Civil aviation passenger demand prediction method and system based on artificial intelligence

Abstract

The invention discloses a civil aviation passenger demand prediction method and system based on artificial intelligence. The method comprises the steps of data acquisition, abnormal demand elimination, civil aviation demand side sample supplementation, civil aviation passenger demand prediction model establishment and civil aviation passenger demand prediction. The invention belongs to the field of civil aviation demand prediction, and particularly relates to a civil aviation passenger demand prediction method and system based on artificial intelligence. According to the scheme, a time sequence weighted distance is introduced, and it is guaranteed that samples with off-season and low demand characteristics are not mistakenly removed; on the basis of route similarity weighting, trusted neighbors are generated, and rare low-demand samples are supplemented in a targeted manner; feature importance scores are introduced, feature sensitivity scores of civil aviation samples are calculated, and influences of feature-time points on demand prediction are evaluated from different angles; the adaptability to festival and holiday peaks and extreme weather scenes is guided through a punishment mechanism; based on the introduction of a holiday association weight enhancement scheduling strategy, the penalty weight growth rate is dynamically improved; and the reliability of civil aviation passenger demand prediction is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of civil aviation demand forecasting, and specifically refers to a civil aviation passenger demand forecasting method and system based on artificial intelligence. Background Art

[0002] The civil aviation passenger demand forecasting method analyzes historical travel data, passenger behavior characteristics, real-time flight information, etc., combines big data, and forecasts the future passenger flow trend to optimize flight arrangements and improve service matching. However, the general civil aviation passenger demand forecasting method is sensitive to abnormal events, easily interfered by extreme values, prone to misidentifying small normal fluctuations before and after seasons as noise, mixing route data with huge differences together, generating pseudo-civil aviation samples that do not have real significance, and thus leading to poor accuracy of demand forecasting; the general civil aviation passenger demand forecasting method has weak forecasting ability for extreme scenarios such as holiday peaks and abnormal weather, and cannot capture this periodic concentrated surge scenario, thus leading to poor reliability of demand forecasting. Summary of the Invention

[0003] In view of the above situation, to overcome the defects of the prior art, the present invention provides a civil aviation passenger demand forecasting method and system based on artificial intelligence. Aiming at the problem that the general civil aviation passenger demand forecasting method is sensitive to abnormal events, easily interfered by extreme values, prone to misidentifying small normal fluctuations before and after seasons as noise, mixing route data with huge differences together, generating pseudo-civil aviation samples that do not have real significance, and thus leading to poor accuracy of demand forecasting, this solution introduces a temporal weighted distance to reduce the probability of excluding points with similar time and normal fluctuations, retain the real fluctuations before and after holidays, and ensure that civil aviation samples with low-season and low-demand characteristics are not mis-excluded; based on weighted route similarity, generate credible neighbors, and specifically supplement rare low-demand civil aviation samples; thereby improving the accuracy of civil aviation passenger demand forecasting; aiming at the problem that the general civil aviation passenger demand forecasting method has weak forecasting ability for extreme scenarios such as holiday peaks and abnormal weather, and cannot capture this periodic concentrated surge scenario, thus leading to poor reliability of demand forecasting, this solution introduces a feature importance score, calculates the feature sensitivity score of civil aviation samples, and evaluates the impact of feature-time points on demand forecasting from different angles in combination with the interaction value; guides the adaptability to holiday peaks and extreme weather scenarios through a penalty mechanism; based on introducing a holiday correlation weight to enhance the scheduling strategy, dynamically increase the growth rate of the penalty weight, and adjust the penalty coefficient according to the holiday distance and the error ratio of the validation set; thereby improving the reliability of civil aviation passenger demand forecasting.

[0004] The technical solution adopted by the present invention is as follows: The civil aviation passenger demand forecasting method based on artificial intelligence provided by the present invention includes the following steps:

[0005] Step S1: Data collection;

[0006] Step S2: Abnormal demand elimination;

[0007] Step S3: Civil aviation demand-side sampling supplement;

[0008] Step S4: Establish a civil aviation passenger demand prediction model;

[0009] Step S5: Civil aviation passenger demand prediction.

