluation is improved. The safety value target is defined, the cause-and-effect chain is constructed, the cause-and-effect contribution degree of features to the safety value is quantified, the direct correlation between the features and the safety value is established, the multidimensional confidence is dynamically fused to generate the final flight confidence, and the problem that the confidence is high but the decision is unsafe is avoided.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically a method and system for calculating flight confidence based on multi-dimensional sensor data. Background Technology

[0002] Flight confidence level calculation methods utilize multi-source heterogeneous data during flight, extract key influencing factors through feature engineering, and combine machine learning models for dynamic quantitative calculation of confidence levels. The aim is to assess in real-time the reliability of flight mission execution, the safety of equipment operation, and the accuracy of target achievement, enabling timely measures such as process parameter optimization, equipment status adjustment, and flight path correction. However, existing flight confidence level calculation methods suffer from several drawbacks. First, they lack real-time acquisition of multi-dimensional sensor data and dynamic filtering and analysis of historical data, leading to insufficient accuracy in confidence level assessments. This can result in flight risks due to misjudgments of spatial parameters, limiting the reliability of flight equipment and navigation efficiency. Second, the calculation process over-relies on data, failing to establish a direct link with the core value proposition of flight safety, resulting in high flight confidence levels but unsafe flight decisions. Summary of the Invention

[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a flight confidence calculation method and system based on multi-dimensional sensor data. It addresses the technical problems of insufficient accuracy in confidence assessment due to the lack of real-time acquisition of multi-dimensional sensor data and dynamic filtering and analysis of historical data, which can easily lead to flight risks caused by misjudgment of spatial parameters and limit the reliability and navigation efficiency of flight equipment. The method employs dynamic resource allocation to ensure the quality of core data acquisition, dynamic Kalman filtering to reduce noise, accurate filtering of high-value historical data, and error correction through feature fusion. A dual mechanism of multi-sensor cross-calibration and Bayesian inference verification outputs the highest confidence score. This approach optimizes the spatial parameter estimation set and corresponding confidence intervals to improve the accuracy of confidence assessment and reduce spatial parameter misjudgments and flight risks. Addressing the technical issue of excessive reliance on data in flight confidence calculations without establishing a direct link to core flight safety values, resulting in high flight confidence but unsafe flight decisions, this approach defines safety value objectives, constructs causal chains, quantifies the causal contribution of features to safety value, establishes a direct link between features and safety value, and dynamically fuses confidence scores to generate the final flight confidence score. This ensures that confidence calculations meet core safety requirements and avoids the problem of high confidence scores but unsafe decisions.

[0004] The technical solution adopted by this invention is as follows: The flight confidence calculation method based on multi-dimensional sensor data provided by this invention includes the following steps:

[0005] Step S1: Data Acquisition;

[0006] Step S2: Multi-dimensional feature extraction;

[0007] Step S3: Establish a security value mapping;

[0008] Step S4: Multi-dimensional confidence fusion calculation;

[0009] Step S5: Decision Feedback.

[0010] Further, in step S1, the data acquisition includes the following steps:

[0011] Step S11: Collect multi-source data, including the following steps:

[0012] Step S111: Perform scene recognition and risk assessment. The edge device judges the current scene in real time through flight parameters and assigns the corresponding risk level coefficient. Simultaneously, it calls the security correlation of each data stream in the current scene from the cloud.

[0013] Step S112: Dynamic resource allocation, calculate the resource allocation weight of each data stream;

[0014] Step S113: Synchronous acquisition of multi-source data. Multi-source data is synchronously acquired through sensors, databases, and external interfaces according to resource allocation weights. The multi-source data includes real-time status data, environmental interference data, historical benchmark data, and task target data.

[0015] Step S114: Data quality gating, which involves assessing the quality of the collected multi-source data and selecting qualified data.

[0016] Step S12: Dynamic noise filtering. Input qualified data and use Kalman filtering to filter noise based on different sensor characteristics. Dynamically adjust the process noise covariance and observation noise covariance based on scene features.

[0017] Calculate the filtering residual to judge the filtering effect, adjust the process noise covariance and observation noise covariance in reverse, continuously optimize the filtering effect, and obtain the filtered standard data.

[0018] Step S13: Dynamically filter historical data, establish scenario-matching data filtering rules, use filtered standard data as the real-time data benchmark, perform similarity matching with historical data, and filter relevant data with the same scenario, risk level, and equipment status from the historical database according to the current flight scenario, while eliminating irrelevant historical data; use a sliding time window and scenario feature matching algorithm to dynamically adjust the window size; including the following steps:

[0019] Step S131: Two-dimensional similarity matching, including scene matching and working condition matching, calculate the two-dimensional similarity value;

[0020] Step S132: Dynamic window size adjustment, calculating the current window size based on the two-dimensional similarity value;

[0021] Step S133: Historical data filtering. Extract historical data from the historical database that are within the dynamic window and have a two-dimensional similarity greater than or equal to 0.6. Calculate the weight of each historical data using the time decay formula and retain high-value historical data with high weights.

[0022] Step S134: Scene correlation verification. Compare the filtered high-value historical data with the real-time data and calculate the deviation rate. If the average deviation rate is less than 10%, the data correlation is confirmed to be qualified and used for subsequent confidence calculation. If the average deviation rate is greater than or equal to 10%, the window is further narrowed and the data is re-filtered to finally obtain the dynamically filtered historical features.

[0023] Further, in step S2, the multi-dimensional feature extraction includes the following steps:

[0024] Step S21: Feature extraction. Spatiotemporal features are extracted from the filtered standard data using a CNN-LSTM hybrid model. Device status features and environmental features are extracted using statistical methods, and a structured feature set is output.

