Multi-Sensor Altitude Prediction System Integrating LSTM and Kalman Filter

Abstract

The present invention discloses a multi-sensor altitude prediction system integrating LSTM and Kalman filter, which relates to the technical field of multi-sensor altitude prediction. It includes an aircraft altitude monitoring and data acquisition module, a phase mismatch prediction and classification module, an adaptive weight adjustment module, an adjustment mechanism feedback evaluation module, and a flight process monitoring and optimization module. The phase mismatch prediction and classification module analyzes the obtained prediction input information, predicts the degree of phase mismatch between the prediction results of the LSTM model and the Kalman filter, and classifies it into three categories: phase consistency, mild mismatch, and severe mismatch according to the prediction results. The adaptive weight adjustment module constructs an adaptive weight adjustment mechanism according to the classification results and takes different weight adjustment measures for the three situations of phase consistency, mild mismatch, and severe mismatch respectively. The present invention solves the problem of phase mismatch in altitude prediction under sudden environmental changes of the aircraft and realizes the real-time response and long-term stability of altitude prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-sensor altitude prediction, and particularly to a multi-sensor altitude prediction system integrating LSTM and Kalman filter. Background Art

[0002] In recent years, with the rapid development of aerospace technology, the safe flight of aircraft has put forward higher requirements for the accurate acquisition of flight altitude data. To ensure the safe and efficient operation of aircraft in complex airspace environments, the commonly used altitude measurement methods currently mainly include the Global Navigation Satellite System (GNSS), Inertial Navigation System (INS), and barometric altimeter. GNSS can provide high-precision three-dimensional positioning and altitude information. The inertial navigation system relies on gyroscopes and accelerometers to achieve autonomous navigation, and the barometric altimeter calculates altitude by measuring atmospheric pressure. Each type of sensor has its own advantages in different flight scenarios. Through data fusion technology, integrating multi-source sensor data to improve the accuracy and robustness of altitude measurement has become an important research direction in the aviation field. In recent years, deep learning algorithms, especially Long Short-Term Memory Neural Networks (LSTM), have shown excellent performance in processing time series data, while the Kalman filter, as a classic data fusion algorithm, can effectively filter out noise in dynamic systems. Therefore, combining the LSTM neural network with the Kalman filter to predict flight altitude through multi-sensor data fusion is becoming an important technical means to improve the reliability and stability of aviation flight data.

[0003] The existing multi-sensor altitude prediction technology integrating LSTM and Kalman filter realizes accurate prediction of the aircraft altitude by integrating data from the Global Navigation Satellite System (GNSS), Inertial Navigation System (INS), and barometric altimeter. First, the LSTM neural network is used to process the time series data of flight altitude. By learning historical data, a non-linear mapping relationship between input features and altitude output is established to predict the altitude data at the next moment. Then, the LSTM prediction result is combined with the real-time sensor data from GNSS, INS, and barometer and input into the Kalman filter for data fusion. During the fusion process, the Kalman filter takes the measurement values of each sensor as the observation input and iteratively updates based on the system state model. By calculating the Kalman gain, the weights of different data sources are dynamically adjusted to obtain a more stable and accurate altitude estimate. The whole process effectively utilizes the prediction ability of LSTM for time series and the advantage of Kalman filter in noise suppression to ensure the accuracy and reliability of altitude information of the aircraft in various complex environments.

[0004] The existing technology has the following deficiencies:

[0005] In the case where the aircraft encounters sudden airflows or drastic environmental changes, there will be a sharp fluctuation in altitude data. In the existing fusion LSTM and Kalman filter system, the LSTM model relies on long-term sequence data for trend prediction, while the Kalman filter relies on short-term data for error correction. Therefore, when the environment changes suddenly, the long-sequence prediction of the LSTM based on historical data will lag behind the actual situation, and the Kalman filter tries to correct the current altitude data through rapid adjustment, resulting in a phase inconsistency in the prediction results of the two. This phase mismatch causes the system to be unable to effectively balance the long-term trend prediction of the LSTM and the short-term correction of the Kalman filter in a short time, making the altitude prediction results oscillate frequently in the short term and difficult to stabilize. Since the existing technology fails to quickly adjust the weight distribution between the LSTM and the Kalman filter, it cannot effectively handle this sudden phase mismatch problem, resulting in a response delay in the altitude control system when the aircraft rapidly descends or lands, thereby increasing the risk of altitude control failure during landing, further causing the aircraft to land unstably, increasing energy consumption and affecting flight safety.

[0006] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure. Therefore, it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0007] The object of the present invention is to provide a multi-sensor altitude prediction system that fuses LSTM and Kalman filter to solve the problems in the above background art.

[0008] To achieve the above object, the present invention provides the following technical solution: A multi-sensor altitude prediction system that fuses LSTM and Kalman filter, including an aircraft altitude monitoring and data acquisition module, a phase mismatch prediction and classification module, an adaptive weight adjustment module, an adjustment mechanism feedback evaluation module, and a flight process monitoring and optimization module;

[0009] The aircraft altitude monitoring and data acquisition module, during the flight of the aircraft, continuously monitors the altitude data of the aircraft, and when it detects a sharp fluctuation in the altitude data of the aircraft, obtains the prediction input information required for generating the predictions of the LSTM model and the Kalman filter;

[0010] The phase mismatch prediction and classification module analyzes the obtained prediction input information, predicts the degree of phase mismatch between the prediction results of the LSTM model and the Kalman filter, and classifies it into three categories: phase consistency, mild mismatch, and severe mismatch according to the prediction results;

[0011] The adaptive weight adjustment module, according to the classification results, constructs an adaptive weight adjustment mechanism, and performs different weight adjustment measures for the three cases of phase consistency, mild mismatch, and severe mismatch respectively;

[0012] Adjustment mechanism feedback evaluation module. During the process of the adaptive weight adjustment mechanism adjusting the weights, it obtains the adjustment mechanism feedback information in real time, analyzes it after obtaining, evaluates whether the adjustment degree of the adaptive weight adjustment mechanism can meet the expectations, and optimizes the adaptive weight adjustment mechanism according to the evaluation results;

[0013] Flight process monitoring and optimization module. It monitors and analyzes the altitude prediction data and adjustment records during the entire flight process in real time, and continuously optimizes the adaptive weight adjustment mechanism according to the monitoring results to ensure the altitude prediction stability and control accuracy of the aircraft under different flight stages and environmental conditions.

[0014] Preferably, in the phase mismatch prediction and classification module, the obtained prediction input information is preprocessed, and the prediction time synchronization information and prediction error comparison information in the preprocessed prediction input information are extracted, and analyzed after extraction to generate a prediction time deviation index and an altitude prediction error synchronization index respectively; a phase mismatch evaluation model is constructed for the generated prediction time deviation index and altitude prediction error synchronization index to generate a phase mismatch coefficient, and the generated phase mismatch coefficient is compared with a preset phase mismatch coefficient threshold interval, and the phase mismatch degree between the prediction results of the LSTM model and the Kalman filter is predicted according to the comparison result, and it is classified into three categories: phase consistent, mild mismatch, and severe mismatch according to the prediction result.

[0015] Preferably, the acquisition logic of the prediction time deviation index and the altitude prediction error synchronization index is as follows:

[0016] Extract the prediction time synchronization information in the preprocessed prediction input information, specifically including the timestamps of the prediction results generated by the LSTM model at different moments within a period of time, the timestamps of the prediction results generated by the Kalman filter, and the timestamp of the actual acquired aircraft altitude data, and are respectively marked as 、 and , represents the timestamp of the prediction result generated by the LSTM model at the moment within a period of time, represents the timestamp of the prediction result generated by the Kalman filter at the moment within a period of time, represents the timestamp of the actual acquired aircraft altitude data at the moment within a period of time, , is a positive integer;

[0017] Calculate the prediction time deviation index, and the specific calculation formula is as follows:

[0018]

[0019] In the formula, is the prediction time deviation index;

[0020] Extract the prediction error comparison information from the pre - processed prediction input information, specifically including the height prediction values of the LSTM model, the height prediction values of the Kalman filter, and the actual height of the aircraft obtained at different times within a period of time, and label them respectively as , and , represents the height prediction value of the LSTM model at the moment within a period of time, represents the height prediction value of the Kalman filter at the moment within a period of time, represents the actual height of the aircraft obtained at the moment within a period of time;

[0021] Calculate the height prediction error synchronization index, and the specific calculation formula is as follows:

[0022]

[0023] In the formula, is the height prediction error synchronization index.

