Data processing method for predicting occurrence time of flight turnover key event

Abstract

The invention provides a data processing method for predicting flight turnover key event occurrence time. The method comprises the steps of obtaining a to-be-processed data set needing to be processed currently; the to-be-processed data set comprises at least one event associated information; inputting the to-be-processed data set into a plurality of target occurrence time prediction models to obtain a plurality of event occurrence time prediction results; performing weighted fusion on the plurality of event occurrence time prediction results to obtain a predicted value of event occurrence time corresponding to each event associated information in the to-be-processed data set; and based on a preset reference database, adjusting the predicted value of the event occurrence time corresponding to each event associated information to obtain a target predicted value of the event occurrence time corresponding to each event associated information. According to the invention, the predicted event occurrence time can be as accurate as possible.

Description

Technical Field

[0001] The present invention relates to the field of computer technology applications, and in particular, to a data processing method for predicting the occurrence time of key events in flight turnover. Background Art

[0002] In some application scenarios, it is necessary to predict the occurrence time of fixed events executed by a target object, so as to formulate corresponding coping strategies according to the predicted occurrence time. For example, it is necessary to accurately predict the occurrence time of each link in many links such as flight takeoff and landing, ground service, and resource allocation involved in airport flight operation guarantee services, so as to improve airport operation efficiency and passenger experience.

[0003] Patent document CN112836905B discloses an event occurrence time prediction method. This patent document trains a deep learning neural network through the historical data of flight basic information and time series message information to obtain an event occurrence time prediction model. Although this document can improve the accuracy of event occurrence time prediction compared with a rule-based expert system, since this document uses a single model to predict the event occurrence time, the accuracy of event occurrence time prediction is still not accurate enough. Summary of the Invention

[0004] In view of the above technical problems, the technical solution adopted by the present invention is as follows: An embodiment of the present invention provides a data processing method for predicting the occurrence time of key events in flight turnover. The occurrence time of key events in flight turnover is the occurrence time of key events in flight turnover experienced by an event-associated object from arriving at a specified geographical area to leaving the specified geographical area. The method includes the following steps: S100, obtaining a dataset to be processed that needs to be processed currently; the dataset to be processed includes at least one event-associated information, and the event-associated information includes event-associated object basic information, weather data, and the number of users of the event-associated object.

[0005] S200, inputting the dataset to be processed into a plurality of target occurrence time prediction models to obtain a plurality of event occurrence time prediction results.

[0006] S300, performing weighted fusion on the plurality of event occurrence time prediction results to obtain a predicted value of the occurrence time of each event-associated information in the dataset to be processed.

[0007] The S400 adjusts the predicted value of the event occurrence time corresponding to each event association information based on a preset reference database to obtain the target predicted value of the event occurrence time corresponding to each event association information. Multiple reference data are stored in the preset reference database, and each reference data includes the corresponding event association information and the true time of the event occurrence time.

[0008] The present invention has at least the following beneficial effects: A data processing method for predicting the occurrence time of key events in flight turnaround provided by an embodiment of the present invention. The event occurrence time is obtained based on the fusion weighted result of the prediction results of multiple target occurrence time prediction models. Each target occurrence time prediction model is used to predict the corresponding event occurrence time, and the influence of different factor combinations on different event occurrence times is considered during the training process, enabling the model to learn the complex relationships between events. Moreover, historical data is used to correct the prediction results of the model, making the predicted event occurrence time more accurate than the results predicted by existing event occurrence time prediction methods.

[0009] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0011] Figure 1 It is a flowchart of the data processing method for predicting the occurrence time of key events in flight turnaround provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0012] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the 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 of 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 fall within the scope of protection of the present invention.

[0013] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which this invention belongs. The terms used in the description of the invention herein are for the purpose of describing specific embodiments only and are not intended to limit the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0014] It should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts depict the steps as sequential processes, many of the steps can be implemented in parallel, concurrently, or simultaneously. In addition, the order of the steps can be rearranged. The process can be terminated when its operations are completed, but it can also have additional steps not included in the figures. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0015] An embodiment of the present invention provides a data processing method for predicting the occurrence time of key events in flight turnaround. The occurrence time of key events in flight turnaround is the occurrence time of the key events in flight turnaround experienced by the event-associated object from arriving at the designated geographical area to leaving the designated geographical area.

