Flight path dynamic closed-loop control and anomaly diagnosis method

Abstract

The invention provides a flight path dynamic closed-loop control and anomaly diagnosis method. The method comprises the following steps: planning a plurality of candidate flight paths; dividing each candidate path into a plurality of segments according to a first preset rule, performing segmented oil consumption prediction on each segment through a trained oil consumption prediction model to obtain a plurality of segmented predicted oil consumption, and performing accumulation summation on all the segmented predicted oil consumption to obtain overall predicted oil consumption; selecting a candidate path with the least overall predicted fuel consumption from the plurality of candidate paths as an optimal path; when the flight flies by adopting the optimal path, actually measuring the actual fuel consumption of each subsection in the optimal path, and performing difference calculation on the actual fuel consumption and the subsection predicted fuel consumption corresponding to the subsection; when the difference value exceeds a set threshold value, it is judged that oil consumption is abnormal. Risks and energy consumption can be reduced.

Description

Technical Field

[0001] The present invention relates to the field of aviation, and particularly to a method for dynamic closed-loop control and abnormal diagnosis of flight paths. Background Art

[0002] In the field of aviation, traditional methods for dynamic closed-loop control and abnormal diagnosis of flight paths usually rely on overall fuel consumption estimation models for path optimization, but such methods have significant defects. On the one hand, the cumulative error in long-distance flights will cause the fuel consumption prediction to deviate from the actual value, affecting the reliability of decision-making; on the other hand, the abnormal fuel consumption detection mechanism is mostly a global judgment, making it difficult to accurately locate the root cause of the problem. In the prior art, the segmentation rules for model training and path prediction are inconsistent, which is likely to introduce systematic biases, and the segmented features are not comparable with real-time data, restricting the timeliness of abnormal handling. At the same time, fuel consumption anomalies caused by environmental mutations or equipment failures often require complex manual diagnosis, making it difficult to achieve dynamic closed-loop optimization. These limitations make it difficult for existing methods to balance economic and safety requirements when dealing with complex meteorological conditions and real-time path adjustments. Summary of the Invention

[0003] To solve the technical problems in the background art, the present invention proposes a method for dynamic closed-loop control and abnormal diagnosis of flight paths, including: S1. Plan multiple candidate flight paths; S2. Divide each candidate path into multiple segments according to a first preset rule, obtain multiple segmented predicted fuel consumptions by performing segmented fuel consumption prediction on each segment through a trained fuel consumption prediction model, and sum up all the segmented predicted fuel consumptions to obtain the overall predicted fuel consumption; S3. Select the candidate path with the least overall predicted fuel consumption from the multiple candidate paths as the optimal path; S4. When the flight uses the optimal path, measure the actual fuel consumption of each segment in the optimal path and calculate the difference from the corresponding segmented predicted fuel consumption; S5. When the difference exceeds the set threshold, it is determined that the fuel consumption is abnormal; S6. Determine whether the abnormal fuel consumption is caused by environmental factors. If so, generate a route correction instruction. If not, proceed to S7; S7. Determine whether the abnormal fuel consumption is caused by equipment failure. If so, activate the redundant system and generate a recommended sequence of alternate airports. If not, proceed to S8; S8. Continuously and dynamically compare the flight data with the confidence interval of the model prediction. When a continuous abnormal deviation is detected and the output stability of the fuel consumption prediction model is abnormal, trigger the online adaptive optimization mechanism.

[0004] Furthermore, the training process of the "fuel consumption prediction model" specifically includes: S21. Split the historical flight path and extract features to form a training sample set; S22. Set the training objective of the fuel consumption prediction model to minimize the mean square error loss function between the predicted value and the actual fuel consumption to obtain a configured fuel consumption prediction model framework; S23. Optimize the model parameters through cross-validation; S24. Train the fuel consumption prediction model using a machine learning algorithm to obtain a trained fuel consumption prediction model that can output segmented fuel consumption prediction values.

[0005] Further, the step of "S21. Split the historical flight path and extract features to form a training sample set" specifically includes: S211. Split each historical flight path into multiple segments according to a first preset rule; S212. Extract a set of feature parameters for each segment as input; S213. Combine the extracted set of feature parameters with the actual fuel consumption data to form a training sample set.