[0024] Preferably, construct a phase mismatch evaluation model for the generated prediction time deviation index and the height prediction error synchronization index , generate a phase mismatch coefficient through weighted summation, and compare the generated phase mismatch coefficient with the preset phase mismatch coefficient threshold interval to predict the phase mismatch degree between the prediction results of the LSTM model and the Kalman filter according to the comparison result, and classify it into three categories: phase - consistent, mild mismatch, and severe mismatch according to the prediction result. The specific comparison analysis and classification are as follows:

[0025] If , the phase mismatch degree between the prediction results of the LSTM model and the Kalman filter is at a low level, and this situation is classified as phase - consistent;

[0026] If , the phase mismatch degree between the prediction results of the LSTM model and the Kalman filter is at a medium level, and this situation is classified as mild mismatch;

[0027] If , if the degree of phase mismatch between the prediction results of the LSTM model and the Kalman filter is high, this situation is classified as a serious mismatch.

[0028] Preferably, in the adaptive weight adjustment module, according to the classification results, an adaptive weight adjustment mechanism is constructed, specifically: according to the classification results of phase consistency, mild mismatch, and serious mismatch, different weight adjustment parameters are set, corresponding to the weight allocation of the LSTM model and the Kalman filter in prediction fusion respectively, to form an adaptive weight adjustment mechanism; based on the range of the phase mismatch coefficient, this adjustment mechanism automatically determines the weight allocation strategy and adjustment amplitude between the LSTM model and the Kalman filter through a pre-set rule.

[0029] Different weight adjustment measures are taken for the three cases of phase consistency, mild mismatch, and serious mismatch respectively, specifically: in the case of phase consistency, an equal weight strategy is adopted, and equal weights are assigned to the LSTM model and the Kalman filter to maintain stable prediction; in the case of mild mismatch, the weight of the Kalman filter is adjusted to be higher than that of the LSTM model to increase the influence of short-term correction in prediction, so that while the system maintains the balance of long-term and short-term predictions, it enhances the response ability to current environmental changes; in the case of serious mismatch, the weight of the Kalman filter is significantly increased, and at the same time the weight of the LSTM model is significantly reduced, ensuring that the system can quickly respond to sudden changes and enhance the short-term stability of high-altitude prediction.

[0030] Preferably, in the adjustment mechanism feedback evaluation module, after obtaining the adjustment mechanism feedback information in real time, it is preprocessed, and the instantaneous prediction adjustment feedback information and long-term adjustment cumulative feedback information in the preprocessed adjustment mechanism feedback information are extracted, and after extraction, they are analyzed to generate an immediate adjustment deviation index and a cumulative adjustment effect coefficient respectively; an adjustment degree evaluation model is constructed for the generated immediate adjustment deviation index and cumulative adjustment effect coefficient to generate an adjustment coefficient, and the generated adjustment coefficient is compared with a pre-set adjustment coefficient threshold, and according to the comparison result, it is evaluated whether the adjustment degree of the adaptive weight adjustment mechanism can meet the expectation, and the adaptive weight adjustment mechanism is optimized according to the evaluation result.

[0031] Preferably, the acquisition logic of the immediate adjustment deviation index and the cumulative adjustment effect coefficient is as follows:

[0032] Extract the instantaneous prediction adjustment feedback information in the preprocessed adjustment mechanism feedback information, specifically including the difference between the predicted altitude of the LSTM model and the Kalman filter combination and the actual altitude of the aircraft at different times within a period during the process of the adaptive weight adjustment mechanism adjusting the weight, the weight difference between the LSTM model and the Kalman filter, and the time interval of the system's response to the altitude change of the aircraft after the weight adjustment, and they are respectively calibrated as , and , represents the difference between the predicted height of the combined LSTM model and Kalman filter and the actual height of the aircraft at a certain moment within a period of time during the process of weight adjustment by the adaptive weight adjustment mechanism, ; represents the weight difference between the LSTM model and the Kalman filter at a certain moment within a period of time during the process of weight adjustment by the adaptive weight adjustment mechanism, ; represents the time interval for the system to respond to the change in the aircraft height after weight adjustment, , where \(n\) is a positive integer;

[0033] Calculate the immediate adjustment deviation index, and the specific calculation formula is as follows:

[0034]

[0035] In the formula, is the immediate adjustment deviation index;

[0036] Extract the long-term adjustment cumulative feedback information from the preprocessed adjustment mechanism feedback information, specifically including the cumulative prediction error change amount after multiple weight adjustments within the time window of weight adjustment by the adaptive weight adjustment mechanism, the cumulative amplitude of all weight adjustments within this time window, and the time length within this time window, and label them as , and , represents the cumulative prediction error change amount after multiple weight adjustments within the time window of weight adjustment by the adaptive weight adjustment mechanism, represents the cumulative amplitude of all weight adjustments within this time window, represents the time length within this time window;

[0037] Calculate the cumulative adjustment effect coefficient, and the specific calculation formula is as follows:

[0038]

[0039] In the formula, is the cumulative adjustment effect coefficient.

[0040] Preferably, for the generated immediate adjustment deviation index and the cumulative adjustment effect coefficient construct an adjustment degree evaluation model, generate an adjustment coefficient by weighted summation, and compare the generated adjustment coefficient with a pre-set adjustment coefficient threshold Make a comparison, evaluate whether the adjustment degree of the adaptive weight adjustment mechanism can meet the expectation according to the comparison result, and optimize the adaptive weight adjustment mechanism according to the evaluation result. The specific comparison and analysis are as follows:

[0041] If , the adjustment degree of the adaptive weight adjustment mechanism can meet the expectation, and there is no need to optimize the adaptive weight adjustment mechanism;

[0042] If , the adjustment degree of the adaptive weight adjustment mechanism cannot meet the expectation, and it is necessary to optimize the adaptive weight adjustment mechanism. Specifically, it includes: increasing the weight adjustment frequency of the LSTM model and the Kalman filter to improve the response speed of the system to environmental changes; resetting the upper limit of the weight adjustment amplitude to ensure the short-term response effect during the adjustment process; optimizing the weight adjustment parameters to dynamically adjust the weight ratio of the LSTM model and the Kalman filter to adapt to the current flight state requirements; combining historical data and real-time feedback information, analyzing the prediction error and response time, and automatically adjusting the system threshold setting to improve the adaptability and accuracy of the system.

[0043] In the above technical solution, the technical effects and advantages provided by the present invention are as follows:

[0044] 1. By combining the long-term trend prediction of the LSTM model and the short-term error correction of the Kalman filter, the present invention enables the system to quickly adjust the prediction weight under sudden environmental changes. This design is particularly suitable for the altitude prediction requirements of aircraft when there are drastic changes in airflows or environmental mutations, ensuring that the weight distribution between the LSTM and the Kalman filter is dynamically adjusted with environmental changes, and alleviating the problem of phase mismatch. Specifically, the "aircraft altitude monitoring and data acquisition module" captures sensor data in real time and uses the data as prediction input information, enabling the system to promptly sense altitude fluctuations and quickly make weight adjustment decisions, thereby significantly improving the real-time performance and accuracy of the prediction.

[0045] 2. By combining the "phase mismatch prediction and classification module" and the "adaptive weight adjustment module", the present invention realizes a comprehensive evaluation and dynamic feedback of the system adjustment process. The instant adjustment deviation index and the cumulative adjustment effect coefficient provide adjustment effect feedback at both the short-term and long-term levels. By weighted summation to generate an adjustment coefficient and comparing it with a preset threshold, the overall adjustment effect of the system is evaluated. This multi-level feedback mechanism ensures that regardless of any altitude fluctuations during flight, the system can effectively monitor and continuously optimize the adjustment mechanism according to the feedback. In this way, not only the stability of the aircraft altitude prediction is ensured, but also the response sensitivity and long-term stability of the system are improved, providing a solid technical guarantee for flight control safety.

[0046] 3. Through the automatic optimization and adjustment mechanism, the present invention further enhances the self - adaptability and fault tolerance of the system. Specific optimization measures include increasing the weight adjustment frequency, resetting the upper limit of the weight adjustment range, and adjusting the threshold setting based on historical data and real - time feedback, etc. Such an adaptive optimization design enables the system to, in a changing flight environment, such as during rapid climb, landing, encountering sudden airflows, etc., improve the prediction accuracy and control smoothness by dynamically adjusting parameters. This intelligent optimization process significantly improves the applicability of the system in different flight scenarios, enabling it to maintain a high - level prediction stability in a complex and changing environment, reducing the risk of altitude control failure, and lowering energy consumption, thereby enhancing flight safety and the overall performance of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0048] Figure 1 It is a schematic diagram of the modules of the multi - sensor altitude prediction system that integrates LSTM and Kalman filter according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] Now, the exemplary embodiments will be described more comprehensively with reference to the drawings. However, the exemplary embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these exemplary embodiments are provided so that the present disclosure will be more thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art.