[0010] Furthermore, in step S1, the data collection is to collect historical civil aviation passenger demand data; label the demand level as a tag; and perform standardization processing on the collected data to obtain an initial civil aviation sample set.

[0011] Furthermore, in step S2, the abnormal demand elimination is to use fuzzy c-means clustering for all low-demand and high-demand civil aviation samples, eliminate abnormal points, and introduce a time series weighted distance to calculate the civil aviation sample anomaly value , and the formula used is: ; ; where is the distance between the civil aviation sample and the cluster center ; and are the time points of the civil aviation sample and the cluster center respectively; is the time decay factor; F is the total feature dimension, f is the dimension index, and i is the civil aviation sample index; and are respectively and the values of the f-th dimension; K is the total number of clusters, and j is the cluster index; for the civil aviation samples with are eliminated; is the noise threshold.

[0012] Furthermore, in step S3, the civil aviation demand-side sampling supplement is to introduce a route similarity weight to perform civil aviation sample sampling supplement, and perform fuzzy c-means clustering again on the low-demand and high-demand civil aviation samples after abnormal demand elimination; calculate the route feature similarity , expressed as: ; within each clustering cluster, based on the route feature similarity, select neighbors from the remaining low-demand and high-demand civil aviation samples to generate new civil aviation samples , and the formula used is: ; where and are the original civil aviation samples; is the generation coefficient; to obtain the final civil aviation sample set.

[0013] Further, in step S4, the establishment of the civil aviation passenger demand prediction model specifically includes the following:

[0014] Step S41: Model architecture design; based on LSTM, the final civil aviation sample set is split into three types of inputs: static features, historical time series features, and future exogenous indicator sequences; the original feature vector is output, ready for subsequent encoding; the static feature encoder receives the static features and outputs a fixed-length static context vector; the encoder consists of multiple layers of LSTM that receive the historical time series vectors concatenated with the static context vector at each step, with a length of T, and outputs a hidden state sequence; the Attention mechanism receives the hidden state sequence and the previous hidden state of the decoder, and outputs a context vector; the decoder consists of multiple layers of LSTM that receive the previous prediction value, static context, current context, and future exogenous indicator, and outputs the decoder hidden state; the output layer receives the decoder hidden state and outputs the predicted passenger flow result at the k-th step; the basic loss uses the mean squared error loss function;

[0015] Step S42: The loss introduces feature importance scoring to evaluate the impact of feature-time points on prediction from different perspectives, and calculates the feature sensitivity score of civil aviation samples , using a dynamically adjusted integration step size, calculates the feature sensitivity score according to the change of features, and takes the integration step size as a dynamic variable and adjusts it according to the change rate of features, ; the feature sensitivity score is expressed as: ; where, is the actual feature value of the k-th dimension of the i-th civil aviation sample at time point t; is the feature change rate of the k-th dimension of the i-th civil aviation sample at time point t; is the feature mean; is the value of the k-th dimension at time point t; is the predicted passenger flow result; x is the input feature vector of the civil aviation sample, is the mean of the feature vectors; is the integration variable; calculates the interaction impact value of civil aviation samples , used to capture the discrete interaction effects of holidays and weather, and is expressed as: ; where, F is the total set of all feature-time points; S is the subset obtained by removing the feature-time point (k,t) from the total set F; is the predicted passenger flow result that receives the subset as input;

[0016] Step S43: Two-stage training strategy; in the first stage, it is trained with pure mean squared error for rounds, minimizing the mean squared error loss, and calculating and normalizing for each batch, expressed as: ; where ImSe is the normalized feature importance score; batch is the number of civil aviation samples in each training batch; G is the normalization constant; and are weight hyperparameters; in the second stage, importance-oriented penalty guidance is used for retraining rounds, minimizing ; where L is the total loss function; is the dynamic penalty coefficient; is the mean squared error loss function;