[0025] Step S22: Based on the structured feature set, the sensor error is corrected by spatial correlation and fusion of GPS and IMU data, and the consistency coefficient is calculated by comparing real-time features with dynamically filtered historical features through temporal correlation. Finally, the feature set after correlation optimization is output.

[0026] Step S23: Perform weighted feature fusion on the optimized feature set, assign higher weights to security-related features and lower weights to secondary features to obtain the fused global feature vector;

[0027] Step S24: Precise verification of spatial parameters, including the following steps:

[0028] Step S241: Multi-sensor cross-calibration, cross-validating the obstacle distance measured by radar with the distance calculated by the visual sensor. If the deviation between the two exceeds the threshold, it is corrected by combining historical data trends.

[0029] Step S242: Construct a feature verification model based on Bayesian inference, including the following:

[0030] Formula for calculating prior probability distribution based on historical data;

[0031] Calculate the likelihood probability and fit the nonlinear mapping relationship using a neural network;

[0032] According to Bayes' theorem, the posterior probability of a spatial parameter is proportional to the prior probability and the likelihood probability. Since the posterior probability distribution is complex, the maximum a posteriori estimation is used to solve for the optimal spatial parameter estimation set; and based on the variance of the posterior probability distribution, the 95% confidence interval is calculated.

[0033] Furthermore, in step S3, establishing a safety value mapping, and establishing a direct correlation between multi-dimensional features and the core value of flight safety, includes the following steps:

[0034] Step S31: Define security value objectives;

[0035] Step S32: Establish a causal chain and use causal inference operators to verify the true causal relationship between features and security value; the features are core features directly related to security value extracted from the fused global feature vector.

[0036] Based on causal weights, the causal contribution of features to security value is quantified;

[0037] Based on the magnitude of their causal contribution, the features are divided into core security features, important security features, and general security features, and are subject to precise control at different risk management levels.

[0038] Further, in step S4, the multi-dimensional confidence fusion calculation includes the following steps:

[0039] Step S41: Multi-dimensional confidence decomposition calculation, calculate data feature confidence, security value confidence and historical matching confidence;

[0040] Step S42: Confidence fusion calculation. A dynamic weight fusion and entropy weight correction mechanism is used to fuse the confidence of the three dimensions to generate the final flight confidence.

[0041] Furthermore, in step S5, the decision feedback specifically involves making a decision based on the final flight confidence level, and then inputting the actual flight safety status after the decision is executed back into the model adjustment parameters for feedback optimization.

[0042] The flight confidence calculation system based on multi-dimensional sensor data provided by this invention includes a data acquisition module, a multi-dimensional feature extraction module, a safety value mapping module, a multi-dimensional confidence fusion calculation module, and a decision feedback module.

[0043] The data acquisition module specifically collects multi-source data synchronously according to resource allocation weights. After quality gating to screen qualified data, it uses Kalman filtering with dynamic covariance adjustment for noise filtering and optimizes the filtering effect by calculating the filtering residual. Based on two-dimensional similarity matching, dynamic window adjustment, time decay weighting, and scene correlation verification, it selects high-value historical features from the historical database, and finally obtains the filtered standard data and dynamically selected historical features. The data is then sent to the multi-dimensional feature extraction module.

[0044] The multi-dimensional feature extraction module specifically extracts features from the filtered standard data to obtain a structured feature set. Based on the structured feature set, it calculates the consistency coefficient by spatially fusing GPS and IMU data and temporally comparing real-time features with dynamically filtered historical features. Finally, it outputs the feature set after association optimization. Weighted feature fusion is used to obtain the fused global feature vector. Finally, through a dual mechanism of multi-sensor cross-calibration and Bayesian inference verification, it outputs the optimal spatial parameter estimation set and the corresponding confidence interval. The data is then sent to the security value mapping module.

[0045] The security value mapping module specifically defines security value objectives, constructs causal chains and quantifies the causal contribution of features to security value, and then classifies features into levels to implement precise risk control; the data is then sent to the multi-dimensional confidence fusion calculation module.

[0046] The multi-dimensional confidence fusion calculation module specifically calculates the confidence of data features, the confidence of safety value, and the confidence of historical matching, respectively, and uses a dynamic weight fusion and entropy weight correction mechanism to perform confidence fusion to generate the final flight confidence; the data is then sent to the decision feedback module.

[0047] The decision feedback module specifically makes decisions based on the final flight confidence level and inputs the actual safety status after the decision is executed back into the model for adjustment and optimization.

[0048] The beneficial results achieved by the present invention using the above solution are as follows:

[0049] (1) To address the technical problem that the lack of real-time acquisition of multi-dimensional sensor data and dynamic screening and analysis of historical data leads to insufficient confidence assessment accuracy, flight risks caused by misjudgment of spatial parameters, and limitations on the reliability and navigation efficiency of flight equipment, dynamic resource allocation is adopted to ensure the quality of core data acquisition, dynamic Kalman filtering is used to reduce noise, high-value historical data is accurately screened, and the optimal spatial parameter estimation set and corresponding confidence interval are output through the dual mechanism of feature fusion to correct errors, multi-sensor cross-calibration and Bayesian inference verification, thereby improving the accuracy of confidence assessment and reducing spatial parameter misjudgment and flight risks;

[0050] (2) To address the technical problem that the flight confidence calculation process relies too much on data and does not establish a direct connection with the core value requirements of flight safety, resulting in high flight confidence but unsafe flight decisions, we adopt the following approach: define safety value objectives, construct causal chains and quantify the causal contribution of features to safety value, establish a direct connection between features and safety value, and perform dynamic fusion of confidence to generate the final flight confidence, so that the confidence calculation meets the core safety requirements and avoids the problem of high confidence but unsafe decisions. Attached Figure Description

[0051] Figure 1 A flowchart illustrating the flight confidence calculation method based on multi-dimensional sensor data provided by this invention;

[0052] Figure 2 This is a schematic diagram of the flight confidence calculation system based on multi-dimensional sensor data provided by the present invention.