[0050] The present invention provides a multi - sensor altitude prediction system that integrates LSTM and Kalman filter as shown in Figure 1 the figure, which includes an aircraft altitude monitoring and data acquisition module, a phase mismatch prediction and classification module, an adaptive weight adjustment module, an adjustment mechanism feedback evaluation module, and a flight process monitoring and optimization module;

[0051] The aircraft altitude monitoring and data acquisition module, during the flight of the aircraft, monitors the altitude data of the aircraft in real - time, and when detecting a sharp fluctuation in the altitude data of the aircraft, obtains the prediction input information required for generating the LSTM model and Kalman filter predictions. These input information include multiple sensor data, such as data from the Global Navigation Satellite System (GNSS), Inertial Navigation System (INS), and barometric altimeter.

[0052] During the flight of an aircraft, real-time monitoring of the aircraft's altitude data can be achieved through multiple sensors on the aircraft. The Global Navigation Satellite System (GNSS) can provide accurate ground positioning information, thereby obtaining the aircraft's altitude data. The Inertial Navigation System (INS) calculates the altitude change of the aircraft through sensors such as accelerometers and gyroscopes, and can still provide continuous altitude data especially in areas where satellite signals cannot be received (such as tunnels or urban environments with dense high-rise buildings). In addition, the barometric altimeter estimates the relative altitude of the aircraft by measuring the change in air pressure. The software system can regularly collect the altitude data output by the sensors through the Sensor Data Interface (API) and store this data in memory or transmit it to the central processing unit in real-time. The system can determine the data collection frequency based on the timestamp or sampling frequency, usually collecting at a millisecond-level frequency to ensure that the altitude data reflects the state of the aircraft in real-time.

[0053] Detecting a sharp fluctuation in the aircraft's altitude data can be achieved by calculating the rate of change of the altitude data. The software system analyzes the altitude data collected each time and calculates the difference between adjacent sampling points. If the altitude change amplitude exceeds a preset threshold within a certain period of time, it indicates that there is a sharp fluctuation. For example, if the altitude data of the aircraft changes by more than the set safe fluctuation range (such as a percentage of a certain altitude range or a specific altitude value) within a short period of time (such as within 1 second), it is considered that a sharp fluctuation has occurred. The software can adopt a sliding window algorithm to analyze the altitude change within a certain time window. If the average rate of change or standard deviation within the window exceeds the predetermined range, the alarm mechanism is triggered. In this way, the system can monitor the altitude change of the aircraft in real-time and quickly detect sharp fluctuations in altitude caused by sudden airflows, wind speed changes or other emergencies.

[0054] When the system detects a sharp fluctuation in the aircraft's altitude data, the software can automatically retrieve and combine the data of multiple sensors to generate the prediction input information required for the LSTM model and the Kalman filter. This data includes the real-time altitude data of the aircraft (such as the data provided by GNSS and INS), flight speed (obtained through INS or external sensors), flight direction (through the trajectory information of the Inertial Measurement Unit IMU or GPS), and barometric information (provided by the barometric altimeter). When a sharp fluctuation occurs, the system constructs a multi-dimensional feature vector based on the real-time collected sensor data as the input for the LSTM and Kalman filters. These input information can be standardized and normalized through a data preprocessing module to ensure the accuracy and applicability of the data. The software system can input this data into the models of the LSTM and Kalman filters for further prediction and error correction to ensure the altitude prediction stability of the aircraft in a complex environment.

[0055] The purpose of this is to accurately capture the altitude fluctuations of the aircraft during sudden environmental changes and timely adjust the prediction mechanism to address the phase mismatch problem between the LSTM and the Kalman filter. When the aircraft encounters sudden airflows or drastic environmental changes, the LSTM model relies on long-term sequence data for trend prediction, while the Kalman filter performs error correction based on short-term data. Due to the prediction lag of the LSTM, conflicts may occur with the short-term adjustment of the Kalman filter, resulting in phase mismatch, which in turn causes frequent oscillations and instability in altitude prediction. By real-time monitoring the altitude data of the aircraft, detecting sharp fluctuations, and timely obtaining prediction input information, the system can quickly analyze and identify the degree of phase mismatch when the aircraft encounters sudden environmental changes, thereby adjusting the weight distribution between the LSTM and the Kalman filter to avoid frequent fluctuations or lags in the prediction results and ensure the accuracy of aircraft altitude control and flight stability. This solution can dynamically adapt to the real-time state of the aircraft, ensuring that the altitude prediction of the aircraft under sudden changes is not affected by lags or mismatches, thus greatly improving the safety and control accuracy of the aircraft.

[0056] The phase mismatch prediction and classification module analyzes the obtained prediction input information, predicts the degree of phase mismatch between the prediction results of the LSTM model and the Kalman filter, and classifies it into three categories: phase consistency, mild mismatch, and severe mismatch according to the prediction results.

[0057] In this embodiment, in the phase mismatch prediction and classification module, the obtained prediction input information is preprocessed, and the prediction time synchronization information and prediction error comparison information in the preprocessed prediction input information are extracted. After extraction, analysis is performed to generate a prediction time deviation index and an altitude prediction error synchronization index respectively; a phase mismatch evaluation model is constructed for the generated prediction time deviation index and altitude prediction error synchronization index to generate a phase mismatch coefficient, and the generated phase mismatch coefficient is compared with a pre-set phase mismatch coefficient threshold interval. According to the comparison result, the degree of phase mismatch between the prediction results of the LSTM model and the Kalman filter is predicted, and it is classified into three categories: phase consistency, mild mismatch, and severe mismatch according to the prediction results.

[0058] In the phase mismatch prediction and classification module, preprocessing the obtained prediction input information is to ensure the quality and consistency of the input data, thereby improving the accuracy and reliability of subsequent phase mismatch evaluation. First of all, preprocessing can eliminate noise and outliers, avoiding data errors caused by sensor accuracy limitations or environmental interference from affecting the analysis results; secondly, preprocessing can standardize the scale of the data, enabling data from different sources to be analyzed under the same dimension, ensuring unity during model processing. In the specific preprocessing process, denoising can be carried out first. For example, moving average filtering or low-pass filtering methods can be used to smooth the fluctuations of height data and eliminate high-frequency noise. Then, outlier detection is performed. For example, the Z-score or box plot method is used to identify and remove outlier height values that exceed the set threshold range to prevent extreme data from interfering with the results. Finally, normalization is carried out, scaling the data of different sensors to a unified range (such as between 0 and 1) according to a certain ratio. Usually, the min-max normalization method is used to standardize the data. Through these preprocessing steps, noise, outliers, and data scale differences in the prediction input information can be effectively controlled, providing high-quality input data for subsequent phase mismatch evaluation.

[0059] The "pre-set phase mismatch coefficient threshold interval" can be determined by statistical analysis based on historical data and experimental calibration methods to ensure the accuracy and robustness of phase mismatch classification. First, the prediction results of the LSTM model and the Kalman filter can be extracted from a large amount of historical flight data, and the phase mismatch coefficients generated by these data under different flight states (such as coefficient values in normal flight, slight perturbation, and severe perturbation conditions) can be calculated. Then, the phase mismatch coefficients are classified through clustering analysis (such as K-means clustering) to determine the corresponding coefficient range for each category, thereby finding the preliminary demarcation points for phase consistency, mild mismatch, and severe mismatch. To improve accuracy, statistical methods can be used, such as calculating the mean and standard deviation of the phase mismatch coefficients for each category, and setting the threshold interval as the mean of each category of coefficients ± 2 times the standard deviation to cover most data points. Finally, software simulation tests are carried out to verify and fine-tune the set threshold interval, and observe whether the performance of the actual flight data in these intervals meets the expected classification criteria. Through this process, a robust phase mismatch coefficient threshold interval can be obtained, thereby effectively classifying the degree of phase mismatch between the LSTM and the Kalman filter in practical applications.

[0060] In this embodiment, the acquisition logic of the prediction time deviation index and the height prediction error synchronization index is as follows:

[0061] Extract the prediction time synchronization information from the preprocessed prediction input information, specifically including the timestamps of the prediction results generated by the LSTM model at different moments within a period of time, the timestamps of the prediction results generated by the Kalman filter, and the timestamps of the actual acquisition of the aircraft altitude data, and label them respectively as 、 and , represents the timestamp of the prediction result generated by the LSTM model at the moment within a period of time, represents the timestamp of the prediction result generated by the Kalman filter at the moment within a period of time, represents the timestamp of the actual acquisition of the aircraft altitude data at the moment within a period of time, , is a positive integer;

[0062] To extract the prediction time synchronization information from the preprocessed prediction input information, the corresponding timestamps need to be extracted from the LSTM model, the Kalman filter, and the sensor data respectively. First, for the timestamp of the prediction result generated by the LSTM model, the time when the LSTM model outputs the predicted altitude result can be recorded in real time by software. Usually, these timestamps are automatically generated by the internal clock of the system and attached to the output result. Each time the LSTM model generates a new prediction result within a time window, the system will automatically save the prediction result and its generated timestamp to form time series data. Next, the timestamp of the Kalman filter can be recorded in the same way, that is, whenever the Kalman filter corrects the error of the sensor data and generates a new altitude value, the system will also save the corresponding timestamp to ensure that it forms a comparable data pair with the timestamp of the LSTM prediction result. Finally, the timestamp of the actual aircraft altitude data is obtained by collecting through sensors (such as GNSS, INS, barometric altimeter). Each time the sensor collects new altitude data, the system will automatically attach the timestamp of the sensor collection. Usually, these timestamps are directly generated by the system clock provided by the hardware, reflecting the actual data acquisition moment. Through the time synchronization module of the software, the timestamp data from different data sources are aligned along the time axis to form a synchronous record of a set of three types of timestamp data. Then, within the set time window, the timestamp data of the LSTM, Kalman, and sensor are extracted in sequence and stored as the prediction time synchronization information within a period of time, which can ensure that each timestamp data corresponds precisely and provide high-quality input information for subsequent time deviation analysis.

