Freight rate demand prediction method and system, storage medium and equipment
By vectorized encoding and model training on multi-source heterogeneous data, the problem of insufficient accuracy and adaptability of freight demand prediction in the prior art is solved, and a more refined and accurate freight demand prediction is achieved.
Patent Information
- Application Number
- CN202411505806.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-25
- Publication Date
- 2025-03-28
AI Technical Summary
The existing technology is difficult to achieve high-precision civil aviation freight demand forecasts in the long-term and wide markets, due to the single variable analysis method, data sensitivity, and model limitations.
By vectorized encoding of multi-source heterogeneous data, a stable vectorized encoding model is trained using the SEQ2VEC algorithm model, and the basic data is processed and predicted by combining the LSTM network and neural network model.
Comprehensive consideration of multiple factors and multi-source data has been achieved, the refinement and accuracy of freight demand forecasts have been improved, and the freight influencing factors of economic and time changes have been adapted to freight influencing factors.
Smart Images
Figure CN119722140A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of domestic freight rate publication, and in particular to a method, system, storage medium and device for predicting freight rate demand. Background Art
[0002] With the rapid development of the civil aviation industry, the advantages of airlines in market competition are increasingly dependent on high-precision forecasts of fare demand. Civil aviation fare demand is affected by a variety of complex factors, such as changes in the economic situation, emergencies, passenger preferences, fare rules, etc., which makes it difficult for traditional single variable analysis methods to meet current demand forecasting needs.
[0003] At present, the forecasting methods of freight demand mainly rely on short-term (such as within 2 months) historical data for trend forecasting, which is limited to a specific market environment; some studies have tried to combine social media event keywords to improve the forecasting accuracy, but they are still limited by data sensitivity and model limitations; simply analyzing one factor is difficult to maintain the stability and accuracy of the forecast in the long-term and broad market, and it is difficult to fully consider the market positioning and target audience. Therefore, developing a model that can comprehensively consider multi-source data, multiple factors, and effectively predict civil aviation freight demand has become an important research direction for the current digital transformation of the civil aviation industry. Summary of the invention
[0004] To solve the above problems, the present invention provides a freight rate demand forecasting method, system, storage medium and device, which can quickly capture and locate the impact of current influencing factors on the demand of a given customer group through vectorized encoding of multi-source heterogeneous data, thereby achieving refined demand forecasting.
[0005] The above objectives can be achieved through the following solutions:
[0006] A method for predicting freight demand comprises collecting data information of factors influencing freight demand to obtain basic data information; using the basic data information to train a SEQ2VEC algorithm model into a stable vectorized coding model; using the vectorized coding model to vectorize the basic data information; receiving the vectorized basic data information, and inputting the basic data information into a neural network model after analysis and processing to obtain demand forecast data.
[0007] Optionally, using the basic data information to train the SEQ2VEC algorithm model into a stable vectorized coding model includes: selecting data from the basic data information to establish a training data set; and training the SEQ2VEC algorithm model on the training data set to obtain a vectorized coding model.
[0008] Optionally, the basic data information includes customer model data, structured data, unstructured data and semi-structured data.
[0009] Optionally, the method also includes: performing dimensionality reduction processing on the customer model data after vectorization processing to obtain first basic data; inputting the first basic data into the LSTM network model for processing to obtain first feature data; inputting the first feature data into the neural network model for processing to obtain the first predicted data.
[0010] Optionally, the method further includes: performing dimensionality reduction processing on the structured data after vectorization processing to obtain second basic data; and inputting the second basic data into a neural network model for processing to obtain the second predicted data.
[0011] Optionally, the method further includes: inputting the vectorized unstructured data into an LSTM network model for processing to obtain second feature data; and inputting the second feature data into a neural network model for processing to obtain the third prediction data.
[0012] Optionally, the method also includes: extending the structured part of the vectorized semi-structured data by adding a lateral dimension of the unstructured part to obtain third basic data; performing dimensionality reduction processing on the third basic data to obtain fourth basic data; inputting the unstructured data and the fourth basic data in the vectorized semi-structured data into the LSTM network model for processing to obtain third feature data; inputting the third feature data into the neural network model for processing to obtain the fourth predicted data.