[0053] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0054] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0055] Example 1, see Figure 1 The present invention provides a method for calculating flight confidence based on multi-dimensional sensor data, which includes the following steps:

[0056] Step S1: Data collection, specifically, multi-source data is collected synchronously according to resource allocation weights. After qualified data is screened by quality gating, noise is filtered by Kalman filtering with dynamic adjustment of covariance. The filtering effect is optimized by calculating the filtering residual. Based on two-dimensional similarity matching, dynamic window adjustment, time decay weighting and scene correlation verification, high-value historical features are screened from the historical database. Finally, the filtered standard data and dynamically screened historical features are obtained.

[0057] Step S2: Multi-dimensional feature extraction, specifically, feature extraction is performed on the filtered standard data to obtain a structured feature set. Based on the structured feature set, the consistency coefficient is calculated by spatially fusing GPS and IMU data and temporally comparing real-time features with dynamically filtered historical features. Finally, the feature set after association optimization is output. Weighted feature fusion is used to obtain the fused global feature vector. Finally, the optimal spatial parameter estimation set and the corresponding confidence interval are output through a dual mechanism of multi-sensor cross-calibration and Bayesian inference verification.

[0058] Step S3: Establish a security value mapping, specifically by defining security value objectives, constructing causal chains and quantifying the causal contribution of features to security value, and then classifying features into levels to implement precise risk control.

[0059] Step S4: Multi-dimensional confidence fusion calculation, specifically, calculating the confidence of data features, confidence of safety value, and confidence of historical matching respectively, and using dynamic weight fusion and entropy weight correction mechanism to fuse the confidence and generate the final flight confidence;

[0060] Step S5: Decision feedback, specifically, making decisions based on the final flight confidence level, and inputting the actual safety status after the decision is executed back into the model for adjustment and optimization.

[0061] Example 2, based on the above example, in step S1, the data acquisition includes the following steps:

[0062] Step S11: Collect multi-source data, including the following steps:

[0063] Step S111: Perform scene recognition and risk assessment. The edge device uses flight parameters (obstacle distance, attitude change rate, etc.) to determine the current flight scenario (cruise, obstacle avoidance, emergency braking) in real time and assigns corresponding risk level coefficients. Simultaneously, it calls the safety correlation of each data stream under the current flight scenario from the cloud.

[0064] Step S112: Dynamic resource allocation. Calculate the resource allocation weight for each data stream. For example, allocate 80% of the edge computing power and bandwidth to core data streams with a security correlation coefficient greater than or equal to 0.5 (such as flight attitude and equipment operating parameters), and allocate 20% of the edge computing power and bandwidth to auxiliary data streams (such as temperature and humidity). The formula used is as follows:

[0065] ;

[0066] In the formula, This represents the resource allocation weight of data stream d1 under scenario s. This represents the risk level coefficient under scenario s (cruising). =1, obstacle avoidance =3, Emergency Braking =5), n1 represents the security correlation between data stream d1 and scene s, and n1 represents the total number of data streams.

[0067] Step S113: Synchronous acquisition of multi-source data through sensors, databases, and external interfaces, with weights allocated according to resources. Simultaneously collect multi-source data, increasing the core data stream acquisition frequency to 100Hz and reducing the auxiliary data stream to 10Hz; the multi-source data includes real-time status data, environmental interference data, historical benchmark data, and task target data;

[0068] The real-time status data includes flight attitude, flight parameters, and equipment operating parameters;

[0069] The environmental interference data includes wind speed, air pressure, temperature, and humidity;

[0070] The historical benchmark data is obtained by calling up historical datasets of flight missions and selecting data with a similarity of ≥80% to the current mission conditions as the benchmark.

[0071] The mission objective data includes a preset flight path, mission priority, and allowable error threshold.

[0072] Step S114: Data quality gating. The quality of the collected multi-source data is assessed, retaining qualified data with a quality score greater than or equal to 0.6 and discarding distorted data with a score less than 0.6. The formula used is as follows:

[0073] ;

[0074] In the formula, This represents the quality score of data stream d1. This represents the weighting coefficient, with a value of 0.7, used to prioritize signal stability. This indicates the signal strength of data stream d1. This represents the maximum signal strength of data stream d1. This represents the real-time collected value of data stream d1. This represents the normal reference value for data stream d1;

[0075] Step S12: Dynamic noise filtering. Input qualified data and use Kalman filtering for noise filtering based on different sensor characteristics. Dynamically adjust the process noise covariance and observation noise covariance based on scene features to improve noise filtering accuracy and make it more suitable for complex flight environments. The formula used is as follows:

[0076] ;

[0077] ;

[0078] In the formula, This represents the process noise covariance at time k. This represents the covariance of the basic process noise. The observation noise covariance at time k is represented. This represents the covariance of the basic observation noise. Indicates the interference adaptation coefficient, ( , (obtained through training with historical data). This represents the intensity of environmental disturbance at time k. This represents the distance correction coefficient between the sensor and the target at time k;