[0006] Further, the set of feature parameters includes: the longitude and latitude of the starting and ending points of the segment, cruising altitude, airspeed, wind direction, wind speed, atmospheric temperature, and aircraft model parameters.

[0007] Further, the step of "S23. Optimize the model parameters through cross-validation" specifically includes: S231. Divide the historical flight path data set into multiple mutually exclusive subsets according to the time series; S232. Perform multiple rounds of parameter optimization training. In each round, retain one subset as the validation set, and use the remaining subsets as the training set to input into the fuel consumption prediction model; S233. Calculate the segmented fuel consumption prediction mean square error and the overall cumulative error of the validation set for the model after each round of training; S234. Based on the error results of multiple rounds, select a combination of model parameters that significantly reduces the segmented prediction error and the standard deviation of the overall cumulative error is less than a preset precision threshold.

[0008] Further, the step of "Determine whether the abnormal fuel consumption is caused by environmental factors" specifically includes: S61. Obtain real-time environmental parameters, where the real-time environmental parameters include: sudden change characteristics of the air flow direction, offset characteristics of the atmospheric pressure gradient, and distortion characteristics of the external temperature field; S62. Dynamically correlate and compare the real-time environmental parameters with the reference parameters in the historical normal flight environment database, and calculate the parameter offset; S63. According to the comparison result between the parameter offset and the system tolerance threshold range, if the parameter offset exceeds the system tolerance threshold range, the determination result is yes; if the parameter offset is within the system tolerance threshold range, the determination result is no.

[0009] Further, the "determining whether the abnormal fuel consumption is caused by equipment failure" specifically includes: S71. Extract the real-time vibration spectrum of the engine and calculate the main frequency band energy attenuation characteristics; S72. Synchronously monitor the discrete characteristics of the fuel flow sensor and the entropy value characteristics of the fuel supply pipeline pressure fluctuation; S73. Based on the main frequency band energy attenuation characteristics, the discrete characteristics of the fuel flow sensor, and the entropy value characteristics of the fuel supply pipeline pressure fluctuation, if the main frequency band energy attenuation characteristics exceed the normal baseline range and the entropy value characteristics of the fuel supply pipeline pressure fluctuation exceed the reference level range, the determination result is yes, otherwise the determination result is no.

[0010] Further, it also includes: S9. Feedback the deviation data in a closed loop to the ground system, and use the compensation coefficient dynamic adjustment algorithm to iteratively improve the prediction accuracy of the fuel consumption prediction model.

[0011] The present invention proposes a method for dynamic closed-loop control and abnormal diagnosis of flight paths. By forcing the segmentation rules in the unified planning stage and the flight monitoring stage, the segmentation prediction results of a single fuel consumption prediction model can be directly used as the benchmark for real-time comparison. This can not only eliminate the feature space misalignment problem caused by segmentation logic differences in traditional multi-model solutions, but also significantly reduce the real-time computing load of the airborne system; the step-by-step difference analysis of the measured fuel consumption and the pre-stored prediction data can achieve the rapid positioning and tracing of abnormal events without additional model operations, while ensuring the consistency of the error evaluation benchmark throughout the entire process from path optimization to flight monitoring, laying a high-precision and low-latency data foundation for subsequent environmental interference identification and equipment failure diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a flow chart of a method for dynamic closed-loop control and abnormal diagnosis of flight paths according to the present invention; Figure 2 It is a flow chart of a method for dynamic closed-loop control and abnormal diagnosis of flight paths according to the present invention; Figure 3 It is a flow chart of a method for dynamic closed-loop control and abnormal diagnosis of flight paths according to the present invention; Figure 4 It is a flow chart of a method for dynamic closed-loop control and abnormal diagnosis of flight paths according to the present invention; Figure 5Schematic flowchart of a method for dynamic closed-loop control and abnormal diagnosis of flight paths according to the present invention; Figure 6 Schematic flowchart of a method for dynamic closed-loop control and abnormal diagnosis of flight paths according to the present invention. Detailed implementation manner

[0013] Refer to Figures 1-6 , the present invention proposes a method for dynamic closed-loop control and abnormal diagnosis of flight paths, including: S1. Plan multiple candidate flight paths.