[0016] In a specific application scenario, the event-associated object can be a flight, the designated geographical area can be an airport, and the key events in flight turnaround experienced by the event-associated object from arriving at the designated geographical area to leaving the designated geographical area can be flight operation support events. Flight operation support events refer to the events that occur in the flight operation support process and can include events such as wheel chock time, cabin door opening time, start of passenger disembarkation time, completion of passenger disembarkation time, start of cleaning time, completion of cleaning time, start of refueling time, completion of refueling time, start of catering time, end of catering time, start of boarding time, end of boarding time, cabin door closing time, removal of wheel chocks time, estimated departure time, etc.

[0017] Further, in the embodiment of the present invention, the data processing method for predicting the occurrence time of key events in flight turnaround provided by the embodiment of the present invention may include the steps as Figure 1 shown: S100, obtaining a dataset to be processed that needs to be processed currently; the dataset to be processed includes at least one event association information, and the event association information includes event-associated object basic information, weather data, and the number of users of the event-associated object.

[0018] In the embodiment of the present invention, when the event-associated object is a flight, the event-associated object basic information may include aircraft number, aircraft type, route, planned departure date, operation status, planned parking bay, and actual arrival time. The operation status may include: normal, early, and delayed statuses.

[0019] In the embodiments of the present invention, the weather data in the event association information may be the weather data within a set time period after the event association object is scheduled to arrive at a specified geographical area. In one illustrative embodiment, the duration of the set time period may be 3 hours. The weather data includes at least parameters such as temperature, wind speed, humidity, and weather characteristics. Among them, the weather characteristics may include light rain, moderate rain, thunderstorm, heavy rain, heavy fog, sandstorm, etc.

[0020] In the embodiments of the present invention, the number of users of the event association object refers to the number of users who use the event association object. In the case where the event association object is a flight, the user may be a passenger.

[0021] S200, input the dataset to be processed into multiple target event occurrence time prediction models to obtain multiple event occurrence time prediction results.

[0022] S300, perform weighted fusion on the multiple event occurrence time prediction results to obtain the predicted value of the event occurrence time corresponding to each event association information in the dataset to be processed; In the embodiments of the present invention, the predicted value t of the v-th event occurrence time corresponding to each event association information v pre satisfies the following condition: t v pre = w1 × t 1v pre + w2 × t 2v pre + …… + w z × t zv pre + …… + w Q × t Qv pre , where v takes values from 1 to Q, Q is the number of event occurrence times to be predicted, that is, the number of critical event occurrence times of flight turnover to be predicted, w z is the weight of the z-th target event occurrence time prediction model, and t zv pre is the predicted value of the v-th event occurrence time predicted by the z-th target event occurrence time prediction model, and z takes values from 1 to Q.

[0023] Furthermore, w z satisfies the following condition: w z = AC z / (AC1 + AC2 + …… + AC z + …… + AC Q ), where AC z is the prediction accuracy rate of the z-th target event occurrence time prediction model for the corresponding event occurrence time.

[0024] S400 adjusts the predicted value of the event occurrence time corresponding to each event association information based on a preset reference database to obtain the target predicted value of the event occurrence time corresponding to each event association information.

[0025] Further, S400 specifically includes: S401, for each event association information, obtains the similarity between the event association information and the event association information corresponding to each reference data in the preset reference database, and obtains the similarity set corresponding to the event association information; wherein, multiple reference data are stored in the preset reference database, and each reference data includes the corresponding event association information and the true time of the event occurrence time.

[0026] In the embodiment of the present invention, the similarity can be the cosine similarity.

[0027] S402, uses the reference data corresponding to the similarity greater than or equal to the set similarity threshold in the similarity set corresponding to each event association information as the associated data of the event association information.