[0017] Step S44: Holiday-related weight enhancement scheduling; introduce the holiday-related weight enhancement scheduling strategy, and define the distance to the nearest major holiday , expressed as: ; construct the holiday enhancement factor , expressed as: ; where h is the date of the holiday closest to the average sampling date of the civil aviation samples in the current batch; H is the set of all holiday dates; is the enhancement amplitude; is the smoothing scale; calculate the validation set error ratio on the key window within 7 days before and after the holiday in the current training round , expressed as: ; construct the error-driven factor , expressed as: ; the final weight scheduling is expressed as: ; where is the validation set loss on the window within 7 days before and after the holiday in the current epoch, g is the current training number; is the average validation set loss within the window; is the previous training validation set error ratio; is the error amplification coefficient; and are the minimum penalty coefficient and the maximum penalty coefficient respectively.

[0018] Furthermore, in step S5, the civil aviation passenger demand prediction is to collect civil aviation passenger demand data in real time and input it into the civil aviation passenger demand prediction model, and use the demand level output by the civil aviation passenger demand prediction model as the civil aviation passenger demand prediction result.

[0019] The civil aviation passenger demand prediction system based on artificial intelligence provided by the present invention includes a data collection module, an abnormal demand elimination module, a civil aviation demand side resampling module, a civil aviation passenger demand prediction model establishment module, and a civil aviation passenger demand prediction module;

[0020] The data collection module collects historical civil aviation passenger demand data and constructs an initial civil aviation sample set;

[0021] The abnormal demand elimination module eliminates the abnormal demand records in the initial civil aviation sample set;

[0022] The civil aviation demand side sampling supplement module selects neighbors based on route feature similarity for the civil aviation samples after eliminating abnormal demands to generate new civil aviation samples, obtaining the final civil aviation sample set;

[0023] The civil aviation passenger demand prediction model establishment module, based on the final civil aviation sample set, uses LSTM as the basis, introduces feature importance scoring, and combines holiday correlation weight enhanced scheduling strategies to construct a civil aviation passenger demand prediction model;

[0024] The civil aviation passenger demand prediction module predicts the civil aviation passenger demand based on the real-time collected civil aviation passenger demand data using the civil aviation passenger demand prediction model.

[0025] The beneficial effects achieved by the present invention using the above solution are as follows:

[0026] (1) Aiming at the problems existing in general civil aviation passenger demand prediction methods, such as being sensitive to abnormal events, being easily interfered by extreme values, easily misinterpreting small normal fluctuations before and after seasons as noise, mixing route data with huge differences together, and the generated pseudo-civil aviation samples not having real significance, thus resulting in poor demand prediction accuracy. This solution introduces a time series weighted distance to reduce the elimination probability of points with similar time and normal fluctuations, retain the real fluctuations before and after holidays, and ensure that civil aviation samples with low demand characteristics in the off-season are not miseliminated; based on route similarity weighting, generate credible neighbors, and specifically supplement scarce low-demand civil aviation samples; thereby improving the accuracy of civil aviation passenger demand prediction.

[0027] (2) Aiming at the problems existing in general civil aviation passenger demand prediction methods, such as weak prediction ability for extreme scenarios such as holiday peaks and abnormal weather, being unable to capture such periodic concentrated surges, and thus resulting in poor demand prediction reliability. This solution introduces feature importance scoring, calculates the feature sensitivity score of civil aviation samples, and combines the interaction influence value to evaluate the influence of features - time points on demand prediction from different angles; through a penalty mechanism, guide the adaptability to holiday peaks and extreme weather scenarios; based on the introduction of holiday correlation weight enhanced scheduling strategies, dynamically increase the growth rate of the penalty weight, and adjust the penalty coefficient according to the holiday distance and the error ratio of the validation set; thereby improving the reliability of civil aviation passenger demand prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic flow chart of the civil aviation passenger demand prediction method based on artificial intelligence provided by the present invention;

[0029] Figure 2 It is a schematic diagram of the civil aviation passenger demand prediction system based on artificial intelligence provided by the present invention;

[0030] Figure 3 It is a flow diagram of step S4.

[0031] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, and do not constitute a limitation to the present invention. Specific embodiments

[0032] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0033] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating the orientation or position relationship are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the system or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention.