[0013] Based on the same inventive concept, the present invention also provides a freight demand forecasting system, which includes: a data collection module, used to collect data information of factors affecting freight demand to obtain basic data information; a vectorized coding model generation module, used to use the basic data information to train the SEQ2VEC algorithm model into a stable vectorized coding model; a vectorized coding module, used to use the vectorized coding model to vectorize the basic data information; a freight demand forecasting module, used to receive the basic data information after vectorization processing, and input it into a neural network model after analysis and processing to obtain demand forecasting data.
[0014] Based on the same inventive concept, the present invention also provides a computer storage medium storing one or more programs, which can implement any of the aforementioned methods when the one or more programs are executed.
[0015] Based on the same inventive concept, the present invention also provides a device, including a processor, a communication interface, a memory and a communication bus; the processor, the communication interface and the memory communicate with each other through the communication bus; the processor is used to execute the program stored in the aforementioned computer-readable storage medium.
[0016] Compared with the prior art, the present invention has the following advantages:
[0017] 1. The present invention fully considers the multiple factors that affect civil aviation fares. By vectorizing and encoding multi-source heterogeneous data, it can quickly capture and locate the impact of current influencing factors on the demand of a given customer group and achieve refined demand forecasting;
[0018] 2. Through vector coding and LSTM networks, structured, semi-structured, and unstructured data that affect civil aviation freight demand can be processed. It is easy to expand and can adapt to changes in freight factors that change with the economy and time.
[0019] 3. Vectorized processing facilitates the analysis of similarities and correlations between influencing factors;
[0020] 4. It can support distributed data processing and realize learning applications of massive data.
[0021] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0023] Figure 1 The figure is a flow chart of a method for predicting freight demand according to an embodiment of the present invention.
[0024] Figure 2 It is a flowchart of vectorized coding learning according to an embodiment of the present invention.
[0025] Figure 3 It is a schematic diagram of customer group classification simulation projection in three-dimensional space according to an embodiment of the present invention.
[0026] Figure 4 1 is a diagram of a preprocessing architecture of factors affecting freight demand according to an embodiment of the present invention.
[0027] Figure 5 It is a schematic diagram of a demand forecast curve of a product approaching the sales period according to an embodiment of the present invention.
[0028] Figure 6 It is a structural schematic diagram of a freight rate demand forecasting system according to an embodiment of the present invention.
[0029] Figure 7 It is a structural schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0030] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0031] Reference Figure 1 An embodiment of the present invention proposes a method for predicting freight demand. By vectorizing and encoding multi-source heterogeneous data, the method can quickly capture and locate the impact of current influencing factors on the demand of a given customer group, thereby achieving refined demand prediction.
[0032] The prediction method of this embodiment specifically includes:
[0033] Collect data information on factors affecting freight rate demand and obtain basic data information;
[0034] Specifically, factors that have a significant impact on the demand for civil aviation products include the following aspects: customer selection preferences, sales time characteristics, and currently available products. At the same time, the occurrence of extreme weather and hot events will also significantly affect travel demand in related areas. These data and information are collected to obtain basic data information.
[0035] Using the basic data information, the SEQ2VEC algorithm model is trained into a stable vectorized coding model;
[0036] Vectorizing the basic data information using the vectorized coding model;
[0037] The basic data information after vectorization processing is received, and after analysis and processing, it is input into the neural network model to obtain demand forecast data.
[0038] Exemplarily, data information on factors affecting freight demand is collected to generate basic data information; a SEQ2VEC algorithm model is established, and the basic data information is used to train the SEQ2VEC algorithm model into a stable vectorized coding model; the basic data information is then input into the vectorized coding model for vectorized processing, and then the vectorized basic data information is analyzed and processed and input into the Shenjiang network model, thereby outputting demand forecast data.