[0079] The filtering effect is judged by calculating the filter residuals. The process noise covariance and observation noise covariance are adjusted in reverse to continuously optimize the filtering effect. When the filter residuals at multiple consecutive time points are stable within the preset accuracy threshold, the filtering is considered to have converged, and the filtered standard data is obtained. The formula used is as follows:

[0080] ;

[0081] ;

[0082] ;

[0083] In the formula, This represents the filtered residual at time k. Let H represent the actual observed data at time k, and let H represent the observation matrix, which is used to map the state prediction values ​​in the filtering model to the observation dimension, thereby achieving dimensionality matching between the state space and the observation space. This represents the predicted state at time k based on the data at time k-1. This represents the process noise covariance updated at time k+1. This indicates the correction step size for process noise, to avoid sudden parameter changes. The product of the filtered residuals at time k is represented by T, where T denotes the transpose sign. This represents the observation noise covariance updated at time k+1. Indicates the correction step size for observation noise;

[0084] Step S13: Dynamically filter historical data, establish scene-matching data filtering rules, use filtered standard data as the real-time data benchmark, perform similarity matching with historical data, and filter data with the same scene, risk level, and equipment status from the historical database according to the current flight scenario, while eliminating irrelevant historical data; use a sliding time window and scene feature matching algorithm to dynamically adjust the window size to ensure the correlation between historical data and the real-time scenario; including the following steps:

[0085] Step S131: Two-dimensional similarity matching, including scene matching and working condition matching. Scene matching compares the current scene with historical data scene labels to obtain scene type similarity. Working condition matching extracts real-time working condition parameters and historical working condition parameters, calculates the working condition parameter similarity using cosine similarity, and calculates the two-dimensional similarity value using the following formula:

[0086] ;

[0087] ;

[0088] In the formula, For scene weighting coefficients, Indicates two-dimensional similarity. Indicates the similarity of scene types. This represents the similarity of operating condition parameters, where m represents the number of dimensions of the operating condition parameters. This represents the real-time operating condition parameters of the i-th dimension. This represents the historical operating condition parameters of the i-th dimension;

[0089] Step S132: Dynamic window size adjustment. The current window size is calculated based on the two-dimensional similarity value. When the two-dimensional similarity is high, it indicates that the current scene and working conditions are highly consistent with historical data. The window expands towards the maximum window to capture more historical data with high matching degree, thereby increasing the sample size and representativeness of the benchmark data. The formula used is as follows:

[0090] ;

[0091] In the formula, W represents the current size of the sliding window. Indicates the maximum window size. Indicates the minimum window size;

[0092] Step S133: Historical data filtering. Extract historical data from the historical database that are within the dynamic window and have a two-dimensional similarity greater than or equal to 0.6. Calculate the weight of each historical data point using the time decay formula, and retain high-value historical data with higher weights. The formula used is as follows:

[0093] ;

[0094] In the formula, This represents the weight of the historical data at time t. Indicates the attenuation coefficient. Indicates the current moment. Indicates the time of historical data collection;

[0095] Step S134: Scene correlation verification. Compare the filtered high-value historical data with the real-time data and calculate the deviation rate. If the average deviation rate is less than 10%, the data correlation is confirmed to be qualified and used for subsequent confidence calculation. If the average deviation rate is greater than or equal to 10%, the window is further narrowed and the data is re-filtered to finally obtain the dynamically filtered historical features.

[0096] Example 3, based on the above examples, in step S2, the multi-dimensional feature extraction includes the following steps:

[0097] Step S21: Feature extraction. Spatiotemporal features are extracted from the filtered standard data using a CNN-LSTM hybrid model. Device status features and environmental features are extracted using statistical methods, and a structured feature set is output.

[0098] The spatiotemporal features include real-time position accuracy, attitude stability, flight path deviation rate, and relative motion trajectory of obstacles;

[0099] The equipment status characteristics include sensor data reliability, engine operating parameters, and avionics system response delay.

[0100] The environmental characteristics include wind speed, wind direction, visibility, and air traffic control compliance boundaries;

[0101] Step S22: Based on the structured feature set, GPS and IMU data are fused through spatial correlation to correct sensor errors. Real-time features are compared with dynamically filtered historical features through temporal correlation to calculate the consistency coefficient. Finally, the optimized feature set is output, including the following:

[0102] Based on the spatial core features extracted from the structured feature set and the device reliability features, accurate fusion of GPS location data and IMU attitude data is achieved, correcting the measurement errors of a single sensor; the formula used is as follows:

[0103] ;

[0104] ;

[0105] ;

[0106] In the formula, This represents the data fused from GPS and IMU at time t. Represents the spatial weights of GPS. Indicates the spatial weights of the IMU. This represents the initial GPS position at time t. This represents the raw attitude data of the IMU at time t. This represents the IMU attitude data after correction at time t. The Kalman gain at time t reflects the strength of the IMU's error correction relative to GPS. This represents the GPS measurement error covariance at time t. This represents the measurement error covariance of the IMU at time t;

[0107] Based on the full-dimensional features of the structured feature set, real-time features are compared with dynamically filtered historical features to calculate feature consistency; the formula used is as follows:

[0108] ;

[0109] In the formula, Represents the feature consistency coefficient at time t ( This indicates that the real-time features and historical features are in good agreement. (This indicates poor consistency in features, suggesting a risk of anomalies). Let represent the real-time feature of the k1-th dimension at time t. express The historical feature values ​​of the k1th dimension after dynamic filtering at any given time. This represents the temporal offset corresponding to the k1-th dimension feature. This represents the similarity threshold for the k1-th dimension feature. The weight of the k1th dimension feature is represented by M, and M represents the total number of feature dimensions participating in the consistency calculation.