[0034] Embodiment 1, refer to Figure 1 , the civil aviation passenger demand prediction method based on artificial intelligence provided by the present invention includes the following steps:

[0035] Step S1: Data collection; collect historical civil aviation passenger demand data and construct an initial civil aviation sample set;

[0036] Step S2: Abnormal demand elimination; eliminate the abnormal demand records in the initial civil aviation sample set;

[0037] Step S3: Civil aviation demand side sampling; for the civil aviation samples after eliminating abnormal demands, select neighbors based on route feature similarity to generate new civil aviation samples, and obtain the final civil aviation sample set;

[0038] Step S4: Establish a civil aviation passenger demand prediction model; based on the final civil aviation sample set, with LSTM as the basis, introduce feature importance scoring, and combine holiday correlation weight enhancement scheduling strategies to construct a civil aviation passenger demand prediction model;

[0039] Step S5: Civil aviation passenger demand prediction; based on the civil aviation passenger demand prediction model, conduct civil aviation passenger demand prediction on the real-time collected civil aviation passenger demand data.

[0040] Embodiment 2, refer to Figure 1, this embodiment is based on the above embodiment. In step S1, data collection is to collect historical civil aviation passenger demand data; the historical civil aviation passenger demand data includes route passenger flow data, external influence data, and timestamps; the route passenger flow data includes historical seat occupancy rate, passenger capacity, ticket sales volume, flight segments, departure and arrival airports, flight distance, aircraft type, and fare information; the external influence data includes weather, holiday data, promotion events, and flight schedule changes; the demand level is marked as a label; the demand level includes low demand, normal demand, and high demand; for each dimension of the historical civil aviation passenger demand data, an index weight is set, and the value range is (0, 1), and the index weights are combined to be 1; the grading thresholds for low demand and high demand are set, and the value range is; the sum of each dimension of the historical civil aviation passenger demand data multiplied by the index weight is used to obtain the data demand evaluation value; if the data demand evaluation value is lower than the grading threshold for low demand, it is marked as a low demand civil aviation sample; if the data demand evaluation value is not lower than the grading threshold for low demand but lower than the grading threshold for high demand, it is marked as a normal demand civil aviation sample; if the data demand evaluation value is not lower than the grading threshold for high demand, it is marked as a high demand civil aviation sample; the collected data is standardized to obtain an initial civil aviation sample set.

[0041] Embodiment Three, refer to Figure 1 , this embodiment is based on the above embodiment. In step S2, abnormal demand elimination is that there are records of abnormally low or suddenly increased demands in the initial civil aviation sample set due to temporary promotions, flight cancellations, or system failures, which will distort the demand prediction judgment. Therefore, for all low demand civil aviation samples and high demand civil aviation samples, fuzzy c-means clustering is used to eliminate the abnormal points closest to the high demand area, and a time series weighted distance is introduced to reduce the elimination probability for normal fluctuations with close time, ensuring smoothness around seasons and holidays, and calculating the abnormal values of civil aviation samples , the formula used is: ; ; where, is the civil aviation sample and the cluster center 's distance; and are respectively the civil aviation sample and the cluster center 's time points; is the time decay factor, and the value range is [0.01, 0.1]; F is the total feature dimension, f is the dimension index, and i is the civil aviation sample index; and are respectively and the values of the f-th dimension; K is the total number of clusters, and the value range is [0.01, 0.1], and j is the cluster index; for the civil aviation samples are eliminated; is the noise threshold, with a value range of [3, 10]; only retain civil aviation samples with stable off-season and low-demand characteristics, without participating in oversampling, to avoid copying boundary noise into new civil aviation samples.

[0042] Example 4, refer to Figure 1 , based on the above example, in step S3, for civil aviation demand-side sampling, introduce the route similarity weight to perform civil aviation sample sampling, balance the quantity distribution of low-demand civil aviation samples and mainstream high-demand civil aviation samples, and avoid excessive interpolation to pseudo-neighbors with large differences from the target route. Therefore, perform fuzzy c-means clustering again on the low-demand civil aviation samples and high-demand civil aviation samples after abnormal demand elimination; calculate the route feature similarity , expressed as: ; within each clustering cluster, select neighbors based on the route feature similarity for the remaining low-demand civil aviation samples and high-demand civil aviation samples to generate new civil aviation samples , and the formula used is: ; where and are the original civil aviation samples; is the generation coefficient, with a value range of [0.1, 1.0]; obtain the final civil aviation sample set.