[0028] In the embodiment of the present invention, the set similarity threshold can be an empirical value, for example, 0.9.

[0029] Those skilled in the art know that if there is no similarity greater than or equal to the set similarity threshold in the similarity set corresponding to a certain event association information, then the event association information has no associated data, which also means that there is no need to correct the event occurrence time corresponding to the event association information.

[0030] S403, based on the true time of the event occurrence time corresponding to the associated data of each event association information, adjusts the predicted value of the event occurrence time corresponding to each event association information to obtain the target predicted value of the event occurrence time corresponding to each event association information.

[0031] Specifically, the target predicted value t of the v-th event occurrence time corresponding to each event association information v satisfies the following conditions: t v =t v pre +SUM(w e ×△t ev ) e=1……H .

[0032] Wherein, w e is the weight of the e-th associated data corresponding to the event association information, w eIt is positively correlated with the similarity between the event-associated information and the event-associated information in the corresponding e-th associated data, and specifically can be equal to the similarity between the event-associated information and the event-associated information in the corresponding e-th associated data, where e ranges from 1 to H, and H is the number of associated data of the event-associated information. △t ev is t v pre the difference between and the true time of the corresponding associated data, specifically t v pre and the difference between the event-associated information and the true time of the occurrence time of the v-th event in the corresponding e-th associated data.

[0033] Furthermore, in the embodiments of the present invention, the multiple target occurrence time prediction models can be obtained through the following steps: S10. Based on the basic data set, train multiple pre-constructed initial time prediction models to obtain initial time prediction models for predicting the occurrence time of each event.

[0034] S20. Adjust the hyperparameter combinations of the initial time prediction models for the occurrence time of each event, obtain the optimal hyperparameter combinations of the initial time prediction models for the occurrence time of each event, and obtain intermediate time prediction models for the occurrence time of each event.

[0035] S30. Based on the initial input features and the intermediate occurrence time prediction models for the occurrence time of each event, obtain the associated combined features for the occurrence time of each event.

[0036] S40. Add the associated combined features for the occurrence time of each event to the initial input features to obtain the target input features for the occurrence time of each event, and based on the target input features for the occurrence time of each event and the basic data set, train the intermediate time prediction models for the occurrence time of each event to obtain the trained intermediate time prediction models for the occurrence time of each event, which are used as the target occurrence time prediction models for the occurrence time of each event.

[0037] In the embodiments of the present invention, the multiple target occurrence time prediction models can be completely different or there can be partially identical models.

[0038] Furthermore, S10 can specifically include: S11. Preprocess the obtained basic data set to obtain the preprocessed basic data set as the training data set.

[0039] In the embodiments of the present invention, the basic data set includes multiple basic data within a set historical time period, and each basic data includes the corresponding event-associated information and the true time of the event occurrence.

[0040] In an embodiment of the present invention, the preprocessing may include: filling the missing data in each piece of basic data to obtain the filled basic data; and removing duplicate data and outliers in the filled basic data.

[0041] In an embodiment of the present invention, if the missing rate of a piece of basic data in the basic data set is less than the set missing rate, then fill the data, otherwise, delete the data. The set missing rate can be set based on actual needs. For example, it can be 0.8.

[0042] In an embodiment of the present invention, existing data filling methods can be used to fill the data. For example, filling with the mean or median, etc.

[0043] In an embodiment of the present invention, the standard score method can be used to remove outliers in the basic data.

[0044] In an embodiment of the present invention, the training data set includes an input feature data set and an output feature data set. The input feature data set is the data set formed corresponding to the input features. The input features may include machine number, aircraft type, route, planned departure date, operating status, planned docking apron, actual arrival time, weather data, and the number of users. The output feature data set is the data set formed based on the actual time of event occurrence.

[0045] S12. Using the training data set, train each initial time prediction model to obtain multiple trained time prediction models, which are respectively used as multiple first time prediction models.