[0110] The full-dimensional features that have undergone spatial error correction and temporal consistency verification are integrated to obtain the feature set after association optimization. The full-dimensional features represent spatiotemporal features, device status features, and environmental features.

[0111] Step S23: Perform weighted feature fusion on the optimized feature set, assigning higher weights to security-related features and lower weights to secondary features to obtain the fused global feature vector; the formula used is as follows:

[0112] ;

[0113] ;

[0114] ;

[0115] In the formula, This represents the global weight of the i-th type of feature (spatiotemporal feature F1, device state feature F2, and environmental feature F3). This represents the security priority coefficient for the i-th type of feature. The data reliability coefficient represents the i-th type of feature. This represents the dynamic security priority weighting coefficient, used to balance the proportion of security priority and data reliability in the feature weights. This represents the basic security priority coefficient. This represents the fused global feature vector, and Norm() represents Min-Max normalization. This represents the feature set after association optimization;

[0116] Step S24: Precise verification of spatial parameters. Through a dual mechanism of multi-sensor cross-calibration and Bayesian inference verification, the optimal spatial parameter estimation set and the corresponding 95% confidence interval are output, achieving precise spatial parameters and providing data support for flight control decisions and safety boundary determination; including the following steps:

[0117] Step S241: Multi-sensor cross-calibration. This involves cross-validating the obstacle distance measured by radar with the distance calculated by the visual sensor. If the deviation exceeds a threshold, historical data trends are used for correction to avoid misjudgments of spatial parameters caused by a single sensor. The formula used is as follows:

[0118] ;

[0119] ;

[0120] In the formula, This indicates the distance to the obstacle measured by radar. This represents the distance to the obstacle measured by the vision sensor, where dev represents the distance deviation. Indicates the deviation threshold. Indicates the weight of the radar sensor. Represents the weights of the visual sensors. This indicates the distance to obstacles after historical correction. The historical data weights are represented by , and D represents the calibrated obstacle distance.

[0121] Step S242: Construct a feature verification model based on Bayesian inference, including the following:

[0122] The formula for calculating the prior probability distribution based on historical data is as follows:

[0123] ;

[0124] In the formula, X represents the three-dimensional location parameters (longitude, latitude, altitude). Represents three-dimensional attitude parameters (roll angle, pitch angle, yaw angle). Indicates a normal distribution. This represents the historical mean and variance of the three-dimensional position parameters. This represents the historical mean and variance of the three-dimensional attitude parameters. This represents the historical mean and variance of obstacle distances. Represents the prior probability distribution of the three-dimensional position parameters. This represents the prior probability distribution of the three-dimensional attitude parameters. The prior probability distribution representing the distance to the obstacle;

[0125] The likelihood probability is calculated and a nonlinear mapping relationship is fitted using a neural network. The formula used is as follows:

[0126] ;

[0127] In the formula, This represents the likelihood probability, i.e., the probability given spatial parameters. Under the given conditions, the probability of observing the fused global feature vector is... This represents the Sigmoid activation function. The weight matrix of the neural network. This represents the bias term of the neural network. This represents a vector concatenation operation;

[0128] According to Bayes' theorem, the posterior probability of a spatial parameter is proportional to its prior probability and likelihood probability. Since the posterior probability distribution is complex, maximum a posteriori estimation is used to solve for the optimal spatial parameter estimation set. The formula used is as follows:

[0129] ;

[0130] ;

[0131] In the formula, This represents the posterior probability of the spatial parameters, i.e., given the observed fused global feature vector, the probability of the spatial parameters... The probability of taking the true value. Indicates the direct proportion sign. This represents the optimal set of spatial parameter estimates, including the optimal 3D position, 3D pose, and obstacle distances. This represents the spatial parameters required to maximize the objective function within the parentheses. Values, Represents the set of spatial parameters to be estimated;

[0132] The 95% confidence interval is calculated based on the variance of the posterior probability distribution using the following formula:

[0133] ;

[0134] In the formula, This represents the 95% confidence interval of the optimal spatial parameter estimation set. That is, given the observed fused global feature vectors, there is a 95% probability that the true values ​​of the spatial parameters fall within this interval. This represents the variance of the optimal estimate given that the fused global eigenvectors have been observed.

[0135] By performing the above operations, dynamic resource allocation is used to ensure the quality of core data acquisition, dynamic Kalman filtering reduces noise, high-value historical data is accurately screened, and the optimal spatial parameter estimation set and corresponding confidence interval are output through a dual mechanism of feature fusion to correct errors and multi-sensor cross-calibration and Bayesian inference verification. This improves the accuracy of confidence assessment, reduces spatial parameter misjudgment and flight risks, and solves the technical problem of insufficient accuracy of confidence assessment due to the lack of real-time acquisition of multi-dimensional sensor data and dynamic screening and analysis of historical data, which easily leads to flight risks caused by spatial parameter misjudgment and limits the reliability and navigation efficiency of flight equipment.