[0014] S2. Divide each candidate path into multiple segments according to a first preset rule, perform segmented fuel consumption prediction on each segment through a trained fuel consumption prediction model to obtain multiple segmented predicted fuel consumptions, and sum up all the segmented predicted fuel consumptions to obtain the overall predicted fuel consumption.

[0015] In terms of technical details, each candidate path is divided into multiple segments according to a fixed distance or time interval, and fuel consumption prediction is performed on each segment through a pre-trained fuel consumption prediction model. After the model outputs the segmented predicted fuel consumption, using the accumulation formula, the overall predicted fuel consumption is equal to the sum of the segmented predicted fuel consumptions. Process the data segment by segment and aggregate the results to provide a quantitative basis for path optimization. In terms of technical effects, the segmented prediction mechanism significantly reduces the cumulative error, enhances the accuracy of fuel consumption estimation, and thus supports more reliable path decisions.

[0016] When dividing each candidate path into multiple segments according to the first preset rule, segmenting based on the first preset rule has two core constraints of two-dimensional locking; First, absolute time window cutting, all segments are strictly cut according to a fixed duration (for example, the fixed duration is defaulted to 90 seconds) to ensure that fuel consumption comparison can be triggered at the end of each segment during flight; and it is exactly the same as the segment time window in the planning stage; Second, waypoint subordination principle, if a physical waypoint falls within the current time window, set it as the segment end point (for example, if the originally planned flyover point A arrives at 60 seconds, then end the current segment in advance); if no waypoint is reached at the end of the time window, generate a virtual segment point (the coordinates are calculated by the FMS in real time). During the process of segmenting based on the first preset rule, the time window is used as the primary key throughout the process, and the virtual point mechanism maintains the path continuity, ensuring that the number of planned segments is equal to the number of flight monitoring segments, and making the longitude and latitude of the start and end points of each segment exactly match.

[0017] Among them, the training process of the "fuel consumption prediction model" specifically includes: S21. Divide historical flight paths and extract features to form a training sample set. Specifically including: S211. Divide each historical flight path into multiple segments according to the first preset rule.

[0018] In terms of technical details, each historical flight path is segmented into multiple segments according to the first preset rule, and the rule is based on the segment length or the change of key waypoints to ensure segment consistency. This step, as a pre-operation for training sample preparation, transforms the original path data into structured segment units, facilitating feature extraction. Standardized segmentation can strengthen the comparability of data and the generalization ability of the trained fuel consumption prediction model, laying a foundation for the subsequent steps.

[0019] S212. Extract a set of feature parameters for each segment as input.

[0020] In terms of technical details, for each segment, a set of feature parameters is extracted. The set of feature parameters includes the latitude and longitude of the starting and ending points of the segment, cruise altitude, airspeed, wind direction, wind speed, atmospheric temperature, and aircraft model parameters. These parameters are processed through a normalization formula to eliminate the influence of dimension. This step transforms the segmented segments into model input feature vectors and directly feeds them into the training process. Multi-dimensional features can capture the key physical characteristics of fuel consumption driving factors, improving the robustness and interpretability of model prediction.

[0021] S213. Combine the extracted set of feature parameters with the actual fuel consumption data to form a training sample set.

[0022] In terms of technical details, the extracted set of feature parameters is paired with the actual fuel consumption data to form a training sample set. Each sample is represented as a mapping relationship between the input feature vector and the actual fuel consumption scalar. This step integrates the outputs of S211 and S212 to form a training sample set, i.e., a supervised learning data set, for the training phase of the fuel consumption prediction model. Constructing a high-quality training sample set can ensure the effectiveness of model learning, reduce the risk of overfitting, and improve the prediction generalization performance.

[0023] S22. Set the training objective of the fuel consumption prediction model to minimize the mean squared error loss function between the predicted value and the actual fuel consumption, so as to obtain a fuel consumption prediction model framework with the training objective configured.