[0063] Calculate the prediction time deviation index, and the specific calculation formula is as follows:

[0064]

[0065] In the formula, is the prediction time deviation index;

[0066] This formula is used to calculate the prediction time deviation index , whose purpose is to quantify the synchronization degree between the LSTM model and the Kalman filter on the time axis, so as to evaluate the time deviation when they generate prediction results. This formula consists of several key steps. First, represents the time difference between the prediction result generated by the LSTM model at time point and the time when the actual height data is obtained, which reflects the delay or advance of the LSTM prediction result. Similarly, represents the time difference between the prediction result of the Kalman filter and the actual height data, reflecting the time offset of the Kalman filter. Adding "+1" to the denominator part is to prevent the occurrence of division by zero and make the calculation more stable. Next, the absolute ratio of the LSTM time difference to the Kalman time difference is squared to enhance the penalty effect when the deviation is large and highlight the situation of time asynchrony. The steps of summing and taking the average unify the time deviations at each time point into an average value, thereby reflecting the time synchronization between the LSTM and the Kalman filter within the entire time window. Through this formula, the overall level of the time deviation between the LSTM and the Kalman filter can be quantified, providing an accurate reference index for the subsequent analysis and classification of phase mismatch.

[0067] The prediction time deviation index directly reflects the synchronization degree between the prediction results of the LSTM model and the Kalman filter on the time axis, and further reflects the degree of phase mismatch between them. Specifically, a smaller value indicates a smaller time deviation between the LSTM model and the Kalman filter, that is, the time when they generate prediction results is closer, indicating that their prediction results are more synchronized in time and the phases are basically the same; while a larger value indicates a significant time deviation between the prediction results of the LSTM and the Kalman filter, and the generated prediction times are inconsistent, resulting in out-of-phase synchronization. The greater this asynchrony, the more serious the degree of phase mismatch. Therefore, the larger the

[0068] value, the higher the degree of phase mismatch between the two, which may cause the system to be unable to balance the long-term trend prediction of the LSTM and the short-term correction of the Kalman filter, thereby affecting the stability and accuracy of the overall prediction. Extract the prediction error comparison information in the preprocessed prediction input information, specifically including the height prediction values of the LSTM model, the height prediction values of the Kalman filter, and the actual height of the aircraft obtained at different times within a period of time, and label them as , and , represent the predicted height values of the LSTM model at time over a period of time, represent the predicted height values of the Kalman filter at time over a period of time, and represent the actual height of the aircraft obtained at time

[0069] Extract the prediction error comparison information from the pre - processed prediction input information, which mainly involves obtaining the predicted height values of the LSTM model, the predicted height values of the Kalman filter, and the actual height of the aircraft over a period of time. These data can be extracted from different modules through a software system and aligned within the same time window. First, the predicted height values of the LSTM model are generated by a deep - learning algorithm based on historical sensor data. They can be automatically recorded by the software each time the prediction result is calculated. These predicted values are usually used as the model output and are stored in the database with a timestamp attached when generated, forming a time series of height predictions. Second, the predicted height values of the Kalman filter are based on real - time error correction of the current sensor data. Each time the Kalman filter updates the height prediction result, the software will also store the predicted value and the corresponding timestamp. To ensure data consistency, the system will ensure that each height prediction value is stored in chronological order for alignment with the LSTM model output. Finally, the actual height data of the aircraft is collected in real - time by sensors (such as GNSS, INS, and barometric altimeters). The height value output by the sensor is directly obtained through a software interface, along with the sensor's timestamp. The software system will extract the values corresponding to the time points within the set time window from the LSTM predicted values, Kalman filter values, and sensor actual height values respectively, and synchronize these data to ensure that the three types of values are obtained at the same time point. Through these methods, the system can generate consistent prediction error comparison information, ensuring that the LSTM predicted values, Kalman filter predicted values, and actual height data at each moment accurately correspond, providing high - quality input data for subsequent error analysis and phase mismatch assessment.

[0070] Calculate the height prediction error synchronization index. The specific calculation formula is as follows:

[0071]

[0072] where is the height prediction error synchronization index.

[0073] This formula is used to calculate the height prediction error synchronization index. , whose purpose is to quantify the synchronization between the LSTM model and the Kalman filter in predicting altitude errors to evaluate the phase consistency between them. In the formula, represents the error between the altitude prediction value of the LSTM model and the actual altitude value at time point . This term quantifies the prediction bias of the LSTM model. Similarly, represents the error between the altitude prediction value of the Kalman filter and the actual altitude value, which reflects the prediction bias of the Kalman filter. By calculating the difference between these two errors, the formula can obtain the degree of asynchrony of the prediction errors between the LSTM model and the Kalman filter. The in the denominator serves as a normalization function to reduce the impact on error calculation due to different ranges of actual altitude changes, and “+1” can avoid division-by-zero errors when the actual altitude is zero. The absolute value operation is used to eliminate the influence of positive and negative biases, making the calculation of the synchronization index more stable. Finally, by summing and averaging the degree of asynchrony of errors at all time points, can measure the error synchronization between the LSTM and Kalman filter prediction results over the entire time window. The larger the value, the more asynchronous the error changes between the two, and the higher the degree of phase mismatch; while a smaller value indicates that the error changes between the two are more consistent, that is, the phase synchronization is better.

[0074] Altitude prediction error synchronization index directly reflects the synchronization between the LSTM model and the Kalman filter in predicting altitude errors, and further reflects the degree of phase mismatch between them. Specifically, a smaller value indicates that the changes in prediction errors of the LSTM model and the Kalman filter are synchronized, that is, the deviations of the two from the actual altitude are basically the same, indicating that their prediction results are more consistent in phase; on the contrary, a larger value indicates that the changes in prediction errors of the LSTM and the Kalman filter are asynchronous, indicating that there are significant differences in the deviations of the two from the actual altitude, resulting in phase mismatch. Therefore, the larger the value, the higher the degree of phase mismatch, which will affect the accurate prediction of the aircraft altitude and the control stability of the system. While a smaller value indicates that the LSTM and the Kalman filter are more synchronized in time, with a lower degree of phase mismatch and better coordination of the prediction results.

[0075] In this embodiment, a phase mismatch evaluation model is constructed for the generated prediction time deviation index and the altitude prediction error synchronization index . A phase mismatch coefficient is generated through weighted summation, and the generated phase mismatch coefficient Compare with the pre-set phase mismatch coefficient threshold interval Perform a comparison, predict the phase mismatch degree between the prediction results of the LSTM model and the Kalman filter according to the comparison result, and classify it into three categories: phase consistency, mild mismatch, and severe mismatch according to the prediction result. The specific comparison analysis and classification are as follows:

[0076] If , and the phase mismatch degree between the prediction results of the LSTM model and the Kalman filter is at a low level, then this situation is classified as phase consistency;

[0077] This situation indicates that the prediction results of the LSTM model and the Kalman filter are highly synchronized in time, and the error changes are basically the same, indicating that there is almost no deviation in phase between the two prediction results. In this case, the long-term trend prediction of the LSTM model and the short-term correction of the Kalman filter can work in coordination to achieve accurate prediction of the aircraft altitude. The state of phase consistency can ensure the stability and accuracy of the system prediction, provide reliable altitude data support for the aircraft, and contribute to smooth flight and precise control.

[0078] If , and the phase mismatch degree between the prediction results of the LSTM model and the Kalman filter is at a medium level, then this situation is classified as mild mismatch;

[0079] This situation shows that there is a certain time deviation or asynchronous error change between the prediction results of the LSTM model and the Kalman filter, but the deviation is small, belonging to the mild mismatch state. In this case, the prediction results of the two are slightly out of sync in phase, which may cause slight fluctuations or instability in the altitude prediction of the system. Although this degree of mismatch will not significantly affect the safety of the aircraft, in flight tasks that require highly accurate altitude prediction, it may cause the aircraft to respond slightly sluggishly or inaccurately. Therefore, the system may need to make minor adjustments to improve the prediction effect.

[0080] If , and the phase mismatch degree between the prediction results of the LSTM model and the Kalman filter is at a high level, then this situation is classified as severe mismatch.