[0039] Optionally, the using the basic data information to train the SEQ2VEC algorithm model into a stable vectorized coding model includes:
[0040] Selecting data from the basic data information to establish a training data set;
[0041] Specifically, vectorization is carried out through distributed learning, learning one sub-batch at a time, and continuously iterating and optimizing. Before the formal vectorization process, a sub-batch accounting for 1% of the basic data information will be pre-trained.
[0042] The SEQ2VEC algorithm model is trained on the training data set to obtain a vectorized coding model.
[0043] For example, Figure 2 As shown, the collected basic data information is input in batch form, and then the batch basic data information is pre-coded and converted into a format suitable for processing by the SEQ2VEC algorithm model. The SEQ2VEC algorithm model converts the pre-coded basic data information into an embedded vector; a regression analysis is performed on the embedded vector to obtain a predicted value of future freight rate demand or flight demand. In this process, the principal component analysis of the embedded vector is performed by a dimensionality reduction algorithm to achieve dimensionality reduction, so that the input sequence content can be comprehensively and targetedly represented using lower-dimensional data, and used for subsequent learning processes; the dimensionality reduction algorithm can be PCA, LDA, MDS, ISOMAP, SNE, T-SNE, AutoEncoder, etc., and the dimensionality reduction algorithm in this scheme mainly uses PCA; at the same time, the result of the embedded vector is also easier to realize the analysis of the similarity of different factors, that is, the similarity of the two factors can be seen through the Euclidean space distance of the two vectors.
[0044] Optionally, the basic data information includes customer model data, structured data, unstructured data and semi-structured data.
[0045] Exemplarily, the customer group model data mainly considers the main selection factors affecting passenger choices and the sensitivity analysis of passengers; the main selection factors affecting passenger choices include departure date, acceptable flight time period, fare range, cabin class and other basic information; passenger sensitivity analysis can be that a certain group of business passengers is sensitive to strict refund and change rules, etc.; structured data includes civil aviation sales data, search data and fare data, which can be directly encoded during the processing process; unstructured data includes event data that has a significant impact on the short-term demand for civil aviation products, including hot events, emergencies, conferences and exhibitions, tourism promotion, etc. Some special policy changes will also significantly affect the demand of related cities, such as the construction of free trade ports, etc. These data can only be extracted through the learning of their keywords, including event content, importance (optional, such as international events and locations District events), place of occurrence, time range (optional, such as there is a Hundred Flowers Festival in a certain place from a certain day to a certain day), scope of impact (optional, such as three cities jointly hosting a certain event), etc. After extracting basic key information from the event data, vectorized learning is performed, and the learning results are input into the subsequent network model; semi-structured data includes recent weather conditions, etc. This part of the data has its structured characteristics. For example, the weather forecast can predict heavy to torrential rain in a certain place, the probability of rainfall, etc. These are all structured, but there are also unstructured parts such as the scope of impact of torrential rain, etc. At the same time, considering the characteristics of the continuous evolution of factors such as events, and the characteristics that may affect each other, before connecting to the final neural network, it first passes through a layer of long short-term memory network to realize the memory of recent important information and the forgetting of "expired" information, and then the demand forecast data is obtained through the output of the neural network model.
[0046] Optionally, the method further comprises:
[0047] Performing dimensionality reduction processing on the customer group model data after vectorization processing to obtain first basic data;
[0048] For example, Figure 3As shown in the figure, through the actual purchase data of passengers of different categories, the characteristics of passengers are analyzed, and the decision direction (passenger sensitivity) that has a strong influence on the decision-making behavior of passengers is determined, so as to realize the classification learning of passengers and support the impact of changes in potential products or influencing factors on passenger demand; for example, for a group of short-term leisure passengers, their group characteristics include: the origin and destination are adjacent cities, they are usually active when there are marathon events, and they have a certain tolerance range for the price of fare products; at the same time, their purchasing behavior shows that this group is more sensitive to the take-off and landing time, that is, flights departing / arriving in the middle of the night are almost not considered; the display of the simulated projection of passenger group classification in three-dimensional space; the display of the simulated projection of passenger group classification in three-dimensional space, the point is the position of the passenger identity vector after quantization and projection to the three-dimensional space, and the arrow is the direction obtained by its sensitivity analysis that can significantly affect its purchase intention.
