Flight plan prediction method based on Prophet and LSTM model fused holiday and seasonal factors

Abstract

The invention discloses a flight plan prediction method based on Prophet and LSTM model fusion holiday and seasonal factors, and the method comprises the steps: collecting historical air traffic flow data, and carrying out the preprocessing of the data, and obtaining a historical data time sequence; constructing a Prophet model, predicting the air traffic flow in a future time period according to the historical data time sequence, and constructing prediction results into a prediction result sequence; and constructing a Seq2Seq model to carry out prediction again to obtain a final air traffic flow prediction result. The prediction result sequence is input into a Seq2Seq model, and joint training is carried out; and carrying out joint prediction on actual data by using the model subjected to joint training, and carrying out visual display on a result. According to the method, seasonal and holiday factors, long-term trend information, historical data and the influence of other characteristics are fused in prediction, and accurate and comprehensive prediction of flight plans in spring and autumn in the future is achieved.

Description

Technical Field

[0001] The present invention relates to a flight schedule prediction method, in particular to a flight schedule prediction method based on Prophet and LSTM models integrating holiday and seasonal factors. Background Art

[0002] This section merely provides background information related to the present disclosure and is not necessarily prior art.

[0003] Spring and autumn are the peak seasons for aviation activities, and airlines have a significant increase in flight schedules during these two seasons. Spring and autumn include many holidays, and air traffic increases significantly during these holidays. In addition, weather conditions in spring and autumn also have an important impact on flight schedules. Existing forecasting methods have difficulty accurately capturing these seasonal characteristics and sudden changes.

[0004] Flight schedule forecasting is a complex task that is affected by many factors, including but not limited to seasonal changes, holidays, weather, air traffic control, etc. At present, although various forecasting methods have been proposed, there are still some challenges in the existing technology. For example, traditional time series methods may not be able to fully capture complex seasonal and holiday effects, and machine learning-based methods, although they can better handle nonlinear relationships, often require a large amount of data and complex models to achieve satisfactory forecasting performance.

[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0006] Purpose of the invention: The technical problem to be solved by the present invention is to provide a flight schedule prediction method based on Prophet and LSTM models that integrates holiday and seasonal factors in order to address the deficiencies in the prior art.

[0007] In order to solve the above technical problems, the present invention discloses a flight schedule prediction method based on Prophet and LSTM model integrating holiday and seasonal factors, the method comprising the following steps:

[0008] Step 1: collect historical air traffic flow data and preprocess it to obtain a historical data time series;

[0009] Step 2: construct a Prophet model, and predict the air traffic flow in the future time period based on the historical data time series, and construct the prediction results into a prediction result sequence;

[0010] Step 3: construct a Seq2Seq model to perform prediction again based on the prediction result sequence obtained in step 2 to obtain the final air traffic flow prediction result;

[0011] Step 4: Input the prediction result sequence in step 2 into the Seq2Seq model constructed in step 3 for joint training;

[0012] Step 5: Use the jointly trained Prophet model and Seq2Seq model to jointly predict the actual data and visualize the results of the joint prediction, ultimately achieving flight schedule prediction based on the Prophet and LSTM models that integrate holiday and seasonal factors.

[0013] Furthermore, the collection of historical air traffic flow data in step 1 includes: counting daily or hourly air traffic flow.

[0014] Furthermore, the preprocessing described in step 1 includes: processing missing values, abnormal values ​​and duplicate values, and formatting the historical air traffic flow data into a time series to obtain a historical data time series.

[0015] Furthermore, the Prophet model described in step 2 is a time series prediction model based on the additive trend model and the decomposition method.

[0016] Furthermore, the prediction of air traffic flow in the future time period described in step 2 specifically includes:

[0017] Step 2-1, input the historical data time series into the Prophet model;

[0018] Step 2-2, use the Prophet model to decompose the key factors based on the historical data time series;

[0019] Step 2-3, make predictions based on key factors to obtain future traffic forecasts that include point estimates and prediction intervals.