[0043] By performing the above operations, for the general civil aviation passenger demand prediction method, there are problems such as being sensitive to abnormal events, being easily interfered by extreme values, easily misidentifying small normal fluctuations before and after seasons as noise, mixing route data with huge differences together, and the generated pseudo-civil aviation samples not having real significance, which in turn leads to poor demand prediction accuracy. This solution introduces the time series weighted distance to reduce the probability of eliminating points with similar time and normal fluctuations, retain the real fluctuations before and after holidays, and ensure that civil aviation samples with off-season and low-demand characteristics are not miseliminated; based on the weighted route similarity, generate credible neighbors, and specifically supplement scarce low-demand civil aviation samples; thereby improving the accuracy of civil aviation passenger demand prediction.

[0044] Example 5, refer to Figure 1 and Figure 3 , based on the above example, in step S4, establishing a civil aviation passenger demand prediction model specifically includes the following content:

[0045] Step S41: Model architecture design; Based on LSTM, the final civil aviation sample set is split into three types of inputs: static features, historical time-series features, and future exogenous indicator sequences; The original feature vector is output and prepared for subsequent encoding; The static feature encoder receives the static features and outputs a fixed-length static context vector, which is used to map categorical and continuous static information to the same dimension and provide the global route background; The encoder consists of multiple LSTMs that receive the historical time-series vectors concatenated with the static context vector at each step, with a length of T, ranging from [7, 30], and outputs a hidden state sequence, which is used to capture the historical time-series dependencies and form a compressed representation of the past T steps; The Attention mechanism receives the hidden state sequence and the hidden state of the previous step of the decoder, and outputs a context vector, which focuses on the most important historical moments in the encoder for the current prediction at each decoding step; The decoder consists of multiple LSTMs that receive the previous prediction value, static context, current context, and future exogenous indicators, and outputs the decoder hidden state, which is used to combine historical information and context to gradually generate the internal representation of each future moment; The output layer receives the decoder hidden state and outputs the passenger flow prediction result at the k-th step; The basic loss uses the mean squared error loss function;

[0046] Step S42: The loss introduces feature importance scores to evaluate the impact of feature-time points on the prediction from different perspectives, and calculates the feature sensitivity scores of civil aviation samples , which is used to evaluate the continuous features of ticket prices and load factors. However, the impact of different features on the prediction at different time points may be dynamically changing. Therefore, using a dynamically adjusted integration step size can calculate the feature sensitivity scores more flexibly according to the changes of features. The integration step size is used as a dynamic variable and adjusted according to the change rate of the feature, ; The feature sensitivity score is expressed as: ; where, is the actual feature value of the k-th dimension of the i-th civil aviation sample at time point t; is the feature change rate of the k-th dimension of the i-th civil aviation sample at time point t; is the feature mean; is the value of the k-th dimension at time point t; is the passenger flow prediction result; x is the input feature vector of the civil aviation sample, is the feature vector mean; is the integration variable; Calculate the interaction impact value of civil aviation samples , which is used to capture the discrete interaction effects of holidays and weather, and is expressed as: ; where, F is the total set of all feature-time points; S is the subset obtained by removing the feature-time point (k, t) from the total set F; is the passenger flow prediction result that receives the subset as input;

[0047] Step S43: Two-stage training strategy; in the first stage, use pure mean squared error to let the model first learn the temporal and feature associations robustly, prevent the penalty from being interfered by noise, and train for rounds, minimize the mean squared error loss, calculate and normalize for each batch, expressed as: ; where ImSe is the normalized feature importance score; batch is the number of civil aviation samples in each training batch; G is the normalization constant, and its value range is [1, 50]; and are weight hyperparameters, and their value range is [0.1, 1.0]; in the second stage, use importance-guided penalty to boost the prediction accuracy for holiday peaks, extreme weather scenarios, and then train for rounds, minimize ; where L is the total loss function; is the dynamic penalty coefficient; is the mean squared error loss function;