[0049] Inputting the first basic data into the LSTM network model for processing to obtain first feature data;
[0050] The first feature data is input into a neural network model for processing to obtain the first prediction data.
[0051] For example, Figure 4 As shown, the customer group model data can be structured. For example, there are some customer group models in the current academic community that support the classification of passengers into groups based on factors such as age and gender, and then predict their needs. Therefore, after this part of the data is vectorized by the SQE2VEC algorithm model, it is also necessary to reduce the dimension through a dimensionality reduction algorithm. The customer group model data can also be unstructured. For example, some studies only focus on some specific factors that affect passenger behavior, including the impact of the need to stay at the airport on passenger choices. Therefore, this part of the data needs to be memorized by the LSTM network model for recent important information and forgotten for "expired" information. The analyzed and processed data is input into the neural network model to output the first prediction data for predicting the fare demand.
[0052] Optionally, the method further comprises:
[0053] Performing dimensionality reduction processing on the structured data after vectorization to obtain second basic data;
[0054] Specifically, structured data includes civil aviation sales data, search data and freight rate data. The recent demand trends can be obtained through sales data and search data analysis, and the current market status can be obtained through freight rate data analysis.
[0055] For example, Figure 5As shown in the figure, an example of a demand forecast curve for a product approaching its sales period is shown. Usually, the overall demand for freight products has a strong time correlation, including whether it is winter or summer vacation, whether it is a statutory holiday, whether it is a peak or off-season, midweek or weekend... In terms of the overall demand trend, there is also a demand for purchasing in advance. For example, long-distance leisure travelers usually start planning their trips early, and have determined their round-trip needs one to two months in advance and purchased related products; while some business travel needs only arise within a week in advance; the influencing factors of emergencies are more complex, and many influencing factors must be considered comprehensively, such as the scope of impact of these events, the people involved, etc.; therefore, these trend ranges can be learned through sales data and search data, and abnormal search ratios can also assist in determining the authenticity of the existence of influencing factors.
[0056] For example, the current market status is obtained through fare data analysis. The market status data is mainly reflected in the fare products that can be selected in the current market. This part of data is mainly the products released by airlines that are bound to multiple fare rules. When passengers have travel needs, if there are no completely satisfactory or satisfactory products, they will also adjust their needs in combination with the products available in the current market. This part is combined with the passenger sensitivity analysis results of the passenger model data to learn the behavior of passengers adjusting their needs according to market conditions; that is, when there is no suitable product, if the cosine similarity of the product characteristics of the available products and the passenger sensitivity analysis feature vector is closer to -1, the passenger is more likely to have a demand transfer. The cosine similarity formula is as follows:
[0057]
[0058] Among them, S is the product feature vector of the reachable product; P s is the passenger sensitivity analysis feature vector; i is the index from 1 to n, n is the dimension of the vector; S i is the value of the product feature vector of the reachable product in the i-th dimension (or feature); P s_i is the value of the passenger sensitivity analysis feature vector in the i-th dimension (or feature);
[0059] For example, when there are no fare products with better departure times, passengers who are less sensitive to departure and arrival times may choose flights with later departure times; similar measurement criteria also include: "Tanimoto" coefficient, Ochiai coefficient, etc.
[0060] The second basic data is input into a neural network model for processing to obtain the second predicted data.
[0061] For example, Figure 4As shown, the sales data, search data and freight data are structured data, so the sales data, search data and freight data can be directly encoded and vectorized through the SQE2VEC algorithm model and then processed through a dimensionality reduction algorithm to obtain the second basic data, and then the second basic data is input into the neural network model to obtain the second prediction data of the freight demand.