[0024] In terms of technical details, the training objective of the fuel consumption prediction model is defined as minimizing the mean squared error loss function between the predicted value and the actual fuel consumption. The specific formula is that the mean squared error loss is equal to the mean of the square of the predicted fuel consumption minus the actual fuel consumption. This step is used to configure the optimization objective of the fuel consumption prediction model and guide the gradient descent algorithm to update the weight parameters. Minimizing the mean squared error directly optimizes the prediction accuracy, making the model output closer to the true fuel consumption distribution and enhancing the overall reliability.

[0025] S23. Optimize the model parameters through cross-validation. Specifically, it includes: S231. Divide the historical flight path data set into multiple mutually exclusive subsets according to the time series.

[0026] Sorted by the flight date, the historical data set is cut into a mutually exclusive subset (such as = 5); Example: Divide the data from January to December 2024 into 12 subsets monthly, each subset containing all flight paths for a single month; It is mandatory that there is no time overlap between subsets (e.g., subset 1 is the data for January; subset 2 is the data for February), ensuring that there is no time intersection between the training set and the validation set.

[0027] S232. Perform multiple rounds of parameter optimization training. In each round, retain one subset as the validation set, and use the remaining subsets as the training set to input into the fuel consumption prediction model.

[0028] Conduct rounds of training ( equal to the number of subsets). In each round of iteration, combine subsets (such as subset 1 + 2 + 3 + 4) as the training set, retain 1 subset (such as subset 5) as the validation set, and use the gradient descent algorithm (such as the Adam optimizer) to update the model weights with the goal of minimizing the mean squared error of the segmented fuel consumption prediction for the training set. At the same time, each subset participates in the role of the validation set once, making full use of the limited data to evaluate the model performance. The comprehensive results of multiple rounds can offset the contingency of a single division (such as the evaluation distortion caused by a subset containing flights in abnormal weather), expose the generalization ability defects of the model in different time periods (such as the prediction deviation caused by seasonal changes), and drive the selection of a more adaptable parameter combination.

[0029] S233. Calculate the mean squared error of the segmented fuel consumption prediction and the overall cumulative error of the validation set for the model after each round of training.

[0030] Specifically, the mean squared error of the segmented fuel consumption prediction , where, ( is the number of segments, , are respectively the predicted fuel consumption and the actual fuel consumption of segment ); The overall cumulative error , where, is the total predicted fuel consumption, is the actual total fuel consumption. The mean squared error of the segmented fuel consumption prediction reflects the accuracy of the model's fuel consumption prediction for local flight states (such as specific altitude / wind speed); The overall cumulative error can reveal the cumulative effect of the segmented error, avoid the global deviation being masked by local high precision, identify whether the model sacrifices the segmented precision for the overall balance (such as overfitting in some segments and underfitting in others), guide the direction of parameter adjustment, and provide a quantitative basis for decision-making. The error value is directly related to the subsequent parameter selection threshold (such as requiring that the overall cumulative error must be less than a preset value).

[0031] S234. Based on the error results of multiple rounds, select a model parameter combination that significantly reduces the segmented prediction error and the standard deviation of the overall cumulative error is less than the preset precision threshold.

[0032] The selection criteria include segmentation is significantly reduced and the standard deviation of the overall cumulative error ( ) meets the standard. For example, for segmentation it decreases by ≥ 15% compared to the baseline model (such as an un-tuned model), and meets the standard, that is , where is a preset threshold, and the preset threshold can limit the cumulative error within an acceptable range (such as <1.5% of the total fuel consumption), ensuring the reliability of path decision-making.

[0033] The selection process is to traverse the training results of multiple rounds, and screen the parameter combinations that meet both conditions; if multiple combinations meet the standard, select the combination with the smallest standard deviation of the overall cumulative error.

[0034] This solution forces the model to balance the learning of segmentation details and the consistency of long paths, avoiding accurate segmentation but overall inaccuracy. A low standard deviation indicates that the model has small fluctuations in performance at different time periods and can adapt to complex environmental changes.

[0035] S24. Use a machine learning algorithm to train the fuel consumption prediction model to obtain a trained fuel consumption prediction model that can output segmented fuel consumption prediction values.