[0020] Furthermore, the decomposition described in step 2-2 yields key factors, including:

[0021] The Prophet model is used to decompose the feature y(t) that affects the time series into four components. The process is expressed as follows:

[0022] y(t)=g(t)+s(t)+h(t)+ε(t)

[0023] Among them, g(t) represents the trend component, which is used to model the long-term change trend of the historical data time series; s(t) represents the seasonal component, which is used to model the periodic change of the historical data time series; h(t) represents the holiday effect component, which is used to model the impact of specific holidays on the historical data time series; ε(t) represents the residual term, which is used to capture the random part; among them, the trend component g(t), the seasonal component s(t) and the holiday effect component h(t) are the key factors.

[0024] Further, in step 2-2,

[0025] The trend component g(t) is a piecewise linear function or a nonlinear regression function, wherein the nonlinear regression function is expressed as:

[0026]

[0027] Wherein, C(t) represents the capacity that changes with time t, and the capacity represents the maximum number of flights that the airport can handle, or the maximum demand for flights in the market; k is the growth rate of the dependent variable; m is the correction parameter, which represents the time offset;

[0028] The piecewise linear function is expressed as:

[0029] g(t)=[k+α(t) R δ]t+[m+α(t) T γ]

[0030] Among them, the change point s j and the growth rate δ at the change point j Composition vector And δ~Laplace(0,τ), γ=-s j δ j , represents the offset correction term; by γ j Composition vector γ, a j (t) forms a(t);

[0031] The change point s j It is the specific time point at which the trend in the time series changes suddenly, that is, the event that affects the flight flow, where the change point sequence number j = 1, 2, ..., h, h represents the number of change points;

[0032] Compare the prediction effects of piecewise linear function or nonlinear regression function by cross-validation method, and choose to use piecewise linear function or nonlinear regression function to calculate trend component g(t) according to the prediction effect;

[0033] The seasonal component s(t) is expressed as:

[0034]

[0035] Where s(t) is the standard discrete Fourier series; P is the period parameter; N is the number of parameters, and 2N parameters form a set β = [a 1 , b 1 , a 2 , b 2 , …, a N , b N ] T , and β~Normal(0,σ 2 ), σ represents the significance of seasonal effect;

[0036] The holiday effect component h(t) is expressed as:

[0037]

[0038] Among them, let the date be D i ,parameter Indicates D i The influence range of surrounding dates is v represents the standard deviation of the normal distribution; the indicator function Z(t) is used to identify whether time t belongs to the date set D of holiday l i , expressed in L holidays as:

[0039] Z(t)=[l(t∈D 1 ),…,l(t∈D L )].

[0040] Furthermore, the step 2 of constructing the prediction results into a prediction result sequence includes:

[0041] The prediction results of the Prophet model, that is, the future traffic forecast values, are combined into a time series to obtain a prediction result sequence.

[0042] Furthermore, the construction of the Seq2Seq model described in step 3 includes:

[0043] Input layer, encoder, decoder and output layer, where

[0044] The input layer is used to input the prediction result sequence obtained in step 2;

[0045] The encoder encodes the input prediction result sequence into a vector representation of a fixed length;

[0046] The decoder uses a long short-term memory network model LSTM as a decoder. The decoder generates an output sequence time by time step according to the input vector representation. At each time step, the decoder generates a prediction value and uses its own hidden state to guide the prediction of the next time step to obtain the future traffic prediction value;

[0047] The output layer outputs the decoder prediction result as the final prediction result.

[0048] Further, the joint training described in step 4 includes:

[0049] Step 4-1, use the prediction result sequence of the Prophet model in step 2 as the input of the encoder in the Seq2Seq model;

[0050] Step 4-2, defining a loss function as an optimization target for joint training, wherein the loss function includes: mean square error MSE or logarithmic loss;

[0051] Step 4-3, by adjusting the parameters in the Prophet model and the Seq2Seq model, calculating the loss value according to the loss function, optimizing the parameters through iterative training, and obtaining the jointly trained Prophet model and Seq2Seq model.