[0046] In an embodiment of the present invention, the initial time prediction model can be an existing data analysis model. For example, it may include: linear regression model, random forest model, long short-term memory network model, etc. The hyperparameters of the initial time prediction model can be randomly initialized hyperparameters.

[0047] As is known to those skilled in the art, any method of using the training data set to train each initial time prediction model to obtain multiple trained time prediction models belongs to the protection scope of the present invention. For example, in a schematic embodiment of the present invention, time series cross-validation can be used for training, such as 5-fold cross-validation. Among them, when training with the training set data obtained by each division each time, the mean squared error loss function can be used to calculate the loss.

[0048] S13. Obtain the prediction performance evaluation value of each first time prediction model for each event occurrence time. For each event occurrence time, use the first time prediction model with the highest prediction performance evaluation value for that event occurrence time as the initial time prediction model for that event occurrence time.

[0049] In the embodiments of the present invention, the prediction performance evaluation value includes a prediction stability evaluation value representing the prediction stability of the model and at least one prediction error evaluation value representing the prediction error of the model. The larger the prediction stability evaluation value is, the more stable the model is. The smaller the prediction error evaluation value is, the more accurate the model is.

[0050] In a schematic embodiment, the prediction performance evaluation value satisfies the following condition: P = W F × PF + (1 - SUM(W a × PC a )) a=1……d ), where PF is the prediction stability evaluation value, and W F is the weight of PF, PC a is the a-th prediction error evaluation value, the value of a ranges from 1 to d, and d is the number of prediction accuracy evaluation values. W a is the weight of PC a . W F > W a . In a schematic embodiment, the weights of all prediction accuracy evaluation values are the same. In a schematic embodiment, W F > 0.5, W a < 0.5. In a specific embodiment, W F = 0.7, W a = 0.3. SUM(W a × PC a ) a=1……d represents the summation operation on W1×PC1 to W d × PC d , that is, SUM(W a × PC a ) a=1……d = (W1×PC1) + …… + (W a × PC a ) + …… + (W d × PC d ).

[0051] In the embodiments of the present invention, the prediction error evaluation value may include mean square error, mean absolute error, root mean square error, mean absolute percentage error, etc.

[0052] In the embodiments of the present invention, the prediction stability evaluation value of each first-time prediction model for the occurrence time of the j-th event satisfies the following condition: PF j = 1 / (1 + (1 / (A1 / ((1 / N)×SUM(|F ij - Y ij |))) i=1……N ), where A1 is an intermediate quantity, and A1 = 1 / (N - 1)×SUM((|F（k+1）j -Y （k+1）j |-|F kj -Y kj |) k=1……N-1 ), SUM(|F ij -Y ij |) i=1……N ) represents the summation operation on |F 1j -Y 1j | to |F Nj -Y Nj | for the summation operation, SUM((|F （k+1）j -Y （k+1）j |-|F kj -Y kj |) k=1……N-1 ) represents the summation operation on (|F 2j -Y 2j |-|F 1j -Y 1j |) to (|F Nj -Y Nj |-|F （N-1）j -Y （N-1）j |) for the summation operation, F ij represents the predicted time of the j-th event occurrence time obtained by each first-time prediction model based on the i-th training data in the training dataset, Y ij is the true time of the j-th event occurrence time corresponding to the i-th training data, and the value of i ranges from 1 to N, where N is the number of training data. F kj represents the predicted time of the j-th event occurrence time obtained by each first-time prediction model based on the k-th training data in the training dataset, Y kj is the true value of F kj that is, the true time of the j-th event occurrence time corresponding to the k-th training data, where the value of k ranges from 1 to N - 1, the value of j ranges from 1 to Q, and the initial value of j is 1.

[0053] Furthermore, in S20, the intermediate-time prediction model for the j-th event occurrence time is obtained through the following steps: S201, for the initial-time prediction model of the j-th event occurrence time, randomly generate m hyperparameter combinations.

[0054] In the embodiments of the present invention, m can be set based on actual needs and is a value greater than 1.