[0048] Step S44: Holiday-related weight enhancement scheduling; the passenger flow on routes often shows obvious peaks before and after major holidays. Relying solely on linear or cosine scheduling cannot capture such periodic concentrated surge scenarios. Therefore, a holiday-related weight enhancement scheduling strategy is introduced to dynamically increase the growth rate of the penalty weight and guide the model to pay more attention to the key features during that period earlier and more strongly; define the distance to the nearest major holiday , expressed as: ; construct the holiday enhancement factor , expressed as: ; where h is the date of the holiday closest to the average sampling date of the civil aviation samples in the current batch, and the date is normalized; H is the set of all holiday dates; is the enhancement amplitude, and its value range is [0.5, 2.0]; is the smoothing scale, and its value range is [0.01, 0.2]; when the holiday is approaching, approaches 0, reaches the maximum; when the distance is far, approaches 1; if the validation error on the key window within 7 days before and after the holiday or during the bad weather window is still high, it means that the model has insufficient learning for these extreme scenarios and additional penalty amplification is required; otherwise, it can be appropriately slowed down; therefore, calculate the validation set error ratio on the key window within 7 days before and after the holiday for the current training round, expressed as: ; construct the error-driven factor , expressed as: ; the final weight scheduling is expressed as: ; where, is the validation set loss on the window of 7 days before and after holidays in the current epoch, and g is the current training times; is the average validation set loss within the window; is the error ratio of the previous training validation set; is the error amplification factor, and its value range is [0.1, 2.0]; and are the minimum penalty coefficient and the maximum penalty coefficient respectively, and their value ranges are [0.1, 1.0] and [1.0, 10.0] respectively.

[0049] By performing the above operations, aiming at the problem that the general civil aviation passenger demand prediction method has weak prediction ability for extreme scenarios such as holiday peaks and abnormal weather, and cannot capture this periodic concentrated sudden increase scenario, resulting in poor reliability of demand prediction, this solution introduces feature importance scoring, calculates the feature sensitivity score of civil aviation samples, and combines the interaction influence value to evaluate the influence of features - time points on demand prediction from different perspectives; guides the adaptability to holiday peaks and extreme weather scenarios through a penalty mechanism; enhances the scheduling strategy based on introducing holiday correlation weights, dynamically increases the growth rate of penalty weights, and adjusts the penalty coefficient according to the holiday distance and the validation set error ratio; thereby improving the reliability of civil aviation passenger demand prediction.

[0050] Example Six, refer to Figure 1 , based on the above example, in step S5, the civil aviation passenger demand prediction is to collect civil aviation passenger demand data in real time and input it into the civil aviation passenger demand prediction model, and use the demand level output by the civil aviation passenger demand prediction model as the civil aviation passenger demand prediction result.

[0051] Example Seven, refer to Figure 2 , based on the above example, the civil aviation passenger demand prediction system based on artificial intelligence provided by the present invention includes a data collection module, an abnormal demand elimination module, a civil aviation demand side sample supplement module, a civil aviation passenger demand prediction model establishment module, and a civil aviation passenger demand prediction module;

[0052] The data collection module collects historical civil aviation passenger demand data and constructs an initial civil aviation sample set;

[0053] The abnormal demand elimination module eliminates abnormal demand records in the initial civil aviation sample set;

[0054] The civil aviation demand side sample supplement module selects neighbors based on route feature similarity for the civil aviation samples after eliminating abnormal demands to generate new civil aviation samples, and obtains the final civil aviation sample set;

[0055] The civil aviation passenger demand prediction model establishment module is based on the final civil aviation sample set, with LSTM as the basis, introduces feature importance scores, and combines holiday correlation weights to enhance the scheduling strategy to construct a civil aviation passenger demand prediction model;

[0056] The civil aviation passenger demand prediction module predicts the civil aviation passenger demand based on the civil aviation passenger demand prediction model for the real-time collected civil aviation passenger demand data.