[0136] Example 4, based on the above examples, in step S3, establishing a safety value mapping, establishing a direct correlation between multi-dimensional features and the core value of flight safety, quantifying the actual contribution of each feature to safety value, replacing the traditional evaluation logic that only relies on data correlation, includes the following steps:

[0137] Step S31: Define safety value objectives, which include reducing collision risk, not exceeding compliance boundaries, and reducing equipment failure risk;

[0138] Step S32: Establish a causal chain, constructing a causal chain of feature-decision-safety value (e.g., obstacle distance feature-steering decision-collision risk reduction, compliance boundary distance feature-route adjustment decision-compliance probability improvement), and use the do operator of causal reasoning to verify the true causal relationship between features and safety value; the formula used is as follows:

[0139] ;

[0140] In the formula, This represents the causal transitivity entropy of the j-th feature to the c-th security value. Represents the expectation operator. This represents multi-dimensional features, which are derived from... The core features extracted are directly related to security value, such as obstacle distance features and compliance boundary distance features. The quantitative value representing the safety value target (c=1 corresponds to a reduction in collision risk, c=2 corresponds to an increase in compliance probability, and c=3 corresponds to a reduction in equipment failure risk). This indicates that the optimal value of the causal interference estimator is intervened, forcibly setting the value of the j-th feature to the optimal state in engineering terms. This indicates that the worst-case intervention of the causal interference estimator forces the value of the j-th feature to be set to the worst-case state in engineering.

[0141] Based on causal weights, the causal contribution of features to security value is quantified; the formula used is as follows:

[0142] ;

[0143] ;

[0144] In the formula, This represents the causal weight of the j-th feature on the c-th security value. J represents the causal contribution of the j-th feature to the c-th security value, and J represents the total number of features.

[0145] Based on the magnitude of causal contribution, the features are divided into core security features ( ), key safety features ( ), General safety features ( ), and carry out precise management according to different risk control levels.

[0146] Example 5, based on the above examples, in step S4, the multi-dimensional confidence fusion calculation, combining data feature accuracy, safety value contribution, and historical scenario matching degree, calculates the final flight confidence, including the following steps:

[0147] Step S41: Calculate the multi-dimensional confidence decomposition;

[0148] The confidence score of data features is calculated to reflect the reliability of the data at the data level; the formula used is as follows:

[0149] ;

[0150] ;

[0151] ;

[0152] In the formula, Indicates the confidence level of data features. This represents the data weight of the j-th feature. This represents the relative measurement error of the j-th feature. This represents the data missing rate of the j-th feature;

[0153] The confidence score for security value is calculated, reflecting the reliability of the contribution of the decision to security value; the formula used is as follows:

[0154] ;

[0155] ;

[0156] In the formula, Indicates the confidence level of the security value. Indicates the contextualized security value weight. This represents the maximum causal transitivity of all features to the c-th security value. This indicates the causal consistency between the j-th feature and the c-th security value. This indicates the reliability of the association between the j-th feature and the c-th security value;

[0157] The formula for calculating the confidence score of historical matches is as follows:

[0158] ;

[0159] In the formula, Indicates the confidence level of historical matches. Represents the real-time global feature vector. This represents the historical feature vector after the k2th dynamic filtering. This represents the security label of the k2th dynamically filtered historical sample, where K represents the total number of dynamically filtered historical feature samples, and k2 represents the index of the dynamically filtered historical sample. This represents the similarity value between the real-time global feature vector and the dynamically filtered historical feature vector.

[0160] Step S42: Confidence fusion calculation. A dynamic weight fusion and entropy weight correction mechanism is used to fuse the confidence scores of the three dimensions. This not only reflects the core value of each dimension but also corrects the weights for their rationality through data entropy, generating the final flight confidence score. The formula used is as follows:

[0161] ;

[0162] ;

[0163] ;

[0164] In the formula, Indicates the final flight confidence level. This indicates that the weights are dynamically adjusted. Indicates the confidence level of a single dimension (d=1 corresponds to) d=2 corresponds to d=3 corresponds to ), This represents the pre-defined baseline weights for the dimensions. Let N represent the information entropy of the d-th dimension, and let N represent the number of historical confidence samples. This represents the information entropy normalization factor. This represents the single-dimensional confidence score at the nth time step, where n represents the historical time step index.

[0165] By performing the above operations, defining safety value objectives, constructing causal chains and quantifying the causal contribution of features to safety value, establishing a direct correlation between features and safety value, and dynamically fusing confidence scores to generate the final flight confidence score, the confidence score calculation meets the core safety requirements, avoiding the problem of high confidence scores but unsafe decision-making. This solves the technical problem that the flight confidence score calculation process over-relies on data and fails to establish a direct correlation with the core value requirements of flight safety, resulting in high flight confidence scores but unsafe flight decisions.

[0166] Example 6, based on the above examples, in step S5, the decision feedback specifically involves making a decision based on the final flight confidence level;

[0167] When the final flight confidence level is [0.8, 1], it is considered high confidence, and the current flight strategy is maintained.

[0168] When the final flight confidence level is within [0.5, 0.8), it is considered a medium confidence level, and a safety warning is activated to closely monitor the core characteristics.

[0169] When the final flight confidence level is in the range [0, 0.5), it is considered low confidence, triggering emergency adjustments (such as route correction, speed reduction, return to base, etc.), and the data is recorded for model feedback optimization.

[0170] The actual flight safety status after the decision is implemented is fed back into the model to adjust parameters, continuously improving the accuracy of confidence assessment.

[0171] Example 7, see Figure 2 Based on the above embodiments, the flight confidence calculation system based on multi-dimensional sensor data provided by the present invention includes a data acquisition module, a multi-dimensional feature extraction module, a safety value mapping module, a multi-dimensional confidence fusion calculation module, and a decision feedback module.