[0055] S202, use the training dataset to train the initial-time prediction model of the j-th event occurrence time with the h-th hyperparameter combination to obtain the prediction performance evaluation value P of the initial-time prediction model of the j-th event occurrence time under the h-th hyperparameter combination jh, the value of h ranges from 1 to m, and the initial value of h is 1.

[0056] In the embodiment of the present invention, P jh =W F ×PF ih +(1 - SUM(W a ×PC ih-a )) a=1……d ), PF ih is the prediction stability evaluation value of the initial time prediction model of the j-th event occurrence time under the h-th hyperparameter combination, and PC ih-a is the a-th prediction error evaluation value of the initial time prediction model of the j-th event occurrence time under the h-th hyperparameter combination. In the embodiment of the present invention, during the process of training the initial time prediction model of the j-th event occurrence time with the h-th hyperparameter combination using the training data set, the loss of the model can be the cross-entropy loss of the j-th event occurrence time.

[0057] S203, take the hyperparameter combination corresponding to the maximum of the m prediction performance evaluation values corresponding to the initial time prediction model of the j-th event occurrence time as the optimal hyperparameter combination of the initial time prediction model of the j-th event occurrence time, and obtain the intermediate time prediction model for each event occurrence time.

[0058] In the embodiment of the present invention, when tuning the parameters of the optimal model for each event occurrence time, the prediction stability evaluation value representing the prediction reliability level is used as an important reference index to guide the tuning direction. When adjusting different parameters, under each set of parameter combinations, the model is trained multiple times. In addition to observing the conventional prediction error index, the corresponding prediction stability evaluation value is calculated at the same time, so that the prediction performance of the selected prediction model can be better.

[0059] Further, in S30, the associated combined features of the j-th event occurrence time are obtained through the following steps: S301, obtain the associated feature set corresponding to the initial input features.

[0060] In an embodiment of the present invention, the associated feature set is a set formed by classification features corresponding to each feature in the initial input features, that is, each associated feature in the associated feature set is a classification feature. The classification feature corresponding to each feature is determined based on the feature type included in the feature. For example, for the feature of machine number, according to the basic data set, it is known that there are g machine numbers, then the classification features corresponding to the machine number include g classification features, respectively corresponding to g machine numbers. Another example is that for the feature of temperature, according to the basic data set, the temperature can be divided into three temperature types: high, medium, and low, then the classification features corresponding to the temperature include three classification features: high temperature, medium temperature, and low temperature. Another example is that for the feature of the number of users, according to the basic data set, the number of users can be divided into three quantity types: many, standard, and few, then the classification features corresponding to the number of users include three classification features: many users, standard number of users, and few users.

[0061] S302. Based on the associated feature set, obtain a plurality of associated feature combinations; wherein, any associated feature combination includes a first combination and a second combination, the first combination includes at least one associated feature, the second combination includes at least one associated feature, and there are no identical associated features between the first combination and the second combination. For example, an associated feature combination is: (heavy rain, few users), and another example is that an associated feature combination is: (heavy rain + strong wind speed, few users).

[0062] S303. Based on the intermediate time prediction model of the j-th event occurrence time, obtain the associated weight of each associated feature combination; wherein, the associated weight corresponding to each associated feature combination satisfies the following condition: I(X1, X2) = log2[P(X1, X2) / P(X1)P(X2)], where I(X1, X2) is the associated weight of the associated feature combination (X1, X2), X1 is the first combination in the associated feature combination (X1, X2), X2 is the second combination in the associated feature combination (X1, X2), P(X1, X2) is the joint weight of the associated feature combination (X1, X2), P(X1) is the marginal weight of X1, and P(X2) is the marginal weight of X2.

[0063] In an embodiment of the present invention, P(X1, X2) = N(X1, X2) / TN(X1, X2), N(X1, X2) is the number of data with a decrease in prediction performance in the prediction result of the intermediate time prediction model of the j-th event occurrence time when the associated feature combination (X1, X2) is used as the input feature, and TN(X1, X2) is the total number of data with the associated feature combination (X1, X2) in the preprocessed basic data set.