[0036] In terms of technical details, use a machine learning algorithm such as gradient boosting decision tree to train the fuel consumption prediction model. The algorithm optimizes the weights to minimize a preset loss function and outputs a model that can predict segmented fuel consumption. As the final stage of the training process, S24 generates a deployable prediction model. The trained model can achieve high-precision segmented fuel consumption output and provide a core engine for real-time path optimization.

[0037] S3. Select the candidate path with the least overall predicted fuel consumption from multiple candidate paths as the optimal path.

[0038] In terms of technical details, select the candidate path with the least overall predicted fuel consumption from multiple candidate paths as the optimal path, and select the minimum value path through comparing the cumulative sum formula. This step performs path decision-making based on the output of S2 and directly outputs the optimal flight plan. This selection mechanism minimizes the expected fuel consumption and improves the economy and environmental friendliness of the flight.

[0039] S4. When the flight flies along the optimal path, measure the actual fuel consumption of each segment in the optimal path and calculate the difference from the corresponding segmented predicted fuel consumption.

[0040] By enforcing the segmentation rules for the unified planning phase and the flight monitoring phase, the segmentation prediction results of a single fuel consumption prediction model can be directly used as the benchmark for real-time comparison. This not only eliminates the problem of feature space misalignment caused by segmentation logic differences in traditional multi-model solutions but also significantly reduces the real-time computing load of the airborne system. The step-by-step difference analysis between the measured fuel consumption and the pre-stored prediction data can achieve rapid positioning and tracing of abnormal events without additional model operations. At the same time, it ensures the consistency of the error evaluation benchmark throughout the entire process from route optimization to flight monitoring, laying a high-precision and low-latency data foundation for subsequent environmental interference identification and equipment fault diagnosis.

[0041] S5. When the difference exceeds the set threshold, it is determined that the fuel consumption is abnormal.

[0042] In terms of technical details, when the difference exceeds the set threshold, it is determined that the fuel consumption is abnormal, and the threshold is set based on the historical error distribution. This step serves as the abnormal trigger point. If the difference exceeds the limit, it will enter the subsequent diagnostic process. Threshold judgment can simplify abnormal classification and accelerate the response speed to faults or environmental factors.

[0043] S6. Determine whether the abnormal fuel consumption is caused by environmental factors. If so, generate a route correction instruction; if not, go to S7.

[0044] In terms of technical details, determine whether the abnormal fuel consumption is caused by environmental factors through obtaining and analyzing real-time environmental parameters for decision-making. This step is executed after S5 and serves as the first step in the root cause analysis of abnormalities. Determining environmental factors helps isolate external interferences and can guide subsequent correction operations. The route correction instruction includes instructions to adjust the flight altitude or path.

[0045] The "determine whether the abnormal fuel consumption is caused by environmental factors" specifically includes: S61. Obtain real-time environmental parameters.

[0046] In terms of technical details, obtain real-time environmental parameters, including sudden change characteristics of air flow direction, offset characteristics of atmospheric pressure gradient, and distortion characteristics of external temperature field, and collect data in real time through a sensor network. This step provides the input source for environmental determination and directly feeds it into the comparison and analysis. Multi-parameter collection can enhance the comprehensiveness of environmental diagnosis and cover key meteorological influencing factors.

[0047] S62. Dynamically correlate and compare the real-time environmental parameters with the reference parameters in the historical normal flight environment database, and calculate the parameter offset.

[0048] In terms of technical details, the real-time environmental parameters are dynamically correlated and compared with the reference parameters in the historical normal flight environment database to calculate the parameter offset. The formula is that the offset is equal to the absolute value of the real-time parameter minus the reference parameter. This step performs parameter correlation analysis to quantify the degree of environmental change. The dynamic comparison can reveal the patterns of environmental anomalies and support the objective evaluation of the offset.

[0049] S63. According to the comparison result between the parameter offset and the system tolerance threshold range, if the parameter offset exceeds the system tolerance threshold range, the determination result is yes; if the parameter offset is within the system tolerance threshold range, the determination result is no.