[0172] The data acquisition module specifically collects multi-source data synchronously according to resource allocation weights. After quality gating to screen qualified data, it uses Kalman filtering with dynamic covariance adjustment for noise filtering and optimizes the filtering effect by calculating the filtering residual. Based on two-dimensional similarity matching, dynamic window adjustment, time decay weighting, and scene correlation verification, it selects high-value historical features from the historical database, and finally obtains the filtered standard data and dynamically selected historical features. The data is then sent to the multi-dimensional feature extraction module.

[0173] The multi-dimensional feature extraction module specifically extracts features from the filtered standard data to obtain a structured feature set. Based on the structured feature set, it calculates the consistency coefficient by spatially fusing GPS and IMU data and temporally comparing real-time features with dynamically filtered historical features. Finally, it outputs the feature set after association optimization. Weighted feature fusion is used to obtain the fused global feature vector. Finally, through a dual mechanism of multi-sensor cross-calibration and Bayesian inference verification, it outputs the optimal spatial parameter estimation set and the corresponding confidence interval. The data is then sent to the security value mapping module.

[0174] The security value mapping module specifically defines security value objectives, constructs causal chains and quantifies the causal contribution of features to security value, and then classifies features into levels to implement precise risk control; the data is then sent to the multi-dimensional confidence fusion calculation module.

[0175] The multi-dimensional confidence fusion calculation module specifically calculates the confidence of data features, the confidence of safety value, and the confidence of historical matching, respectively, and uses a dynamic weight fusion and entropy weight correction mechanism to perform confidence fusion to generate the final flight confidence; the data is then sent to the decision feedback module.

[0176] The decision feedback module specifically makes decisions based on the final flight confidence level and inputs the actual safety status after the decision is executed back into the model for adjustment and optimization.

[0177] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0178] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0179] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for calculating flight confidence based on multi-dimensional sensor data, characterized in that: The method includes the following steps: Step S1: Data collection, specifically, multi-source data is collected synchronously according to resource allocation weights. After qualified data is screened by quality gating, noise is filtered by Kalman filtering with dynamic adjustment of covariance. The filtering effect is optimized by calculating the filtering residual. Based on two-dimensional similarity matching, dynamic window adjustment, time decay weighting and scene correlation verification, high-value historical features are screened from the historical database. Finally, the filtered standard data and dynamically screened historical features are obtained. Step S2: Multi-dimensional feature extraction, specifically, feature extraction is performed on the filtered standard data to obtain a structured feature set. Based on the structured feature set, the consistency coefficient is calculated by spatially fusing GPS and IMU data and temporally comparing real-time features with dynamically filtered historical features. Finally, the feature set after association optimization is output. Weighted feature fusion is used to obtain the fused global feature vector. Finally, the optimal spatial parameter estimation set and the corresponding confidence interval are output through a dual mechanism of multi-sensor cross-calibration and Bayesian inference verification. Step S3: Establish a security value mapping, specifically by defining security value objectives, constructing causal chains and quantifying the causal contribution of features to security value, and then classifying features into levels to implement precise risk control. Step S4: Multi-dimensional confidence fusion calculation, specifically, calculating the confidence of data features, confidence of safety value, and confidence of historical matching respectively, and using dynamic weight fusion and entropy weight correction mechanism to fuse the confidence and generate the final flight confidence; Step S5: Decision feedback, specifically, making decisions based on the final flight confidence level, and inputting the actual safety status after the decision is executed back into the model for adjustment and optimization; Step S1 includes the following steps: Step S12: Dynamic noise filtering. Input qualified data and use Kalman filtering to filter noise based on different sensor characteristics. Dynamically adjust the process noise covariance and observation noise covariance based on scene features. Calculate the filtering residual to judge the filtering effect, adjust the process noise covariance and observation noise covariance in reverse, continuously optimize the filtering effect, and obtain the filtered standard data. Step S13: Dynamically filter historical data, establish scenario-matching data filtering rules, use filtered standard data as the real-time data benchmark, perform similarity matching with historical data, and filter relevant data with the same scenario, risk level, and equipment status from the historical database according to the current flight scenario, while eliminating irrelevant historical data; use a sliding time window and scenario feature matching algorithm to dynamically adjust the window size; including the following steps: Step S131: Two-dimensional similarity matching, including scene matching and working condition matching, calculate the two-dimensional similarity value; Step S132: Dynamic window size adjustment, calculating the current window size based on the two-dimensional similarity value; Step S133: Historical data filtering. Extract historical data from the historical database that are within the dynamic window and have a two-dimensional similarity greater than or equal to 0.