[0081] This situation indicates that there is a significant asynchrony in time between the prediction results of the LSTM model and the Kalman filter, with a large difference in the variation of errors, being in a severe mismatch state. In this case, the phases of the prediction results of the two are significantly asynchronous, and they cannot work effectively in cooperation, resulting in a substantial reduction in the prediction accuracy of the aircraft altitude by the system. Frequent prediction errors and instabilities may occur. This severe mismatch will affect the altitude control of the aircraft, increasing the risk when the aircraft performs precise operations (such as landing or obstacle avoidance), and may lead to ineffective or unstable control responses. Therefore, positive adjustment measures need to be taken to improve the phase synchrony and ensure flight safety.

[0082] It is necessary to construct a phase mismatch evaluation model and generate a phase mismatch coefficient , first, it is necessary to perform a weighted sum on the generated prediction time deviation index and the altitude prediction error synchrony index . Specifically, is used to measure the synchrony between the LSTM model and the Kalman filter on the time axis, while is used to evaluate the synchrony between the two in terms of prediction errors. To generate , two weight coefficients and can be set to adjust the influence of and in the final phase mismatch coefficient respectively. The selection of the weight coefficients should be based on the degree of attention to time synchrony and error synchrony in practical applications. For example, in flight missions with higher requirements for time synchrony, can be set slightly larger to increase the influence of , while in scenarios with higher precision requirements, the proportion of can be increased. The final phase mismatch coefficient is calculated by the following formula: , where the values of and need to be optimized through a large amount of experimental data analysis or actual flight tests to ensure that the phase mismatch evaluation model can accurately reflect the phase synchrony of the LSTM and the Kalman filter, and thus provide a reliable basis for the classification of the degree of phase mismatch.

[0083] An adaptive weight adjustment module, according to the classification results, constructs an adaptive weight adjustment mechanism to perform different weight adjustment measures for the three situations of phase consistency, mild mismatch, and severe mismatch respectively;

[0084] In this embodiment, in the adaptive weight adjustment module, an adaptive weight adjustment mechanism is constructed according to the division result. Specifically: according to the division results of phase consistency, mild mismatch, and severe mismatch, different weight adjustment parameters are set, corresponding to the weight allocation of the LSTM model and the Kalman filter in predictive fusion, forming an adaptive weight adjustment mechanism; based on the range of the phase mismatch coefficient, this adjustment mechanism automatically determines the weight allocation strategy and adjustment amplitude between the LSTM model and the Kalman filter through pre-set rules.

[0085] In the adaptive weight adjustment module, to achieve automatic adjustment of the weight allocation between the LSTM model and the Kalman filter according to the division results of phase consistency, mild mismatch, and severe mismatch, it can be realized through a dynamic weight allocation algorithm and rule setting. Specifically, first, the range of the phase mismatch coefficient and the corresponding weight adjustment rules need to be defined in the software, that is, the numerical interval of the pre-set phase mismatch coefficient (such as the thresholds for phase consistency, mild mismatch, and severe mismatch) and the weight ratios of the LSTM and Kalman filters corresponding to each interval are set. For example, in the software, an equal weight is set for the phase consistency state, a slightly higher weight of the Kalman filter is set for the mild mismatch state, and a higher weight of the Kalman filter is set for the severe mismatch state to ensure that the system has appropriate response strategies under different mismatch states. Then, the system calculates the current phase mismatch coefficient in real time and automatically matches the corresponding weight adjustment rules according to the value of this coefficient. In this way, the software can automatically adjust the weight allocation of the LSTM and Kalman filters based on the current mismatch degree, thereby dynamically balancing the relationship between long-term trend prediction (LSTM) and short-term error correction (Kalman filtering). Through this implementation method, the system can adapt to different phase mismatch situations, improve the accuracy and stability of prediction, and thus ensure the reliability of aircraft altitude prediction.

[0086] Different weight adjustment measures are taken for the three cases of phase consistency, mild mismatch, and severe mismatch respectively. Specifically: in the case of phase consistency, an equal weight strategy is adopted, and equal weights are assigned to the LSTM model and the Kalman filter to maintain stable prediction; in the case of mild mismatch, the weight of the Kalman filter is adjusted to be higher than that of the LSTM model to increase the influence of short-term correction in prediction, so that while the system maintains the balance of long-term and short-term predictions, it enhances the response ability to current environmental changes; in the case of severe mismatch, the weight of the Kalman filter is significantly increased, and the weight of the LSTM model is significantly reduced to ensure that the system can quickly respond to sudden changes and enhance the short-term stability of altitude prediction.

[0087] To implement different weight adjustment measures for three cases of phase consistency, mild mismatch, and severe mismatch, it can be automatically achieved in software through a dynamic weight control algorithm combined with conditional judgment rules. Specifically, the software can first set the weight allocation rules for phase consistency, mild mismatch, and severe mismatch. For example, when the phase is consistent, the LSTM model and the Kalman filter are assigned equal weights; when there is a mild mismatch, the weight ratio of the Kalman filter is increased, and the weight of the LSTM is slightly decreased; when there is a severe mismatch, the weight of the Kalman filter is significantly increased, and the weight of the LSTM is significantly decreased. The system will judge the current mismatch state based on the phase mismatch coefficient calculated in real time and automatically select the corresponding weight configuration strategy. Whenever the mismatch state changes, the software will automatically adjust the weights of the LSTM and the Kalman filter based on the preset weight rules to adapt to the current environmental requirements. The purpose of this dynamic regulation method is to maintain the smoothness of prediction in the phase-consistent state, moderately improve the short-term correction ability in the mild mismatch state, and give priority to quickly responding to sudden changes in the severe mismatch state, thereby enhancing the overall prediction accuracy and stability of the system.

[0088] The adjustment mechanism feedback evaluation module, during the process of the adaptive weight adjustment mechanism adjusting the weights, obtains the adjustment mechanism feedback information in real time, analyzes it after obtaining, evaluates whether the adjustment degree of the adaptive weight adjustment mechanism can meet the expectations, and optimizes the adaptive weight adjustment mechanism according to the evaluation results;

[0089] In this embodiment, in the adjustment mechanism feedback evaluation module, after obtaining the adjustment mechanism feedback information in real time, it performs preprocessing on it, extracts the instantaneous prediction adjustment feedback information and the long-term adjustment cumulative feedback information in the preprocessed adjustment mechanism feedback information, analyzes them after extraction, and generates an immediate adjustment deviation index and a cumulative adjustment effect coefficient respectively; constructs an adjustment degree evaluation model for the generated immediate adjustment deviation index and cumulative adjustment effect coefficient, generates an adjustment coefficient, compares the generated adjustment coefficient with a preset adjustment coefficient threshold, evaluates whether the adjustment degree of the adaptive weight adjustment mechanism can meet the expectations according to the comparison result, and optimizes the adaptive weight adjustment mechanism according to the evaluation results.

[0090] During the process of weight adjustment by the adaptive weight adjustment mechanism, real-time acquisition of the feedback information of the adjustment mechanism can be achieved through data monitoring and automatic recording. Specifically, the software system can set up a real-time monitoring module to automatically capture the output prediction values, real-time errors, and current adjusted weight parameters of the LSTM model and the Kalman filter after each adjustment. Whenever the weight adjustment mechanism fine-tunes the weights of the LSTM and the Kalman filter, the system will immediately call the corresponding sensor interface to obtain the actual altitude data from the altitude sensors of the aircraft (such as GNSS, INS, etc.). At the same time, the system will take the current predicted altitude, actual altitude, real-time error, and weight parameters as part of the feedback information, attach timestamps, and store them in the temporary data cache to ensure the integrity of the time series of the data. At each time interval during the entire adjustment process, the system automatically records this feedback information and synchronizes it to the feedback evaluation module for subsequent analysis and processing. Through this method of automatic software monitoring and recording, the system can achieve real-time acquisition of the feedback information of the adjustment mechanism, ensure the timeliness and integrity of the feedback data, and thus support the immediate evaluation of the adjustment effect.

[0091] In the feedback evaluation module of the adjustment mechanism, the real-time acquired feedback information of the adjustment mechanism needs to be preprocessed to ensure the accuracy, consistency, and reliability of the data, thereby improving the accuracy of the evaluation. First, the feedback information may contain sensor noise and outliers, so filtering (such as Kalman filtering or moving average filtering) is required to remove the noise, and threshold filtering is used to eliminate abnormal readings. Then, to ensure that the feedback information from the LSTM model, the Kalman filter, and the actual altitude can be compared and analyzed at the same time point, the system needs to align the data in time to make the data sequences consistent. In addition, data normalization is also necessary to standardize the data from different sources to a unified scale or relative ratio to ensure comparison in the same dimension. Through such preprocessing, the feedback data is smoother, more standardized, and more consistent, providing a reliable basis for the subsequent evaluation of the adjustment effect.