[0057] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0058] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention.

[0059] The above describes the present invention and its embodiments, and this description is not restrictive. What is shown in the drawings is only one of the embodiments of the present invention, and the actual structure is not limited thereto. All in all, if those of ordinary skill in the art are inspired by it and design similar structural ways and embodiments without creative efforts without departing from the purpose of the present invention, they should all fall within the protection scope of the present invention.

Claims

1. An artificial intelligence-based civil aviation passenger demand prediction method, characterized in that: The method includes the following steps: Step S1: Data collection; collect historical civil aviation passenger demand data and construct an initial civil aviation sample set; Step S2: Abnormal demand elimination; eliminate the abnormal demand records in the initial civil aviation sample set; Step S3: Civil aviation demand side sampling supplement; for the civil aviation samples after eliminating abnormal demands, select neighbors based on route feature similarity to generate new civil aviation samples and obtain the final civil aviation sample set; Step S4: Establish a civil aviation passenger demand prediction model; based on the final civil aviation sample set, with LSTM as the basis, introduce feature importance scoring, and combine holiday correlation weight enhanced scheduling strategy to construct a civil aviation passenger demand prediction model; Step S5: Civil aviation passenger demand prediction; based on the civil aviation passenger demand prediction model, conduct civil aviation passenger demand prediction on the real-time collected civil aviation passenger demand data.

2. The method for predicting civil aviation passenger demand based on artificial intelligence according to claim 1, wherein: In step S2, for all low-demand civil aviation samples and high-demand civil aviation samples, fuzzy c-means clustering is used to eliminate outliers, and a time-series weighted distance is introduced to calculate the outlier values of civil aviation samples. , and the formula used is: ; ; where is the civil aviation sample and the cluster center distance; and are respectively the civil aviation sample and the cluster center time points; is the time decay factor; F is the total feature dimension, f is the dimension index, and i is the civil aviation sample index; and are respectively and the values of the f-th dimension; K is the total number of clusters, and j is the cluster index; for civil aviation samples are eliminated; is the noise threshold.

3. The method for predicting civil aviation passenger demand based on artificial intelligence according to claim 2, characterized in that: In step S3, the civil aviation demand-side sample supplementation is to introduce the route similarity weight, perform civil aviation sample supplementation, and perform fuzzy c-means clustering on the low-demand civil aviation samples and high-demand civil aviation samples after abnormal demand elimination again; calculate the route feature similarity , which is expressed as: ; within each clustering cluster, select neighbors based on the route feature similarity for the remaining low-demand civil aviation samples and high-demand civil aviation samples to generate new civil aviation samples , and the formula used is: ; where and are the original civil aviation samples; is the generation coefficient; the final civil aviation sample set is obtained.

4. The method for predicting civil aviation passenger demand based on artificial intelligence according to claim 1, wherein: In step S4, the establishment of the civil aviation passenger demand prediction model specifically includes the following content: Step S41: Model architecture design; with LSTM as the basis, split the final civil aviation sample set into three types of inputs: static features, historical time series features, and future exogenous indicator sequences; output the original feature vector for subsequent encoding; the static feature encoder receives the static features and outputs a fixed-length static context vector; the encoder consists of multiple layers of LSTM that receive the historical time series vectors concatenated with the static context vector at each step, with a length of T, and outputs a hidden state sequence; the Attention mechanism receives the hidden state sequence and the previous hidden state of the decoder and outputs a context vector; The decoder consists of multiple layers of LSTM that receive the previous prediction value, static context, current context, and future exogenous indicator, and outputs the decoder hidden state; the output layer receives the decoder hidden state and outputs the predicted passenger flow result at the kth step; the basic loss uses the mean square error loss function; Step S42: Loss introduction feature importance scoring, evaluating the impact of feature-time point on prediction from different perspectives, and calculating the feature sensitivity score of civil aviation samples , using dynamic adjustment of the integration step size, calculating the feature sensitivity score according to the change of features, and taking the integration step size as a dynamic variable, adjusting it according to the change rate of features, ; The feature sensitivity score is expressed as: ; where is the actual feature value of the k-th dimension of the i-th civil aviation sample at time point t; is the feature change rate of the k-th dimension of the i-th civil aviation sample at time point t; is the feature mean; is the value of the k-th dimension at time point t; is the passenger flow prediction result; x is the input feature vector of the civil aviation sample, is the feature vector mean; is the integration variable; calculating the interaction effect value of the civil aviation sample , used to capture the discrete interaction effects of holidays and weather, expressed as: ; where F is the total set of all feature-time points; S is the subset obtained by removing the feature-time point (k,t) from the total set F; is the passenger flow prediction result that receives the subset as input; Step S43: Two-stage training strategy; Step S44: Holiday correlation weight enhanced scheduling.