[0050] In terms of technical details, according to the comparison result between the parameter offset and the system tolerance threshold range, if the offset exceeds the threshold, the determination result is yes, otherwise it is no. This step completes the decision-making on environmental factors and outputs a boolean result. The threshold comparison can achieve automatic judgment, reduce subjective errors and improve response efficiency.

[0051] S7. Determine whether the abnormal fuel consumption is caused by equipment failure. If so, activate the redundant system and generate a recommended sequence of alternate airports. If not, go to S8.

[0052] In terms of technical details, determine whether the abnormal fuel consumption is caused by equipment failure, which is achieved by monitoring the characteristics of the engine and fuel system. This step is executed after the environmental determination is no and serves as a continuation of the abnormal analysis. The equipment failure determination locates internal problems and ensures the timely activation of safety measures.

[0053] The "determine whether the abnormal fuel consumption is caused by equipment failure" specifically includes: S71. Extract the real-time vibration spectrum of the engine and calculate the energy attenuation characteristics of the main frequency band.

[0054] In terms of technical details, extract the real-time vibration spectrum of the engine and calculate the energy attenuation characteristics of the main frequency band, and analyze the spectral energy change through Fourier transform. This step provides vibration data input for equipment diagnosis. The energy attenuation characteristics detect engine mechanical failures and can improve the sensitivity of fault identification.

[0055] S72. Synchronously monitor the discrete characteristics of the fuel flow sensor and the entropy value characteristics of the pressure fluctuation in the fuel supply pipeline.

[0056] In terms of technical details, synchronously monitor the discrete characteristics of the fuel flow sensor and the entropy value characteristics of the pressure fluctuation in the fuel supply pipeline. The discrete characteristics reflect data jumps, and the entropy value characteristics quantify the degree of pressure disorder. This step collects fuel system data in parallel and collaborates with vibration analysis. The multi-characteristic synchronous monitoring covers the entire fuel supply chain and can reduce the diagnostic blind area.

[0057] S73. Based on the main frequency band energy attenuation feature, the discrete feature of the fuel flow sensor, and the entropy value feature of the fuel supply pipeline pressure fluctuation, if the main frequency band energy attenuation feature exceeds the normal baseline range and the entropy value feature of the fuel supply pipeline pressure fluctuation exceeds the reference level range, the determination result is yes; otherwise, the evaluation result is no.

[0058] The dual - condition check can strengthen the reliability of the criterion and ensure a high confidence level in fault confirmation.

[0059] S8. Continuously and dynamically compare the flight data with the model prediction confidence interval. When a continuous abnormal deviation is detected and the model output stability is abnormal, trigger the online adaptive optimization mechanism.

[0060] In terms of technical details, continuously and dynamically compare the flight data with the model prediction confidence interval. When a continuous abnormal deviation is detected and the model output stability is abnormal, trigger the online adaptive optimization mechanism to adapt to the new data by adjusting the model weights. This step is executed when the anomaly is not resolved by S6 or S7, realizing the self - update of the model. The adaptive optimization can cope with data drift and maintain the timeliness and accuracy of the prediction model.

[0061] S9. Feed the deviation data back to the ground system in a closed - loop manner and use the compensation coefficient dynamic adjustment algorithm to iteratively improve the prediction accuracy of the fuel consumption prediction model.

[0062] In terms of technical details, feed the deviation data back to the ground system in a closed - loop manner and use the compensation coefficient dynamic adjustment algorithm to update the fuel consumption prediction model. The compensation coefficient formula is calculated based on the historical deviation mean. This step is the closed - loop end point of the process, transmitting the flight data back to the training system. The feedback mechanism can iteratively improve the prediction accuracy of the model and form a continuously optimized learning loop.