[0052] Beneficial effects:

[0053] By combining the Prophet and LSTM models, this method integrates Prophet's precise grasp of seasonality and long-term trends and LSTM's sensitivity to short-term fluctuations, providing more comprehensive and accurate flight schedule forecasts for spring and autumn, better coping with the complexity of flight traffic in spring and autumn, and ensuring the reliability and comprehensiveness of forecasts on different time scales. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments, and the above and / or other advantages of the present invention will become more clear.

[0055] Figure 1 It is a schematic diagram of an overall process according to an embodiment of the present invention.

[0056] Figure 2 is a diagram of a Seq2seq network structure according to an embodiment of the present invention.

[0057] Figure 3 1 is a diagram of an LSTM network structure according to an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The overall idea of ​​the present invention is as follows:

[0059] Prophet and Long Short-Term Memory (LSTM) are two advanced models widely used in time series forecasting. The Prophet model was developed by Facebook and is specifically designed to process time series data with seasonal and holiday factors. It can accurately capture seasonality and long-term trends in flight schedule forecasting, making it excellent in processing cyclical changes in data. The LSTM model is a deep learning model suitable for sequence data. It is good at processing short-term dynamic changes and nonlinear characteristics, and is more sensitive to the prediction of emergencies and irregular fluctuations. It has achieved remarkable results in the field of time series forecasting.

[0060] The fusion method based on Prophet and LSTM (Long Short-Term Memory) model is a potential solution, which combines the good capture ability of Prophet model for seasonal and holiday effects, and the modeling ability of LSTM model for complex nonlinear relationships. The present invention aims to propose a flight schedule prediction method based on Prophet and LSTM model fusion of holiday and seasonal factors, so as to overcome the limitations of the existing technology and improve the prediction accuracy and reliability of flight schedule prediction.

[0061] The present invention proposes a method combining the Prophet model and the LSTM model for flight schedule forecasting in spring and autumn. The method combines the Prophet model's prediction of seasonality and long-term trends with the LSTM model's processing of short-term fluctuations and holiday impacts, covering the characteristics and changes on different time scales in spring and autumn, thereby more effectively coping with the complexity of flight traffic in spring and autumn and improving the comprehensiveness and accuracy of forecasting.

[0062] In order to achieve the above object, the technical solution adopted by the present invention is:

[0063] (1) Data collection and preprocessing: The historical flight schedules and actual air traffic flow data in spring and autumn are collected, including daily or hourly traffic statistics, with a focus on data during holidays such as the Spring Festival, Qingming Festival, Labor Day, Mid-Autumn Festival, and National Day. Missing values, outliers, and duplicate values ​​are processed to ensure data quality.

[0064] (2) Prophet model construction: Use the Prophet model to predict flight schedules in spring and autumn, capturing information on seasonality, holiday factors, and long-term trends, especially flight changes during the Spring Festival, Qingming Festival, Labor Day, Mid-Autumn Festival, and National Day.

[0065] (3) Sequence construction and encoder input: The prediction results of the Prophet model for spring and autumn are constructed into a sequence as the input of the Seq2Seq (sequence to sequence) model encoder, combining the seasonal characteristics of spring and autumn and holiday factors.

[0066] (4) LSTM decoder design: Design and construct LSTM as the decoder of the Seq2Seq model, combining historical data of spring and autumn, seasonal characteristics and holiday factors to predict future flight schedules.

[0067] (5) Model training and prediction: The Prophet prediction output is used as the encoder input to train the Seq2Seq model to predict the flight schedules in spring and autumn. The loss function is defined for optimization, taking into account the seasonal fluctuations and holiday effects in spring and autumn.

[0068] (6) Information fusion and prediction output: The LSTM model outputs prediction results for future air traffic flow by integrating the seasonality, holiday factors, and long-term trend information provided by Prophet, combined with historical data and other features.

[0069] Example:

[0070] In order to explain the specific technical solutions of the present invention in detail, the following preferred embodiments are given and described in conjunction with the accompanying drawings.