5. The method for predicting civil aviation passenger demand based on artificial intelligence according to claim 4, characterized in that: In step S43, the two-stage training strategy is to use pure mean squared error for the first stage of training for a number of rounds to minimize the mean squared error loss, and calculate and normalize for each batch, expressed as: ; where ImSe is the normalized feature importance score; batch is the number of civil aviation samples in each training batch; G is the normalization constant; and are weight hyperparameters; for the second stage, use importance-guided penalty to guide and retrain for a number of rounds to minimize ; where L is the total loss function; is the dynamic penalty coefficient; is the mean squared error loss function.

6. The method for predicting civil aviation passenger demand based on artificial intelligence according to claim 5, wherein: In step S44, the holiday-associated weight enhanced scheduling introduces a holiday-associated weight enhanced scheduling strategy and defines the distance to the nearest major holiday , which is expressed as: ; constructs a holiday enhancement factor , which is expressed as: ; where h is the date of the holiday closest to the average sampling date of the current batch of civil aviation samples ; H is the set of all holiday dates; is the enhancement amplitude; is the smoothing scale; calculates the validation set error ratio on the key window within 7 days before and after the holiday for the current training round , which is expressed as: ; constructs an error-driven factor , which is expressed as: ; the final weight scheduling is expressed as: ; where is the validation set loss on the window within 7 days before and after the holiday for the current epoch, and g is the current training number; is the average validation set loss within the window; is the previous training validation set error ratio; is the error amplification coefficient; and are the minimum penalty coefficient and the maximum penalty coefficient, respectively.

7. The method for predicting civil aviation passenger demands based on artificial intelligence according to claim 6, wherein: In step S1, the data collection is to collect historical civil aviation passenger demand data; label the demand level as a label; standardize the collected data to obtain the initial civil aviation sample set.

8. The method for predicting civil aviation passenger demand based on artificial intelligence according to claim 7, wherein: In step S5, the civil aviation passenger demand prediction is to collect real-time civil aviation passenger demand data and input it into the civil aviation passenger demand prediction model, and use the demand level output by the civil aviation passenger demand prediction model as the civil aviation passenger demand prediction result.

9. An artificial intelligence-based civil aviation passenger demand prediction system for implementing the artificial intelligence-based civil aviation passenger demand prediction method according to any one of claims 1-8, characterized in that: It includes a data collection module, an abnormal demand elimination module, a civil aviation demand side sampling supplement module, a civil aviation passenger demand prediction model establishment module, and a civil aviation passenger demand prediction module; The data collection module collects historical civil aviation passenger demand data and constructs an initial civil aviation sample set; The abnormal demand elimination module eliminates the abnormal demand records in the initial civil aviation sample set; The civil aviation demand side sampling supplement module selects neighbors based on route feature similarity for the civil aviation samples after eliminating abnormal demands to generate new civil aviation samples and obtain the final civil aviation sample set; The civil aviation passenger demand prediction model establishment module, based on the final civil aviation sample set, with LSTM as the basis, introduces feature importance scoring, and combines holiday correlation weight enhanced scheduling strategy to construct a civil aviation passenger demand prediction model; The civil aviation passenger demand prediction module performs civil aviation passenger demand prediction on the real-time collected civil aviation passenger demand data based on the civil aviation passenger demand prediction model.

Images