[0092] The "predetermined adjustment coefficient threshold" can be determined based on historical data analysis and simulation test results to ensure that the threshold adapts to different flight scenarios and meets the system's stability and response requirements. Specifically, the software system can first import a large amount of flight data, including real-time feedback data of the LSTM model and Kalman filter during the operation of the adjustment mechanism under different flight conditions. Then, the data analysis module statistically processes these historical data, calculates the average value, variance, and distribution range of each adjustment coefficient in different scenarios. Subsequently, the system can use the simulation environment to conduct multiple rounds of weight adjustment tests, simulate various emergencies and altitude changes, and observe the impact of different adjustment coefficients on the system performance. According to the simulation results, the software will automatically record the minimum adjustment coefficient value that ensures prediction stability and the maximum adjustment coefficient value for rapid response in each test, and then determine a suitable threshold range. Finally, the system will use this range as the basis for setting the adjustment coefficient threshold and make fine-tuning according to the evaluation results in actual applications to optimize the adjustment effect and meet the altitude control requirements of the aircraft in various environments.

[0093] In this embodiment, the acquisition logic of the instant adjustment deviation index and the cumulative adjustment effect coefficient is as follows:

[0094] Extract the instantaneous prediction adjustment feedback information from the preprocessed adjustment mechanism feedback information, specifically including the difference between the predicted altitude of the LSTM model and Kalman filter combination and the actual altitude of the aircraft at different moments within a period during the process of the adaptive weight adjustment mechanism adjusting the weights, the weight difference between the LSTM model and Kalman filter, and the time interval of the system's response to the altitude change of the aircraft after the weight adjustment, and calibrate them respectively as 、 and , represents the difference between the predicted altitude of the LSTM model and Kalman filter combination and the actual altitude of the aircraft at the moment within a period during the process of the adaptive weight adjustment mechanism adjusting the weights, represents the weight difference between the LSTM model and Kalman filter at the moment within a period during the process of the adaptive weight adjustment mechanism adjusting the weights, represents the time interval of the system's response to the altitude change of the aircraft after the weight adjustment, , is a positive integer;

[0095] When extracting the instantaneous prediction adjustment feedback information from the pre - processed adjustment mechanism feedback information, the software system can obtain and record the differences between the predicted height and the actual height, the weight differences of the LSTM model and the Kalman filter, and the time interval of the system response through the real - time data acquisition and data synchronization processing module. The specific extraction process is as follows: First, during the adjustment process of the adaptive weight adjustment mechanism, the software monitors and records the predicted height value after the combination of the LSTM and the Kalman filter and the actual height value from the aircraft height sensor (such as data provided by sensors like GNSS, INS, etc.). The system automatically calculates the difference between the two as the real - time prediction error data. This error value can reflect the deviation between the prediction result and the true height, indicating the prediction effect after adjustment. At the same time, the software automatically records the current weight difference between the LSTM and the Kalman filter (i.e., the relative change in their weights), which represents the specific adjustment amplitude of the weights during the adjustment process and helps to quantify the strength of the adjustment mechanism. In addition, the software marks the response time after adjustment, that is, records the time interval of the system's response to the change in the actual height after the weight adjustment takes effect. By tracking the start and end of each adjustment through the built - in timestamp, the response speed of the system can be calculated. Through this set of real - time data acquisition and synchronous calculation mechanisms, the software can automatically extract the required instantaneous prediction adjustment feedback information, providing accurate basic data for subsequent evaluation of the adjustment effect.

[0096] Calculate the immediate adjustment deviation index, and the specific calculation formula is as follows:

[0097]

[0098] In the formula, is the immediate adjustment deviation index;

[0099] The immediate adjustment deviation index The calculation formula is designed to measure the response effect of the adaptive weight adjustment mechanism in a short period of time, and each operation step in the formula serves this goal. First, represents the instantaneous error between the predicted height and the actual height after each adjustment, which reflects whether the weight adjustment can promptly reduce the prediction deviation. In the denominator, represents the weight difference between the LSTM model and the Kalman filter, which is used to quantify the strength of the weight adjustment; represents the system response time after adjustment, reflecting the immediate reaction speed of the adjustment mechanism. Through 's weighted denominator structure, the relationship between the error and the adjustment amplitude can be balanced to ensure that situations with a slower response speed or a larger adjustment amplitude will not overly affect the calculation result. Finally, the sum of the squares of all instantaneous errors is divided by the number of adjustments The averaging process can smooth out short-term fluctuations and obtain an overall immediate response effect evaluation. Generally speaking, this formula effectively balances errors, adjustment amplitudes, and response speeds, and can objectively evaluate the immediate adjustment effect of the adjustment mechanism.

[0100] Immediate adjustment deviation index The magnitude directly reflects the improvement effect of the adaptive weight adjustment mechanism on prediction accuracy in the short term, and further evaluates whether the adjustment mechanism has achieved the expected adjustment degree. Specifically, The smaller the value, the more rapidly the error between the predicted height and the actual height decreases after weight adjustment, and the faster the system responds to height changes. This indicates that the immediate effect of the weight adjustment mechanism is good and meets the expected adjustment goal; conversely, if the value is large, it means that after weight adjustment, the prediction error is still large or the response speed after adjustment is slow, and it cannot quickly adapt to the requirements of height changes, indicating that the immediate adjustment effect of the adjustment mechanism is poor and does not meet the expected requirements. Therefore, by observing the numerical magnitude, it is possible to determine whether the immediate response ability and accuracy of the current adjustment mechanism meet the set performance indicators.

[0101] Extract the long-term adjustment cumulative feedback information from the preprocessed adjustment mechanism feedback information, specifically including the cumulative prediction error change amount after multiple weight adjustments within the time window when the adaptive weight adjustment mechanism adjusts the weight, the cumulative amplitude of all weight adjustments within this time window, and the time length within this time window, and label them as 、 and , represents the cumulative prediction error change amount after multiple weight adjustments within the time window when the adaptive weight adjustment mechanism adjusts the weight, represents the cumulative amplitude of all weight adjustments within this time window, represents the time length within this time window;

[0102] The extraction of long-term adjustment cumulative feedback information can be achieved through the long-term data monitoring and cumulative calculation module of the software system, so that the system can evaluate the cumulative effect of multiple adjustments within the time window when the adaptive weight adjustment mechanism is running. Specifically, the system automatically records the prediction error during each weight adjustment, that is, the difference between the predicted height by the combination of LSTM and Kalman filter and the actual height. The system will accumulate these instantaneous errors to calculate the cumulative change in prediction error, thereby reflecting the overall error change trend within this time window. At the same time, the system also records the magnitude of the weight change of LSTM and Kalman filter during each adjustment, that is, the weight difference between LSTM and Kalman filter during each adjustment, and accumulates these weight differences successively within the time window to calculate the cumulative magnitude of weight adjustment, indicating the total intensity of adjustment within this window. In addition, the system marks the time of each adjustment through timestamps and calculates the time length from the start to the end at the end of the window, that is, the time span within the entire window. These data are automatically extracted and stored through the cumulative calculation and data recording module of the software, providing a complete data basis for subsequent evaluation of the long-term effect of weight adjustment.

[0103] Calculate the cumulative adjustment effect coefficient, and the specific calculation formula is as follows:

[0104]

[0105] In the formula, is the cumulative adjustment effect coefficient.

[0106] Cumulative adjustment effect coefficient The calculation formula of aims to evaluate the error convergence effect of the adaptive weight adjustment mechanism in the long term, and each operation step in the formula has a specific evaluation role. First, in the numerator, represents the cumulative change in prediction error after multiple weight adjustments within the time window, reflecting whether the prediction error gradually decreases after multiple adjustments. If is small, it indicates that the error gradually decreases and the adjustment effect is good. In the denominator, represents the cumulative magnitude of weight adjustment, that is, the total magnitude of all weight adjustments within this window, which measures the intensity of the adjustment mechanism in the long term. By calculating the ratio of to , the convergence situation of the prediction error under multiple adjustments can be reflected. To further quantify the degree of error convergence, this ratio is squared to emphasize the significance of the error change after adjustment. Finally, the entire result is divided by , where is the length of the time window, ensuring that when the time is long, the value decreases, indicating the persistence of the adjustment effect in the long-term evaluation. Therefore, The smaller the value, the better the effect of weight adjustment in the long term and the more stable the error tends to be; A larger value indicates that the adjustment effect is not ideal and the error is difficult to converge.

[0107] Cumulative adjustment effect coefficient The magnitude directly reflects the ability of the adaptive weight adjustment mechanism to control the prediction error in the long term and evaluates whether its adjustment effect meets the expectations. Specifically, The smaller the value, it means that within the time window after weight adjustment, the cumulative change of the prediction error tends to converge, and the magnitude of the weight adjustment is relatively stable. This indicates that the adjustment mechanism effectively reduces the error in the long term, achieves a good error control effect, and meets the expected adjustment goal; on the contrary, if The value is large, indicating that the adjustment mechanism fails to effectively reduce the error in a long time, the error change is large or the adjustment magnitude is unstable, indicating that the long-term effect of the adjustment mechanism is not ideal and does not meet the expected requirements. Therefore, by observing the numerical magnitude, it can be judged whether the current adjustment mechanism achieves the expected adjustment effect in terms of long-term error control.