[0071] like Figure 1 FIG. 1 is an overall flow chart of an embodiment of the present invention, wherein the flow chart includes the following steps:

[0072] Step S1 is data collection and preprocessing. The data collection and preprocessing is to collect historical air traffic flow data, including daily or hourly traffic statistics, covering data for multiple periods. Process missing values, outliers and duplicate values ​​to ensure data quality, and then format the time series. Obtain historical air traffic flow data and holiday and seasonal data to prepare the features required for Prophet model prediction.

[0073] After data collection and preprocessing in step S1, a time series of historical air traffic flow data is obtained. This time series will include daily or hourly traffic statistics, and missing values, outliers, and duplicate values ​​have been processed to ensure data quality. In addition, the time series will be formatted to accommodate time series analysis, such as alignment to specific time intervals (e.g., hourly or daily) and taking into account seasonal and holiday factors.

[0074] Step S2 is the construction and prediction of the Prophet model. A Prophet model is constructed, which is specifically used to predict air traffic flow in future time periods. The Prophet model can effectively capture and simulate seasonal patterns, holiday effects, and long-term trends in the data through its unique time series decomposition method, thereby providing accurate prediction results. The Prophet model was proposed by Facebook data scientists Sean J. Taylor and Benjamin Letham in 2017. The goal of Prophet is to make time series prediction easier and more accessible, especially for time series data with seasonal and holiday effects. It provides a simple and powerful method that can be used to predict time series data in various fields. Prophet is a time series prediction model based on the additive trend model and the decomposition method. The Prophet model decomposes the feature y(t) that affects the time series into 4 components. Through this decomposition, the Prophet model can effectively identify and simulate complex patterns in time series data. In a specific embodiment, the Prophet model receives the time series data preprocessed in step S1 as input, and initializes and adjusts the model parameters according to the data. The model will then automatically fit trends and seasonality, taking into account the impact of holidays, to generate forecasts for air traffic flow in future time periods. Specifically, these factors are represented in the model as:

[0075] y(t)=g(t)+s(t)+h(t)+ε(t)

[0076] Where g(t) represents the trend component, which is used to model the long-term trend of changes in the time series. The trend component reflects the long-term development direction of the time series. In flight traffic, this includes long-term changes due to factors such as economic growth, tourism development, or improved transportation infrastructure. By identifying the trend component, the general growth or decline direction of future traffic can be predicted; s(t) represents the seasonal component, which is used to model the periodic changes in the time series. The seasonal component captures the periodic patterns in the data, such as the peak traffic in a specific season or holiday each year. Understanding seasonal patterns helps predict traffic changes in a specific period in the future, such as predicting an increase in passenger traffic during the peak tourist season; h(t) represents the holiday effect, which is used to model the impact of specific holidays on the time series. The holiday effect takes into account the short-term impact of specific holidays on flight traffic. By identifying the impact of holidays on traffic, these special events can be taken into account in the forecast, thereby providing more accurate traffic forecasts during the holiday period; ε(t) represents the residual term, which captures the random part that cannot be explained by the model. The residual term represents the part of the model that is not explained by the trend, seasonality, and holiday effects. This may include random fluctuations or other factors not captured by the model. While residual terms are often viewed as errors, they are also an important component of the forecasting process because they provide information about the uncertainty in the model's predictions.

[0077] The g(t) trend term can be expressed as two functions; one is a piecewise linear function, which is a straight line with different slopes in different time periods, but is linear within each segment. This function is suitable for situations where the trend changes relatively smoothly and has clear changes at certain points in time. For example, if the flight traffic grows at a relatively stable rate most of the time, but the growth rate changes after certain specific events (such as the opening of new routes), the piecewise linear function can effectively capture this change; the other is a nonlinear regression function, which allows the trend term to change in a more complex way and can capture situations where the growth rate accelerates or slows down over time. This function is suitable for situations where the trend change does not follow a linear pattern but presents a certain curve shape. For example, if the growth rate of flight traffic accelerates over time, the nonlinear regression function can fit this accelerated growth trend more flexibly. The nonlinear regression function is suitable for capturing complex changes in trend terms, such as acceleration or slowing of growth rates, and is suitable for situations where the trend changes in a curve shape. The piecewise linear function is suitable for data with stable trend changes or changes at specific points because it can clearly capture the trend change points. When selecting a function, data characteristics, model complexity, prediction objectives, and model diagnostics should be considered. Although the nonlinear regression function can simulate dynamic changes more accurately, it has a high computational complexity and may increase the computational burden. Therefore, when resources are limited or the model needs to be run quickly, the piecewise linear function may be more practical. The prediction target will also affect the choice. When roughly capturing the direction of the long-term trend, the piecewise linear function is usually sufficient, while a nonlinear regression function is required to finely simulate the trend dynamics. Model diagnostic tools such as residual analysis can be used to evaluate the function fitting effect and select the best fitting function. In actual operation, the prediction effects of the two functions can be compared through methods such as cross-validation to determine the best model in a specific application scenario, ensuring that the selected function can effectively capture data trend changes and provide accurate predictions. Among them, the nonlinear regression function is expressed as:

[0078]

[0079] Where: C(t) represents the capacity that changes at any time, which represents the limit value that the time series may reach when it stabilizes in the long term. In the context of flight traffic forecasting, capacity refers to the maximum number of flights that an airport can handle, or the maximum market demand for flights. For trend forecasting, understanding capacity helps to determine the upper limit of long-term growth, thereby revealing potential limitations to business growth and future market potential. In the process of changing over time, g(t) will approach a relatively fixed capacity value related to the C(t) function; k is the growth rate of the dependent variable; m is a correction parameter, which represents the time offset.

[0080] If a piecewise linear function is used, the function only needs to modify k and m. It can be expressed as:

[0081] g(t)=[k+α(t) R δ]t+[m+α(t) T γ]

[0082] Change point s j (j = 1, 2, ..., h) and the growth rate δ at the change point j Composition vector And δ~Laplace(0,τ), γ=-s j δ j , represents the offset correction term. j Composition vector γ, a j (t) is composed of a(t). A change point refers to a specific time point in a time series when a trend suddenly changes. This change may be caused by some external events or internal changes. In flight traffic, a change point corresponds to the opening of new routes, major holidays, emergencies (such as epidemics), or other events that affect flight traffic.

[0083] The seasonal term s(t) is expressed as:

[0084]

[0085] Where s(t) is the standard discrete Fourier series; P is the period parameter, which is 365.25 if it is in years and 7 if it is in weeks; N is the number of parameters, and if there are 2N parameters, the set β = [a 1 , b 1 , a 2 , b 2 , …, a N , b N ] R , and β~Normal(0,σ 2 );σ represents the degree of seasonal effect. If the unit is year, N=10; if the unit is week, N=3.

[0086] Holidays have a great impact on traffic. Let the date be D i ,parameter Indicates the impact range on surrounding dates, and the holiday term h(t) is expressed as:

[0087]

[0088] set up v represents the standard deviation of the normal distribution, quantifying the parameter The larger the discreteness of v, the greater the impact of the holiday, that is, the impact of the holiday on the data is not limited to the holiday day, but may also extend to a period of time before and after the holiday. Z(t) represents an indicator function used to identify whether time t belongs to a certain holiday date set D i , expressed in L holidays as:

[0089] Z(t)=[l(t∈D 1 ),…,l(t∈D L )]

[0090] In step S2, the present invention predicts future air traffic flow through the Prophet model, which analyzes historical data to identify seasonal patterns, holiday effects, and long-term trends. First, daily or hourly traffic data covering multiple periods are collected and preprocessed, and then a Prophet model is constructed and input into these data to decompose key factors such as trends, seasonality, and holiday effects. The trend analysis of the model reveals the long-term development direction of traffic, the seasonal analysis simulates the impact of seasonal changes through periodic functions, and the holiday effect analysis quantifies the short-term impact of specific holidays. Finally, the Prophet model generates future traffic forecasts containing point estimates and prediction intervals. These forecast results will be used as inputs to the Seq2Seq model to further refine the forecast and provide a basis for subsequent steps.