[0062] Optionally, the method further comprises:
[0063] The vectorized unstructured data is input into the LSTM network model for processing to obtain the second feature data;
[0064] For example, for civil aviation demand, many emergencies will lead to short-term and substantial changes in specific demand, and the ability to capture, adapt and adjust this part of demand is also crucial; such as changes in business transactions caused by the release of policies, changes in travel demand caused by hot events, etc.; this part of data usually comes from hot searches on various websites and search engines, which are usually unstructured, and it is necessary to analyze the characteristics of hot events from the corpus, the scope of impact, the main affected population, the time period involved and other factors; from the perspective of the market economy, simultaneous events also have a complementary or mutually substitutive relationship for demand forecasting. For example, if there are events that attract tourists in nearby cities at the same time, it is very likely to attract tourists to plan a one-time travel arrangement, but the attraction events between distant cities are in a competitive relationship for transportation demand; in this system, through SEQ2VEC encoding, the characteristics of current events can be effectively classified and sorted, similar events can be found, and the impact of events can be located. At the same time, it can be verified with demand trend data to capture the changes in demand caused by hot events; when multiple events occur at the same time, filtering through the LSTM network model can also realize the memory of key events, the memory of relatively important events, and the forgetting of outdated events.
[0065] The second feature data is input into a neural network model for processing to obtain the third prediction data.
[0066] For example, Figure 4As shown in the figure, unstructured data is often text data, such as the hot search information containing "a certain city will hold a certain festival recently", which will lead to changes in related demand; the SQE2VEC algorithm model analyzes and vectorizes these information, and then the vectorized data is processed by the LSTM network model to obtain the second feature data, and then the second feature data is input into the neural network model to obtain the third prediction data to filter the information with low relevance; the LSTM network model includes a forget gate, an add gate and an output gate, which can realize the memory and forgetting of key information; for example, for the same hot search information, the information "a certain city will hold a certain festival recently" has a high probability of being output to the neural network through the add gate; after the hot search continues for a period of time, with the emergence of new hot searches, the original information "a certain city will hold a certain festival recently" will also pass through the forget gate to achieve the dilution of the information.
[0067] Optionally, the method further comprises:
[0068] In the vectorized semi-structured data, the structured part is extended by the horizontal dimension of the unstructured part to obtain the third basic data;
[0069] Performing dimensionality reduction processing on the third basic data to obtain fourth basic data;
[0070] For example, for semi-structured data, the structured part is first encoded, and the horizontal dimension of the unstructured part is extended, and then the dimensionality reduction process is uniformly performed.
[0071] Inputting the unstructured data and the fourth basic data in the semi-structured data after the vectorization processing into the LSTM network model for processing to obtain third feature data;
[0072] For example, considering that semi-structured data also contains unstructured text or keyword information, it is also necessary to memorize key information through the LSTM layer.
[0073] The third feature data is input into a neural network model for processing to obtain the fourth prediction data.
[0074] For example, Figure 4As shown in the figure, the impact of weather data on civil aviation demand is mainly for short-term demand. For example, if a city has a heavy rainstorm and it is expected to last for 5 days, passengers who need to travel during these five days will hardly consider using civil aviation as a means of transportation. As the accuracy of weather forecasts continues to improve, the impact of such bad weather on demand and demand planning is also increasing. Therefore, the introduction of this part of data can assist airlines in adjusting flight plans as early as possible. After these semi-structured data are vectorized by the SQE2VEC algorithm model, the structured part of the vectorized semi-structured data is reduced in dimension by a dimensionality reduction algorithm. The unstructured part of the vectorized semi-structured data is processed using the LSTM network model. All the processed data are input into the neural network model to obtain the fourth forecast data of freight demand.
[0075] Based on the same inventive concept, Figure 6 As shown, the present invention also provides a freight rate demand forecasting system, the system comprising:
[0076] The data collection module is used to collect data information on factors affecting freight rate demand and obtain basic data information;
[0077] A vectorized coding model generation module, used to train the SEQ2VEC algorithm model into a stable vectorized coding model using the basic data information;
[0078] A vectorized coding module, used for performing vectorized processing on the basic data information by using the vectorized coding model;
[0079] The freight demand forecasting module is used to receive the basic data information after vectorization processing, and input it into the neural network model after analysis and processing to obtain demand forecasting data.