[0108] In this embodiment, for the generated instant adjustment deviation index and the cumulative adjustment effect coefficient a degree-of-adjustment evaluation model is constructed, and the adjustment coefficient is generated through weighted summation, and the generated adjustment coefficient is compared with the preset adjustment coefficient threshold to evaluate whether the adjustment degree of the adaptive weight adjustment mechanism can meet the expectations according to the comparison result, and the adaptive weight adjustment mechanism is optimized according to the evaluation result. The specific comparison and analysis are as follows:

[0109] If the adjustment degree of the adaptive weight adjustment mechanism can meet the expectations and there is no need to optimize the adaptive weight adjustment mechanism;

[0110] This situation means that the current adjustment degree of the adaptive weight adjustment mechanism reaches the expected effect, can effectively balance the weight distribution between the LSTM model and the Kalman filter, and accurately track the change of the flight altitude. In this case, the system does not need to further optimize the adjustment mechanism, indicating that the current adjustment strategy has good stability and adaptability in practical applications and can maintain the accuracy of altitude prediction and the sensitivity of system response under the current environmental conditions.

[0111] If , the adjustment degree of the adaptive weight adjustment mechanism cannot meet the expectation, and it is necessary to optimize the adaptive weight adjustment mechanism, specifically including: increasing the weight adjustment frequency of the LSTM model and the Kalman filter to improve the system's response speed to environmental changes; resetting the upper limit of the weight adjustment amplitude to ensure the short-term response effect during the adjustment process; optimizing the weight adjustment parameters to dynamically adjust the weight ratio of the LSTM model and the Kalman filter to meet the current flight state requirements; combining historical data and real-time feedback information to analyze the prediction error and response time, and automatically adjust the system threshold setting to improve the system's adaptability and accuracy.

[0112] This situation indicates that the adjustment effect of the adaptive weight adjustment mechanism fails to meet the expectation, which may lead to an increase in prediction error or insufficient system response speed, making it difficult to cope with complex or sudden flight environments. In this case, it is necessary to optimize the adjustment mechanism. By increasing the adjustment frequency, resetting the weight adjustment amplitude, dynamically adjusting the weight parameters, etc., the system's error convergence speed and adjustment adaptability can be improved to ensure that the aircraft can still maintain a high prediction accuracy and stability in a changing environment.

[0113] To optimize the adaptive weight adjustment mechanism to meet the adjustment requirements under the current conditions, the software can achieve this optimization process in the following ways. First, by adding a monitoring module for the weight adjustment frequency in the system, the software can automatically increase the weight update rate of the LSTM model and the Kalman filter, thereby shortening the weight adjustment cycle, ensuring that the system can respond to environmental changes more timely, and improving the adaptability to emergencies. Second, the software can dynamically set the upper limit of the weight adjustment amplitude, that is, set the maximum allowable weight change amplitude in each adjustment to ensure an effective response in the short term. This setting can be adjusted based on real-time error feedback. When the error increases, the adjustment amplitude also increases accordingly to ensure that the system can make a rapid adjustment when the error suddenly increases. In addition, the software can optimize the weight adjustment parameters through intelligent algorithms, such as adopting an adaptive parameter adjustment method to adjust the weight ratio of the LSTM model and the Kalman filter in real time according to the current flight state (such as speed, attitude, etc.) to optimize the prediction effect for different flight state requirements. Finally, the software can combine historical data and real-time feedback information, calculate the average value and change trend of the prediction error and response time through the data analysis module, and automatically adjust the system threshold setting, so as to appropriately lower the threshold in a more variable environment to improve sensitivity, and appropriately increase the threshold in a stable environment to enhance stability. These optimization measures not only enable the system to more flexibly adapt to different flight scenarios and sudden situations, but also improve the overall prediction accuracy and stability of the system, achieving continuous optimization.

[0114] When constructing the adjustment degree evaluation model, the software system will be based on the generated immediate adjustment deviation index and the cumulative adjustment effect coefficient to evaluate the overall adjustment effect of the adaptive weight adjustment mechanism. Specifically, the system will perform a weighted sum on and to generate a comprehensive adjustment coefficient , which is used to measure the comprehensive performance of the adjustment effect. Assuming the weight coefficients are and respectively, the calculation formula for the adjustment coefficient is: . Among them, and represent the weight coefficients of and respectively, and these coefficients are set according to the requirements of the adjustment model. Usually, can be set to a relatively high value to emphasize the immediate response effect of the system and ensure that the adjustment mechanism responds quickly in the short term; while is used to balance the long-term stability and error convergence effect of the system, so it can be dynamically adjusted according to different flight environments. For example, in a stable environment, the weight of can be increased to emphasize the long-term effect, and in a frequently changing environment, the weight of can be increased to improve the immediate response ability of the system. Through this weighted sum, the adjustment coefficient can comprehensively reflect the adjustment effect of the adaptive weight adjustment mechanism in both the immediate and long-term dimensions, thus achieving an accurate assessment of the adjustment degree.

[0115] The flight process monitoring and optimization module monitors and analyzes the altitude prediction data and adjustment records during the entire flight process in real time, and continuously optimizes the adaptive weight adjustment mechanism according to the monitoring results to ensure the altitude prediction stability and control accuracy of the aircraft under different flight stages and environmental conditions.

[0116] The flight process monitoring and optimization module can achieve real-time monitoring and dynamic optimization of the flight process by establishing a closed-loop system for data collection, analysis, and feedback regulation. Specifically, the software first collects the altitude prediction data and adjustment records of the aircraft in real time through the sensor interface, including information such as the real-time predicted altitude, actual altitude, weight adjustment frequency and amplitude of the LSTM model and Kalman filter. These data are stored in the time series database of the system and are accompanied by accurate timestamps for historical analysis. Then, the software uses the data analysis module to perform statistical and trend analysis on the real-time data through algorithms, such as calculating the mean error, variance, prediction error fluctuation, etc., to evaluate the current prediction stability and accuracy. Based on these analysis results, the software can identify the error characteristics and control requirements under different flight phases (such as climb, cruise, descent) and environmental changes (such as air flow, temperature change). Then, the system will dynamically adjust the parameters of the adaptive weight adjustment mechanism according to the analysis results. For example, when the error is large or the environment changes suddenly, the weight adjustment frequency or response speed will be increased to enhance the adaptability of the system, while the adjustment frequency will be reduced when the environment is stable to ensure the stability and energy efficiency of the system. This closed-loop system of real-time monitoring and feedback optimization ensures that the aircraft can maintain the accuracy of altitude prediction and the stability of control under various flight phases and environmental conditions, providing data support and optimization guarantee for flight safety.

[0117] The above formulas are all dimensionless and take their numerical calculations. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formulas are set by technicians in this field according to the actual situation.

[0118] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that contains one or more collections of available media. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, or a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0119] It should be understood that in various embodiments of the present application, the magnitudes of the sequence numbers of the above processes do not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0120] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. A professional technician can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0121] In several embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the above-described embodiments are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be indirect couplings or communication connections through some interfaces, devices, or units, and can be in an electrical, mechanical, or other form.

[0122] The unit described as a separation component may or may not be physically separated. The component displayed as a unit may or may not be a physical unit, that is, it may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0123] In addition, each functional unit in various embodiments of the present application may be integrated in a processing unit, may exist separately as individual physical units, or two or more units may be integrated in one unit.