[0091] Step S3 is sequence construction and encoder input. The prediction results of the Prophet model are constructed as a sequence, and each time step represents the traffic prediction value at a future time point, which serves as the input sequence of the Seq2Seq model encoder. The Seq2Seq (Sequence-to-Sequence) model is a neural network model for processing sequence-to-sequence problems. It consists of two main parts, namely the encoder and the decoder. The encoder encodes the input sequence into a fixed-length vector representation, and then the decoder decodes this vector representation into an output sequence.

[0092] Step S4 is LSTM decoder design. Design and build LSTM as the decoder of the Seq2Seq model, combining historical data and other features. This step can be divided into two parts, data preparation and LSTM model design. First, prepare the time series data output by the Prophet model, and then input the data into the LSTM decoder model with encoder input to use the prediction provided by the Prophet model as the initial input and combine other features to make future traffic predictions.

[0093] Step S5 is model training and prediction. In this step, a Seq2Seq network model is trained, which uses the prediction results generated by the Prophet model in step S2 as input data. The Seq2Seq model consists of an encoder and a decoder, wherein the encoder receives the output sequence of the Prophet model and encodes it into a vector representation of a fixed length; the decoder generates a flight flow forecast for the future time step based on this vector representation. During the training process, a loss function is defined as the target of model optimization, such as mean squared error (MSE), logarithmic loss, or other evaluation indicators. Through iterative training, the weights and biases in the model are adjusted, and the loss value of each Epoch is calculated according to the defined loss function to minimize the difference between the predicted value and the actual value. In addition, the present invention also takes into account the seasonal fluctuations and holiday effects in spring and autumn to ensure that the model can accurately capture the impact of these factors on flight flow.

[0094] Step S6 is information fusion and prediction output. The LSTM model outputs the prediction results of future air traffic flow by integrating the seasonality, holiday factors and long-term trend information provided by Prophet, combining historical data and other features.

[0095] This method combines the advantages of Prophet and LSTM models to improve the comprehensiveness and accuracy of air traffic flow forecasting.

[0096] like Figure 2 The figure shows the structure of the Seq2seq network. The Seq2Seq network model further refines the flight traffic forecast provided by the Prophet model through its sequence-to-sequence capability, captures more complex time dependencies and short-term dynamics, integrates multi-source information such as weather and event data, improves forecast accuracy, and can dynamically adjust to quickly respond to new data, thereby optimizing the forecast results. The network consists of four parts: input, encoder, decoder, and output. Among them:

[0097] The input is the prediction result of the Prophet model based on historical data, which is input as a sequence to the Seq2Seq model.

[0098] Next, the Encoder part encodes the input sequence and converts it into a fixed-length vector representation. This vector representation contains the key information of the input sequence.

[0099] The Decoder part then receives the vector representation generated by the encoder and generates the output sequence time-step by time-step. At each time-step, the decoder generates a prediction value and uses its own hidden state to guide the prediction of the next time-step. The Decoder model generates the output sequence by referring to the encoded information of the input sequence.

[0100] Finally, the output part outputs the sequence generated by the decoder as the final prediction result.

[0101] like Figure 3 The figure shows the LSTM network architecture. LSTM (Long Short-Term Memory) is a commonly used recurrent neural network (RNN) model for modeling and forecasting sequence data. Compared with the traditional RNN structure, LSTM introduces a gating mechanism that can better capture and process long-term dependencies. The application of LSTM network structure in the field of flight traffic forecasting uses its unique gating mechanism to process time series data, effectively capture long-term dependencies, and adapt to complex data patterns caused by various factors such as seasonal changes, holidays, and weather conditions. This network structure can improve forecasting accuracy by learning complex patterns and relationships in historical data and dynamically adjusting to quickly respond to new data such as emergencies or market changes. In addition, LSTM can be integrated with other models such as Prophet to form a more accurate forecasting system, and can integrate multi-source information such as historical traffic data, weather reports, and economic indicators to provide comprehensive and in-depth insights for airlines and decision makers.

[0102] In the LSTM model, there are three key gating units: input gate, forget gate, and output gate. These gating units allow the model to selectively remember and forget information in the input sequence by learning and controlling the flow of information.

[0103] Specifically, the LSTM model works as follows:

[0104] Input gate: The input gate controls the effect of new input data on the internal cell state. It uses a sigmoid function to decide which information should be saved.