[0064] In an embodiment of the present invention, P(X1) satisfies the following condition: P(X1) = TP(X1) / TN(X1), where TP(X1) is the number of data with a decrease in prediction performance in the prediction results of the intermediate time prediction model of the j-th event occurrence time when the associated feature combination (X1, X2) is used as the input feature, and TN(X1) is the total number of data with the first combination in the preprocessed basic dataset.

[0065] In an embodiment of the present invention, P(X2) satisfies the following condition: P(X2) = TP(X2) / TN(X2), where TP(X2) is the number of data with a decrease in prediction performance in the prediction results of the intermediate time prediction model of the j-th event occurrence time when the associated feature combination (X1, X2) is used as the input feature, and TN(X2) is the total number of data with the second combination in the preprocessed basic dataset.

[0066] In an embodiment of the present invention, N(X1, X2), TN(X1), and TN(X2) can be obtained through the following steps: S1, Add the data corresponding to N(X1, X2) to the data containing (X1, X2) in the input feature dataset to obtain a new input feature dataset; S2, Use the new input feature dataset and the output feature dataset to train the initial time prediction model of the j-th event occurrence time to obtain the corresponding training results, where the training results include the prediction results of the event occurrence time corresponding to the new input feature dataset; S3, Based on the training results and the corresponding true results, obtain the prediction performance evaluation value of the j-th event occurrence time corresponding to the new input feature dataset, and use the prediction performance evaluation value corresponding to the data containing (X1, X2) in the input feature dataset as the first to-be-processed prediction performance evaluation value, use the prediction performance evaluation value corresponding to the data containing X1 in the input feature dataset as the second to-be-processed prediction performance evaluation value, and use the prediction performance evaluation value corresponding to the data containing X2 in the input feature dataset as the third to-be-processed prediction performance evaluation value; S4, Traverse the u-th to-be-processed prediction performance evaluation value. For a certain u-th to-be-processed prediction performance evaluation value traversed, if the comparison result between the u-th to-be-processed prediction performance evaluation value and the corresponding reference prediction performance evaluation value indicates that the prediction performance corresponding to the u-th to-be-processed prediction performance evaluation value is lower than the prediction performance corresponding to the reference prediction performance evaluation value, set the u-th counter Cu = Cu + 1, and the initial value of Cu is 0; the reference prediction performance evaluation value is the prediction performance evaluation value of the j-th event occurrence time obtained by the intermediate time prediction model; u = 1, 2, 3.

[0067] In an embodiment of the present invention, if the prediction stability evaluation value and the prediction error evaluation value in the u-th to-be-processed prediction performance evaluation value are respectively less than the prediction stability evaluation value and the prediction error evaluation value in the corresponding reference prediction performance evaluation value, it indicates that the prediction performance corresponding to the u-th to-be-processed prediction performance evaluation value is lower than the prediction performance corresponding to the corresponding reference prediction performance evaluation value, which means that after adding the associated feature combination, the prediction performance of the model has decreased.

[0068] In an embodiment of the present invention, if the prediction performance of the model decreases, it indicates that the added feature combination has an impact on the prediction performance of the model, which can be used as an influencing feature for the event occurrence time.

[0069] S5. Set N(X1, X2) = C1, TN(X1) = C2, and TN(X2) = C3.

[0070] S304. Use the associated feature combination with the maximum association weight as the associated combined feature for the occurrence times of j events.

[0071] In summary, the data processing method for predicting the occurrence time of key events in flight turnover provided by the embodiment of the present invention obtains the event occurrence time based on the fusion weighted result of the prediction results of multiple target event occurrence time prediction models. Each target event occurrence time prediction model is used to predict the corresponding event occurrence time, and the influence of different factor combinations on different event occurrence times is considered during the training process, which can enable the model to learn the complex relationships between events. In addition, the reliability and error and other performances of the model are comprehensively considered during the training process, which can make the prediction performance of the model more reliable. Furthermore, historical data is used to correct the prediction results of the model, which can make the predicted event occurrence time more accurate compared to the results predicted by the existing event occurrence time prediction methods.