[0124] As mentioned above, the above are only specific implementation manners of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A multi-sensor height prediction system integrating LSTM and Kalman filtering, characterized by: It includes aircraft altitude monitoring and data acquisition module, phase mismatch prediction and classification module, adaptive weight adjustment module, adjustment mechanism feedback evaluation module and flight process monitoring and optimization module; The aircraft altitude monitoring and data acquisition module monitors the aircraft altitude data in real time during the flight of the aircraft, and obtains the prediction input information required for generating LSTM model and Kalman filter prediction when a sharp fluctuation in the aircraft altitude data is detected; The phase mismatch prediction and classification module analyzes the obtained prediction input information, predicts the phase mismatch degree between the prediction results of the LSTM model and the Kalman filter, and divides them into three categories according to the prediction results: phase consistency, mild mismatch, and severe mismatch; In the phase mismatch prediction and classification module, the acquired prediction input information is preprocessed, and the prediction time synchronization information and prediction error comparison information in the preprocessed prediction input information are extracted, and analyzed after extraction to generate a prediction time deviation index and a height prediction error synchronization index respectively; a phase mismatch evaluation model is constructed for the generated prediction time deviation index and the height prediction error synchronization index to generate a phase mismatch coefficient, and the generated phase mismatch coefficient is compared with a pre-set phase mismatch coefficient threshold interval, and the phase mismatch degree between the prediction results of the LSTM model and the Kalman filter is predicted according to the comparison results, and it is divided into three categories according to the prediction results: phase consistency, mild mismatch and severe mismatch; The acquisition logic of the prediction time deviation index and the height prediction error synchronization index is as follows: Extract the prediction time synchronization information from the preprocessed prediction input information, including the timestamps of the LSTM model generating prediction results at different times within a period of time, the timestamps of the Kalman filter generating prediction results, and the timestamps of actually obtaining the aircraft altitude data, and mark them as , and , Indicates that within a period of time The timestamp of the moment when the LSTM model generates the prediction result, Indicates that within a period of time The timestamp of the moment the Kalman filter generates the prediction result, Indicates that within a period of time The timestamp of the actual acquisition of the aircraft altitude data at the moment, , is a positive integer; Calculate the forecast time deviation index. The specific calculation formula is as follows: ； In the formula, is the forecast time deviation index; Extract the prediction error comparison information from the preprocessed prediction input information, including the altitude prediction value of the LSTM model at different times in a period of time, the altitude prediction value of the Kalman filter, and the actual altitude of the aircraft obtained, and calibrate them as , and , Indicates that within a period of time The height prediction value of the LSTM model at the moment, Indicates that within a period of time The height prediction value of the Kalman filter at time instant, Indicates that within a period of time The actual altitude of the aircraft obtained at all times; Calculate the height prediction error synchronization index. The specific calculation formula is as follows: ； In the formula, is the height forecast error synchronicity index; The adaptive weight adjustment module builds an adaptive weight adjustment mechanism based on the division results, and performs different weight adjustment measures for the three situations of phase consistency, mild mismatch and severe mismatch; The adjustment mechanism feedback evaluation module obtains the adjustment mechanism feedback information in real time during the process of the adaptive weight adjustment mechanism adjusting the weight, and analyzes it after obtaining it to evaluate whether the adjustment degree of the adaptive weight adjustment mechanism can meet expectations, and optimizes the adaptive weight adjustment mechanism according to the evaluation results; The flight process monitoring and optimization module monitors and analyzes the altitude prediction data and adjustment records during the entire flight process in real time, and continuously optimizes the adaptive weight adjustment mechanism based on the monitoring results to ensure the altitude prediction stability and control accuracy of the aircraft in different flight phases and environmental conditions.

2. The multi-sensor height prediction system integrating LSTM and Kalman filtering according to claim 1 is characterized in that: The time deviation index of the generated forecast and height forecast error synchronicity index Construct a phase mismatch evaluation model and generate a phase mismatch coefficient by weighted summation , and the generated phase mismatch coefficient The phase mismatch coefficient threshold interval is set A comparison is made, and the phase mismatch degree between the prediction results of the LSTM model and the Kalman filter is predicted based on the comparison results. The prediction results are divided into three categories: phase consistency, mild mismatch, and severe mismatch. The specific comparison analysis and classification are as follows: like , if the phase mismatch between the prediction results of the LSTM model and the Kalman filter is low, the situation is classified as phase consistent; like , the phase mismatch between the prediction results of the LSTM model and the Kalman filter is moderate, then the situation is classified as mild mismatch; like , if the phase mismatch between the prediction results of the LSTM model and the Kalman filter is high, the situation is classified as severe mismatch.

3. The multi-sensor height prediction system integrating LSTM and Kalman filtering according to claim 2 is characterized in that: In the adaptive weight adjustment module, an adaptive weight adjustment mechanism is constructed according to the division results. Specifically, different weight adjustment parameters are set according to the division results of phase consistency, mild mismatch and severe mismatch, which correspond to the weight allocation of LSTM model and Kalman filter in prediction fusion, forming an adaptive weight adjustment mechanism. The adjustment mechanism automatically determines the weight allocation strategy and adjustment range between LSTM model and Kalman filter based on the range of phase mismatch coefficient through pre-set rules. Different weight adjustment measures are taken for the three situations of phase consistency, mild mismatch and severe mismatch. Specifically, in the case of phase consistency, a balanced weight strategy is adopted to assign equal weights to the LSTM model and the Kalman filter to maintain stable prediction; in the case of mild mismatch, the weight of the Kalman filter is adjusted to be higher than that of the LSTM model to increase the influence of short-term correction in the prediction, so that the system can maintain a balance between long-term and short-term predictions while enhancing its responsiveness to current environmental changes; in the case of severe mismatch, the weight of the Kalman filter is significantly increased, and the weight of the LSTM model is significantly reduced to ensure that the system can respond quickly to sudden changes and enhance the short-term stability of highly predicted results.

4. The multi-sensor height prediction system integrating LSTM and Kalman filtering according to claim 3 is characterized in that: In the adjustment mechanism feedback evaluation module, after the adjustment mechanism feedback information is obtained in real time, it is preprocessed, and the instantaneous prediction adjustment feedback information and the long-term adjustment cumulative feedback information in the preprocessed adjustment mechanism feedback information are extracted, and analyzed after extraction to generate an immediate adjustment deviation index and a cumulative adjustment effect coefficient respectively; an adjustment degree evaluation model is constructed for the generated immediate adjustment deviation index and cumulative adjustment effect coefficient to generate an adjustment coefficient, and the generated adjustment coefficient is compared with a pre-set adjustment coefficient threshold, and whether the adjustment degree of the adaptive weight adjustment mechanism can meet expectations is evaluated based on the comparison results, and the adaptive weight adjustment mechanism is optimized based on the evaluation results.

5. The multi-sensor height prediction system integrating LSTM and Kalman filtering according to claim 4 is characterized in that: The acquisition logic of the instant adjustment deviation index and the cumulative adjustment effect coefficient is as follows: The instantaneous prediction adjustment feedback information in the preprocessed adjustment mechanism feedback information is extracted, specifically including the difference between the predicted altitude of the LSTM model and Kalman filter combination and the actual altitude of the aircraft at different times within a period of time during the process of the adaptive weight adjustment mechanism adjusting the weight, the weight difference between the LSTM model and the Kalman filter, and the time interval of the system responding to the change of the aircraft altitude after the weight adjustment, and calibrated as , and , Indicates that during the period of time when the adaptive weight adjustment mechanism adjusts the weight The difference between the predicted altitude of the LSTM model and Kalman filter combination and the actual altitude of the aircraft at that moment, Indicates that during the period of time when the adaptive weight adjustment mechanism adjusts the weight The weight difference between the LSTM model and the Kalman filter at this moment, It represents the time interval for the system to respond to the change of aircraft altitude after weight adjustment. , is a positive integer; Calculate the instant adjustment deviation index. The specific calculation formula is as follows: ； In the formula, To adjust the deviation index instantly; The long-term adjustment cumulative feedback information in the preprocessed adjustment mechanism feedback information is extracted, specifically including the cumulative prediction error change after multiple weight adjustments within the time window in which the adaptive weight adjustment mechanism adjusts the weight, the cumulative amplitude of all weight adjustments within the time window, and the time length within the time window, and they are marked as , and , It represents the cumulative change in prediction error after multiple weight adjustments within the time window in which the adaptive weight adjustment mechanism adjusts the weight. Represents the cumulative magnitude of all weight adjustments within the time window, Indicates the length of time within the time window; Calculate the cumulative adjustment effect coefficient. The specific calculation formula is as follows: ； In the formula, It is the cumulative adjustment effect coefficient.

6. The multi-sensor height prediction system integrating LSTM and Kalman filtering according to claim 5 is characterized in that: Instantly adjust the deviation index for the generated and the cumulative adjusted effect coefficient Construct an adjustment degree assessment model and generate adjustment coefficients through weighted summation , and the resulting adjustment factor With the preset adjustment factor threshold Compare and evaluate whether the adjustment degree of the adaptive weight adjustment mechanism can meet expectations based on the comparison results, and optimize the adaptive weight adjustment mechanism based on the evaluation results. The specific comparison and analysis are as follows: like , the adjustment degree of the adaptive weight adjustment mechanism can meet expectations, and there is no need to optimize the adaptive weight adjustment mechanism; like , the adjustment degree of the adaptive weight adjustment mechanism cannot meet expectations, and the adaptive weight adjustment mechanism needs to be optimized, including: increasing the weight adjustment frequency of the LSTM model and the Kalman filter to improve the system's response speed to environmental changes; resetting the upper limit of the weight adjustment range to ensure the short-term response effect during the adjustment process; optimizing the weight adjustment parameters, dynamically adjusting the weight ratio of the LSTM model and the Kalman filter to adapt to the current flight status requirements; combining historical data and real-time feedback information, analyzing the prediction error and response time, automatically adjusting the system threshold setting, and improving the system's adaptability and accuracy.

Images