[0105] Forget gate: The forget gate determines which information in the cell state at the previous moment should be forgotten. It determines the information to be retained through the sigmoid function and element-by-element multiplication.

[0106] Cell state: The cell state in the LSTM model is the core component of the LSTM network. The cell state is responsible for remembering and transmitting long-term dependencies in the sequence.

[0107] Output gate: The output gate determines the output value generated from the current cell state. It inputs the cell state into the sigmoid function and the tanh function and performs element-by-element multiplication to obtain the output value.

[0108] Through the collaborative work of these gating units, the LSTM model can better handle the dependencies between long sequences, and thus perform well in tasks such as sequence modeling, natural language processing, and time series prediction.

[0109] In a specific implementation, the present application provides a computer storage medium and a corresponding data processing unit, wherein the computer storage medium can store a computer program, and when the computer program is executed by the data processing unit, the invention content of a flight schedule prediction method based on Prophet and LSTM models integrating holiday and seasonal factors provided by the present invention and some or all steps in each embodiment. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0110] Those skilled in the art can clearly understand that the technical solutions in the embodiments of the present invention can be implemented by means of computer programs and their corresponding general hardware platforms. Based on such an understanding, the technical solutions in the embodiments of the present invention can be essentially or partly contributed to the prior art in the form of computer programs, i.e., software products, which can be stored in a storage medium and include several instructions for enabling a device including a data processing unit (which can be a personal computer, a server, a single-chip microcomputer, an MCU or a network device, etc.) to execute the methods described in various embodiments of the present invention or certain parts of the embodiments.

[0111] The present invention provides a method and idea for a flight schedule prediction method based on Prophet and LSTM models integrating holiday and seasonal factors. There are many methods and approaches to implement the technical solution. The above is only a preferred implementation of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention. All components not specified in this embodiment can be implemented using existing technologies.

Claims

1. A flight schedule prediction method based on Prophet and LSTM model integrating holiday and seasonal factors, characterized in that: The method comprises the following steps: Step 1: collect historical air traffic flow data and preprocess it to obtain a historical data time series; Step 2: construct a Prophet model, and predict the air traffic flow in the future time period based on the historical data time series, and construct the prediction results into a prediction result sequence; Step 3: construct a Seq2Seq model to perform prediction again based on the prediction result sequence obtained in step 2 to obtain the final air traffic flow prediction result; Step 4: Input the prediction result sequence in step 2 into the Seq2Seq model constructed in step 3 for joint training; Step 5: Use the jointly trained Prophet model and Seq2Seq model to jointly predict the actual data and visualize the results of the joint prediction, ultimately achieving flight schedule prediction based on the Prophet and LSTM models that integrate holiday and seasonal factors.

2. A flight schedule prediction method based on Prophet and LSTM model integrating holiday and seasonal factors according to claim 1, characterized in that: The collection of historical air traffic flow data described in step 1 includes: statistics on daily or hourly air traffic flow.

3. A flight schedule prediction method based on Prophet and LSTM model integrating holiday and seasonal factors according to claim 2, characterized in that: The preprocessing described in step 1 includes: processing missing values, abnormal values ​​and duplicate values, and formatting the historical air traffic flow data into a time series to obtain a historical data time series.

4. A flight schedule prediction method based on Prophet and LSTM model integrating holiday and seasonal factors according to claim 3, characterized in that: The Prophet model described in step 2 is a time series prediction model based on the additive trend model and the decomposition method.

5. A flight schedule prediction method based on Prophet and LSTM model integrating holiday and seasonal factors according to claim 4, characterized in that: The air traffic flow forecast for the future time period described in step 2 specifically includes: Step 2-1, input the historical data time series into the Prophet model; Step 2-2, use the Prophet model to decompose the key factors based on the historical data time series; Step 2-3, make predictions based on key factors to obtain future traffic forecasts that include point estimates and prediction intervals.