[0072] The embodiment of the present invention further provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are configured to execute the method described in the embodiment of the present invention.

[0073] The embodiment of the present invention further provides a computer-readable storage medium storing computer-executable instructions, and the computer instructions are used to execute the method described in the embodiment of the present invention.

[0074] It should be understood that various forms of the processes shown above can be used, reordering, adding, or deleting steps. For example, the steps recited in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in the present invention can be achieved, and no limitations are imposed herein.

[0075] The above specific embodiments do not constitute a limitation to the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub - combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A data processing method for predicting the occurrence time of key events in flight turnarounds, characterized in that, The occurrence time of the key events in flight turnover is the occurrence time of the key events in flight turnover experienced by the event-related object from arriving at the specified geographical area to leaving the specified geographical area. The method includes the following steps: S100, obtain the dataset to be processed that needs to be processed currently; the dataset to be processed includes at least one event-related information, and the event-related information includes basic information of the event-related object, weather data, and the number of users of the event-related object; S200, input the dataset to be processed into multiple target occurrence time prediction models to obtain multiple event occurrence time prediction results; S300, perform weighted fusion on the multiple event occurrence time prediction results to obtain the predicted value of the occurrence time corresponding to each event-related information in the dataset to be processed; S400, based on the preset reference database, adjust the predicted value of the occurrence time corresponding to each event-related information to obtain the target predicted value of the occurrence time corresponding to each event-related information. Multiple reference data are stored in the preset reference database, and each reference data includes the corresponding event-related information and the real time of the occurrence time.

2. The method according to claim 1, characterized in that, In S300, the predicted value t of the occurrence time of the v-th event corresponding to each event association information v pre satisfies the following condition: t v pre = w1 × t 1v pre + w2 × t 2v pre + …… + w z × t zv pre + …… + w Q × t Qv pre , where the value of v ranges from 1 to Q, Q is the number of event occurrence times to be predicted, and w z is the weight of the z-th target occurrence time prediction model, and t zv pre is the predicted value of the occurrence time of the v-th event predicted by the z-th target occurrence time prediction model, and the value of z ranges from 1 to Q.

3. The method according to claim 2, wherein w z Meet the following conditions: w z =AC z / (AC1 + AC2 + …… + AC z + …… + AC Q ),AC z is the prediction accuracy of the z-th target occurrence time prediction model for the corresponding event occurrence time.

4. The method according to claim 2, wherein S400 specifically includes: S401, for each event-related information, obtain the similarity between the event-related information and the event-related information corresponding to each reference data in the preset reference database to obtain the similarity set corresponding to the event-related information; S402, use the reference data corresponding to the similarity greater than or equal to the set similarity threshold in the similarity set corresponding to each event-related information as the associated data of the event-related information; S403, based on the real time of the occurrence time corresponding to the associated data of each event-related information, adjust the predicted value of the occurrence time corresponding to each event-related information to obtain the target predicted value of the occurrence time corresponding to each event-related information.

5. The method according to claim 4, wherein The target predicted value t of the occurrence time of the v-th event corresponding to each event association information v Satisfies the following conditions: t v = t v pre + SUM(w e × Δt ev ) e=1……H ; where, w e is the weight of the e-th associated data corresponding to the event associated information, w e is positively correlated with the similarity between the event associated information and the event associated information in the corresponding e-th associated data, and the value of e ranges from 1 to H, where H is the number of associated data of the event associated information; △t ev is t v pre and the difference between the real time of the occurrence time of the v-th event in the corresponding e-th associated data and the event associated information.