[0063] The present invention proposes a method for dynamic closed-loop control and anomaly diagnosis of flight paths. By using the same fuel consumption prediction model as both the data source for overall fuel consumption prediction and the real-time monitoring benchmark, the collaborative efficiency of the entire process is significantly improved. The model outputs the predicted fuel consumption for each segment of the candidate path during the path planning phase, and the sum of these predictions is directly used for the optimal path decision-making. When the flight follows the optimal path, the predicted values for each segment of the model also become the comparison benchmarks for real-time fuel consumption monitoring. The same preset rules are strictly used for both segments to ensure the complete alignment of the data structures, enabling the feature space learned during the training phase to precisely match the segment units during the prediction and monitoring phases, and eliminating error transmission caused by fuzzy boundaries. The fuel consumption difference for each segment during actual flight can be directly correlated with the original input features of the model to achieve millisecond-level localization of the root cause of anomalies. The calculation of environmental parameter offsets and equipment fault diagnosis are carried out based on the predicted values for each segment output by the model, avoiding benchmark drift caused by multi-system switching. The continuously detected deviation data dynamically optimize the model weights through a closed-loop feedback mechanism, resulting in a simultaneous enhancement of prediction accuracy and diagnostic sensitivity. This single-model dual-application architecture reduces system complexity and constructs a seamless closed-loop from path planning to anomaly handling, providing dual support for flight safety and economy while ensuring the reliability of fuel consumption prediction.

[0064] The present invention drives carbon emissions reduction through two engines of precise prediction and dynamic optimization by proposing a method for dynamic closed-loop control and anomaly diagnosis of flight paths. During the path planning phase, the global optimal path is selected based on the predicted fuel consumption for each segment, directly avoiding high-energy-consuming routes and compressing the total fuel consumption from the source. During actual flight, the real-time fuel consumption difference monitoring and anomaly diagnosis mechanism can instantaneously trigger route correction instructions, such as dynamically adjusting the flight altitude to avoid strong countercurrent areas or switching to an alternative path to bypass sudden storm areas, to avoid ineffective fuel losses. For fuel consumption anomalies triggered by equipment failures, the system actively activates redundant equipment and guides emergency landings to prevent continuous deterioration of fuel efficiency caused by the failures. The closed-loop feedback mechanism continuously feeds actual flight data back to the ground system, enabling the fuel consumption prediction model to continuously approach the real physical environment and ensuring that subsequent flights continue to adopt more fuel-efficient paths. This full-link optimization from planning, execution to iteration enables the fuel efficiency of each flight to reach the theoretical optimal range, thereby directly reducing carbon dioxide emissions and contributing to the green transformation of the aviation industry.

[0065] The present invention constructs an accuracy evolution closed loop through a model training mechanism in a flight path dynamic closed loop control and anomaly diagnosis method. The cross-validation process adopts a time series partitioning strategy to strictly isolate the risk of time overlap between the training set and the validation set, ensuring that the optimization of model parameters is not contaminated by future data. During multiple rounds of training, dual evaluation metrics of segment prediction error and overall cumulative error are calculated synchronously, forcing the model to balance local feature learning and global consistency, and avoiding the superposition of prediction biases caused by segment optimization. The real-time data in the flight phase is dynamically compared with the model confidence interval, which can sensitively capture signs of environmental drift or equipment degradation, and can promptly trigger an online adaptive optimization mechanism, such as reweighting feature dimensions or adjusting the depth of the tree model, to make the predicted value fit the changes in the physical world in real time. After the deviation data is fed back to the ground system through the closed loop channel, the compensation coefficient dynamic adjustment algorithm converts the historical error mean into a model weight correction amount, driving the parameter update in the next training cycle. This spiral upward mechanism from offline training to online tuning and then to offline iteration enables the model to continuously absorb new type of data experience, and the prediction accuracy grows logarithmically with the operation time, ultimately forming a highly robust prediction ability against environmental disturbances and equipment aging.