[0080] Based on the above disclosed content, the present invention also provides an electronic device. Figure 7 As shown, the electronic device of an embodiment of the present invention includes at least one electrically connected processor and at least one storage medium, wherein the storage medium is electrically connected to the processor, wherein the storage medium stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method as described above.
[0081] Based on the same inventive concept, the present invention also provides a storage medium, which stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method as described above.
[0082] It should be noted that the electrical connection between the above-mentioned units does not necessarily mean the connection between lines. An indirect connection method can be applied to the embodiments of the present invention as long as the purpose of the present invention is achieved.
[0083] The above is only an exemplary embodiment of the present invention and cannot be used to limit the scope of the present invention. That is, any equivalent changes and modifications made according to the teachings of the present invention are still within the scope of the present invention. After considering the disclosure of the specification and the truth of practice, it will be easy for those skilled in the art to think of other embodiments of the present invention. This application is intended to cover any variation, use or adaptive change of the present invention, which follows the general principles of the present invention and includes common knowledge or customary technical means in the technical field not recorded in the present invention.
Claims
1. A method for predicting freight demand, characterized in that: The method comprises: Collect data information on factors affecting freight rate demand and obtain basic data information; Using the basic data information, the SEQ2VEC algorithm model is trained into a stable vectorized coding model; Vectorizing the basic data information using the vectorized coding model; The basic data information after vectorization processing is received, and after analysis and processing, it is input into the neural network model to obtain demand forecast data.
2. A method for predicting freight demand according to claim 1, characterized in that: The step of using the basic data information to train the SEQ2VEC algorithm model into a stable vectorized coding model includes: Selecting data from the basic data information to establish a training data set; The SEQ2VEC algorithm model is trained on the training data set to obtain a vectorized coding model.
3. A method for predicting freight demand according to claim 1, characterized in that: The basic data information includes customer model data, structured data, unstructured data and semi-structured data.
4. A method for predicting freight demand according to claim 3, characterized in that: The method further comprises: Performing dimensionality reduction processing on the customer group model data after vectorization processing to obtain first basic data; Inputting the first basic data into the LSTM network model for processing to obtain first feature data; The first feature data is input into a neural network model for processing to obtain the first prediction data.
5. A method for predicting freight demand according to claim 3, characterized in that: The method further comprises: Performing dimensionality reduction processing on the structured data after vectorization to obtain second basic data; The second basic data is input into a neural network model for processing to obtain the second predicted data.
6. A method for predicting freight demand according to claim 3, characterized in that: The method further comprises: The vectorized unstructured data is input into the LSTM network model for processing to obtain the second feature data; The second feature data is input into a neural network model for processing to obtain the third prediction data.
7. A method for predicting freight demand according to claim 3, characterized in that: The method further comprises: In the vectorized semi-structured data, the structured part is extended by the horizontal dimension of the unstructured part to obtain the third basic data; Performing dimensionality reduction processing on the third basic data to obtain fourth basic data; Inputting the unstructured data and the fourth basic data in the semi-structured data after the vectorization processing into the LSTM network model for processing to obtain third feature data; The third feature data is input into a neural network model for processing to obtain the fourth prediction data.
8. A freight rate demand forecasting system, characterized in that: The system comprises: The data collection module is used to collect data information on factors affecting freight rate demand and obtain basic data information; A vectorized coding model generation module, used to train the SEQ2VEC algorithm model into a stable vectorized coding model using the basic data information; A vectorized coding module, used for performing vectorized processing on the basic data information by using the vectorized coding model; The freight demand forecasting module is used to receive the basic data information after vectorization processing, and input it into the neural network model after analysis and processing to obtain demand forecasting data.
9. A computer storage medium, characterized in that: One or more programs are stored, and when the one or more programs are executed, a freight rate demand forecasting method as described in any one of claims 1-7 can be implemented.
10. A device, characterized in that: The system comprises a processor, a communication interface, a memory and a communication bus; the memory stores at least one program in a computer storage medium as claimed in claim 9 which can be loaded and executed by the processor.