6. Calculate the weight of each historical data using the time decay formula to obtain high-value historical data. Step S134: Scene correlation verification. Compare the filtered high-value historical data with the real-time data and calculate the deviation rate. If the average deviation rate is less than 10%, the data correlation is confirmed to be qualified and used for subsequent confidence calculation. If the average deviation rate is greater than or equal to 10%, the window is further narrowed and the data is re-filtered to finally obtain the dynamically filtered historical features. Step S2 includes the following steps: Step S23: Perform weighted feature fusion on the optimized feature set, assigning high weights to security-related features and low weights to secondary features to obtain the fused global feature vector; Step S24: Precise verification of spatial parameters, including the following steps: Step S241: Multi-sensor cross-calibration, cross-validating the obstacle distance measured by radar with the distance calculated by the visual sensor. If the deviation between the two exceeds the threshold, it is corrected by combining historical data trends. Step S242: Construct a feature verification model based on Bayesian inference, including the following: Formula for calculating prior probability distribution based on historical data; Calculate the likelihood probability and fit a nonlinear mapping relationship using a neural network; the likelihood probability represents the probability in a given space. Under the given conditions, the probability of observing the fused global feature vector; where X represents the three-dimensional position parameter, This represents the three-dimensional attitude parameters, where D represents the calibrated obstacle distance; According to Bayes' theorem, the posterior probability of a spatial parameter is proportional to its prior probability and likelihood probability. Maximum a posteriori (MAP) estimation is used to solve for the optimal spatial parameter estimation set. Based on the variance of the posterior probability distribution, the 95% confidence interval of the optimal spatial parameter estimation set is calculated. The posterior probability of the spatial parameter represents the probability of the spatial parameter being equal to or less than the expected probability of the fused global feature vector. The probability of taking the true value; the 95% confidence interval of the optimal spatial parameter estimation set means that, given the observation of the fused global feature vector, there is a 95% probability that the true value of the spatial parameter falls within this interval; Step S3 includes the following steps: Step S31: Define security value objectives; Step S32: Establish a causal chain and use causal inference operators to verify the true causal relationship between features and security value; the features are core features directly related to security value extracted from the fused global feature vector. Based on causal weights, the causal contribution of features to security value is quantified; Based on the magnitude of their causal contribution, the features are divided into core security features, important security features, and general security features, and are subject to precise control at different risk management levels.

2. The flight confidence calculation method based on multi-dimensional sensor data according to claim 1, characterized in that: In step S2, the multi-dimensional feature extraction further includes the following steps: Step S21: Feature extraction. Spatiotemporal features are extracted from the filtered standard data using a CNN-LSTM hybrid model. Device status features and environmental features are extracted using statistical methods, and a structured feature set is output. Step S22: Based on the structured feature set, the sensor error is corrected by spatial correlation and fusion of GPS and IMU data, and the consistency coefficient is calculated by comparing real-time features with dynamically filtered historical features through temporal correlation. Finally, the feature set after correlation optimization is output.

3. The flight confidence calculation method based on multi-dimensional sensor data according to claim 1, characterized in that: In step S4, the multi-dimensional confidence fusion calculation includes the following steps: Step S41: Multi-dimensional confidence decomposition calculation, calculate data feature confidence, security value confidence and historical matching confidence; Step S42: Confidence fusion calculation. A dynamic weight fusion and entropy weight correction mechanism is used to fuse the confidence of the three dimensions to generate the final flight confidence.

4. The flight confidence calculation method based on multi-dimensional sensor data according to claim 1, characterized in that: In step S1, the data acquisition further includes the following steps: Step S11: Collect multi-source data, including the following steps: Step S111: Perform scene recognition and risk assessment. The edge device judges the current scene in real time through flight parameters and assigns the corresponding risk level coefficient. Simultaneously, it calls the security correlation of each data stream in the current scene from the cloud. Step S112: Dynamic resource allocation, calculate the resource allocation weight of each data stream; Step S113: Synchronous acquisition of multi-source data. Multi-source data is synchronously acquired through sensors, databases, and external interfaces according to resource allocation weights. The multi-source data includes real-time status data, environmental interference data, historical benchmark data, and task target data. Step S114: Data quality gating, which involves assessing the quality of the collected multi-source data and selecting qualified data.

5. The flight confidence calculation method based on multi-dimensional sensor data according to claim 1, characterized in that: In step S5, the decision feedback specifically involves making a decision based on the final flight confidence level; and inputting the actual flight safety status after the decision is executed back into the model adjustment parameters for feedback optimization.

6. A flight confidence calculation system based on multi-dimensional sensor data, used to implement the flight confidence calculation method based on multi-dimensional sensor data as described in any one of claims 1-5, characterized in that: It includes a data acquisition module, a multi-dimensional feature extraction module, a security value mapping module, a multi-dimensional confidence fusion calculation module, and a decision feedback module.

7. The flight confidence calculation system based on multi-dimensional sensor data according to claim 6, characterized in that: The data acquisition module specifically collects multi-source data synchronously according to resource allocation weights. After quality gating to screen qualified data, it uses Kalman filtering with dynamic covariance adjustment for noise filtering and optimizes the filtering effect by calculating the filtering residual. Based on two-dimensional similarity matching, dynamic window adjustment, time decay weighting, and scene correlation verification, it selects high-value historical features from the historical database, and finally obtains the filtered standard data and dynamically selected historical features. The data is then sent to the multi-dimensional feature extraction module. The multi-dimensional feature extraction module specifically extracts features from the filtered standard data to obtain a structured feature set. Based on the structured feature set, it calculates the consistency coefficient by spatially fusing GPS and IMU data and temporally comparing real-time features with dynamically filtered historical features. Finally, it outputs the feature set after association optimization. Weighted feature fusion is used to obtain the fused global feature vector. Finally, through a dual mechanism of multi-sensor cross-calibration and Bayesian inference verification, it outputs the optimal spatial parameter estimation set and the corresponding confidence interval. The data is then sent to the security value mapping module. The security value mapping module specifically defines security value objectives, constructs causal chains and quantifies the causal contribution of features to security value, and then classifies features into levels to implement precise risk control; the data is then sent to the multi-dimensional confidence fusion calculation module. The multi-dimensional confidence fusion calculation module specifically calculates the confidence of data features, the confidence of safety value, and the confidence of historical matching, respectively, and uses a dynamic weight fusion and entropy weight correction mechanism to perform confidence fusion to generate the final flight confidence; the data is then sent to the decision feedback module. The decision feedback module specifically makes decisions based on the final flight confidence level and inputs the actual safety status after the decision is executed back into the model for adjustment and optimization.

Images