6. The method according to claim 1, wherein The multiple target occurrence time prediction models are obtained through the following steps: S10, based on the basic dataset, train multiple pre-constructed initial time prediction models to obtain the initial time prediction models for predicting the occurrence time of each event; S20, adjust the hyperparameter combinations of the initial time prediction models for the occurrence time of each event to obtain the optimal hyperparameter combinations of the initial time prediction models for the occurrence time of each event, and obtain the intermediate time prediction models for the occurrence time of each event; S30, based on the initial input features and the intermediate occurrence time prediction models for the occurrence time of each event, obtain the associated combined features for the occurrence time of each event; S40, add the associated combined features for the occurrence time of each event to the initial input features to obtain the target input features for the occurrence time of each event, and based on the target input features for the occurrence time of each event and the basic dataset, train the intermediate time prediction models for the occurrence time of each event to obtain the trained intermediate time prediction models for the occurrence time of each event, as the target occurrence time prediction models for the occurrence time of each event.

7. The method according to claim 6, characterized in that In S20, the intermediate time prediction model for the jth event occurrence time is obtained through the following steps: S201. For the initial time prediction model of the occurrence time of the j-th event, randomly generate m combinations of hyperparameters. The value of j ranges from 1 to Q, and the initial value of j is 1, where Q is the number of event occurrence times to be predicted. S202. Use the training data set to train the initial time prediction model for the occurrence time of the j-th event with the h-th hyperparameter combination, and obtain the prediction performance evaluation value P of the initial time prediction model for the occurrence time of the j-th event under the h-th hyperparameter combination jh , where h ranges from 1 to m, and the initial value of h is 1; S203. Take the combination of hyperparameters corresponding to the maximum value among the m prediction performance evaluation values of the initial time prediction model of the occurrence time of the j-th event as the optimal combination of hyperparameters of the initial time prediction model of the occurrence time of the j-th event, and obtain the intermediate time prediction model of the occurrence time of the j-th event.

8. The method according to claim 7, wherein In S30, the associated combined features of the occurrence time of the j-th event are obtained through the following steps: S301. Obtain the set of associated features corresponding to the initial input features. The set of associated features is a set formed by the classification features corresponding to each feature in the initial input features. S302. Based on the set of associated features, obtain multiple combinations of associated features. Among them, any combination of associated features includes a first combination and a second combination. The first combination includes at least one associated feature, the second combination includes at least one associated feature, and there are no identical associated features between the first combination and the second combination. S303. Based on the intermediate time prediction model of the occurrence time of the j-th event, obtain the associated weight of each combination of associated features. Among them, the associated weight corresponding to each combination of associated features satisfies the following condition: I(X1, X2) = log2[P(X1, X2) / P(X1)P(X2)], where I(X1, X2) is the associated weight of the combination of associated features (X1, X2), X1 is the first combination in the combination of associated features (X1, X2), X2 is the second combination in the combination of associated features (X1, X2), P(X1, X2) is the joint weight of the combination of associated features (X1, X2), P(X1) is the marginal weight of X1, and P(X2) is the marginal weight of X2. S304. Take the combination of associated features with the maximum associated weight as the associated combined features of the occurrence time of the j-th event.

9. The method according to claim 8, characterized in that, Among them, P(X1, X2) = N(X1, X2) / TN(X1, X2), where N(X1, X2) is the number of data with a decrease in prediction performance in the prediction results of the intermediate time prediction model of the occurrence time of the j-th event when the combination of associated features (X1, X2) is used as the input feature, and TN(X1, X2) is the total number of data with the combination of associated features (X1, X2) in the preprocessed basic dataset. P(X1) satisfies the following condition: P(X1) = TP(X1) / TN(X1), where TP(X1) is the number of data with a decrease in prediction performance in the prediction results of the intermediate time prediction model of the occurrence time of the j-th event when the combination of associated features (X1, X2) is used as the input feature and representing the data containing the first combination X1, and TN(X1) is the total number of data with the first combination in the preprocessed basic dataset. P(X2) satisfies the following condition: P(X2) = TP(X2) / TN(X2), where TP(X2) is the number of data showing a decline in prediction performance among the data representing the inclusion of the second combination X2 in the prediction results of the intermediate time prediction model for the occurrence time of the j-th event when the associated feature combination (X1, X2) is used as the input feature, and TN(X2) is the total number of all data with the second combination in the preprocessed basic dataset.

Images