[0066] The above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A method for dynamic closed-loop control and abnormal diagnosis of flight paths, characterized in that Including: S1. Plan multiple candidate flight paths; S2. Divide each candidate path into multiple segments according to the first preset rule, perform segment fuel consumption prediction on each segment through a trained fuel consumption prediction model to obtain multiple segment prediction fuel consumptions, and accumulate and sum all segment prediction fuel consumptions to obtain the overall predicted fuel consumption; S3. Select the candidate path with the least overall predicted fuel consumption from multiple candidate paths as the optimal path; S4. When the flight uses the optimal path for flight, measure the actual fuel consumption of each segment in the optimal path and calculate the difference from the corresponding segment prediction fuel consumption; S5. When the difference exceeds the set threshold, it is judged as abnormal fuel consumption; S6. Determine whether the abnormal fuel consumption is caused by environmental factors. If so, generate a route correction instruction. If not, go to S7; S7. Determine whether the abnormal fuel consumption is caused by equipment failure. If so, activate the redundant system and generate a recommended sequence of alternate airports. If not, go to S8; S8. Dynamically compare the flight data with the model prediction confidence interval in real time. When continuous abnormal deviation is detected and the output stability of the fuel consumption prediction model is abnormal, trigger the online adaptive optimization mechanism.

2. The method according to claim 1, characterized in that, The training process of the "fuel consumption prediction model" specifically includes: S21. Divide the historical flight path and extract features to form a training sample set; S22. Set the training target of the fuel consumption prediction model to minimize the mean square error loss function between the predicted value and the actual fuel consumption to obtain a fuel consumption prediction model framework with the training target configured; S23. Optimize the model parameters through cross-validation; S24. Train the fuel consumption prediction model using a machine learning algorithm to obtain a trained fuel consumption prediction model that can output segment fuel consumption prediction values.

3. The method according to claim 2, wherein The "S21. Divide the historical flight path and extract features to form a training sample set" specifically includes: S211. Divide each historical flight path into multiple segments according to the first preset rule; S212. Extract a set of feature parameters for each segment as input; S213. Combine the extracted set of feature parameters with the actual fuel consumption data to form a training sample set.

4. The method according to claim 3, characterized in that, The set of feature parameters includes: the longitude and latitude of the segment start and end points, cruising altitude, airspeed, wind direction, wind speed, atmospheric temperature, and aircraft model parameters.

5. The method according to claim 2, characterized in that, The "S23. Optimize the model parameters through cross-validation" specifically includes: S231. Divide the historical flight path data set into multiple mutually exclusive subsets according to the time series; S232. Perform multiple rounds of parameter optimization training. In each round, retain one subset as the validation set, and the remaining subsets as the training set and input them into the fuel consumption prediction model; S233. Calculate the segment fuel consumption prediction mean square error and the overall cumulative error of the validation set for the model after each round of training; S234. Based on the error results of multiple rounds, select a model parameter combination that significantly reduces the segment prediction error and the standard deviation of the overall cumulative error is less than the preset precision threshold.

6. The method according to claim 1, wherein The "Determine whether the abnormal fuel consumption is caused by environmental factors" specifically includes: S61. Obtain real-time environmental parameters, where the real-time environmental parameters include: characteristics of sudden change in air flow direction, characteristics of atmospheric pressure gradient offset, and characteristics of external temperature field distortion; S62. Dynamically correlate and compare the real-time environmental parameters with the reference parameters in the historical normal flight environment database, and calculate the parameter offset; S63. According to the comparison result between the parameter offset and the system tolerance threshold range, if the parameter offset exceeds the system tolerance threshold range, the determination result is yes; if the parameter offset is within the system tolerance threshold range, the determination result is no.

7. The method according to claim 1, characterized in that The specific content of "determining whether the abnormal fuel consumption is caused by equipment failure" includes: S71. Extract the real-time vibration spectrum of the engine and calculate the energy attenuation characteristics of the main frequency band; S72. Synchronously monitor the discrete characteristics of the fuel flow sensor and the entropy value characteristics of the pressure fluctuation in the fuel supply pipeline; S73. Based on the energy attenuation characteristics of the main frequency band, the discrete characteristics of the fuel flow sensor, and the entropy value characteristics of the pressure fluctuation in the fuel supply pipeline, if the energy attenuation characteristics of the main frequency band exceed the normal baseline range and the entropy value characteristics of the pressure fluctuation in the fuel supply pipeline exceed the reference level range, the determination result is yes, otherwise the determination result is no.

8. The method according to claim 1, wherein It further includes: S9. Closed-loop feedback the deviation data to the ground system, and use the compensation coefficient dynamic adjustment algorithm to iteratively improve the prediction accuracy of the fuel consumption prediction model.

Images