6. A flight schedule prediction method based on Prophet and LSTM model integrating holiday and seasonal factors according to claim 5, characterized in that: The decomposition described in step 2-2 yields key factors, including: The Prophet model is used to decompose the feature y(t) that affects the time series into four components. The process is expressed as follows: y(t)=g(t)+s(t)+h(t)+ε(t) Among them, g(t) represents the trend component, which is used to model the long-term change trend of the historical data time series; s(t) represents the seasonal component, which is used to model the periodic change of the historical data time series; h(t) represents the holiday effect component, which is used to model the impact of specific holidays on the historical data time series; ε(t) represents the residual term, which is used to capture the random part; among them, the trend component g(t), the seasonal component s(t) and the holiday effect component h(t) are the key factors.

7. A flight schedule prediction method based on Prophet and LSTM model integrating holiday and seasonal factors according to claim 6, characterized in that: In step 2-2, The trend component g(t) is a piecewise linear function or a nonlinear regression function, wherein the nonlinear regression function is expressed as: Wherein, C(t) represents the capacity that changes with time t, and the capacity represents the maximum number of flights that the airport can handle, or the maximum demand for flights in the market; k is the growth rate of the dependent variable; m is the correction parameter, which represents the time offset; The piecewise linear function is expressed as: g(t)=[k+α(t) T δ]t+[m+α(t) T c] Among them, the change point s j and the growth rate δ at the change point j Composition vector And δ~Laplace(0,τ),γ=-s j δ j , represents the offset correction term; by γ j Composition vector γ, a j (t) forms a(t); The change point s j It is a specific time point in the time series where the trend changes suddenly, that is, the event that affects the flight flow, where the change point number j = 1, 2, ..., h, h represents the number of change points; Compare the prediction effects of piecewise linear function or nonlinear regression function by cross-validation method, and choose to use piecewise linear function or nonlinear regression function to calculate trend component g(t) according to the prediction effect; The seasonal component s(t) is expressed as: Where s(t) is the standard discrete Fourier series; P is the period parameter; N is the number of parameters, and 2N parameters form the set β = [a1, b1, a2, b2, …, a N ,b N ] T , and β~Normal(0,σ 2 ), σ represents the significance of seasonal effect; The holiday effect component h(t) is expressed as: Among them, let the date be D i ,parameter Indicates D i The influence range of surrounding dates is ν represents the standard deviation of the normal distribution; the indicator function Z(t) is used to identify whether time t belongs to the date set D of holiday l i , in L holidays it is expressed as: Z(t)=[l(t∈D1),…,l(t∈D L )]。 8. A flight schedule prediction method based on Prophet and LSTM model integrating holiday and seasonal factors according to claim 7, characterized in that: The prediction results described in step 2 are constructed as a prediction result sequence, including: The prediction results of the Prophet model, that is, the future traffic forecast values, are combined into a time series to obtain a prediction result sequence.

9. A flight schedule prediction method based on Prophet and LSTM model integrating holiday and seasonal factors according to claim 8, characterized in that: The construction of the Seq2Seq model described in step 3 includes: Input layer, encoder, decoder and output layer, where The input layer is used to input the prediction result sequence obtained in step 2; The encoder encodes the input prediction result sequence into a vector representation of a fixed length; The decoder uses a long short-term memory network model LSTM as a decoder. The decoder generates an output sequence time by time step according to the input vector representation. At each time step, the decoder generates a prediction value and uses its own hidden state to guide the prediction of the next time step to obtain the future traffic prediction value; The output layer outputs the decoder prediction result as the final prediction result.

10. A flight schedule prediction method based on Prophet and LSTM model integrating holiday and seasonal factors according to claim 9, characterized in that: Perform joint training as described in step 4, including: Step 4-1, use the prediction result sequence of the Prophet model in step 2 as the input of the encoder in the Seq2Seq model; Step 4-2, defining a loss function as an optimization target for joint training, wherein the loss function includes: mean square error MSE or logarithmic loss; Step 4-3, by adjusting the parameters in the Prophet model and the Seq2Seq model, calculating the loss value according to the loss function, optimizing the parameters through iterative training, and obtaining the jointly trained Prophet model and Seq2Seq model.

Images