CN-121998161-A - Method, device, storage medium and computer equipment for predicting cross-border port entry and exit passenger flow

CN121998161ACN 121998161 ACN121998161 ACN 121998161ACN-121998161-A

Abstract

The application provides a cross-border port entrance and exit passenger flow prediction method, a prediction device, a storage medium and computer equipment. The method comprises the steps of carrying out parameterization processing on vacation features of a cross-border area to obtain time sequence features, constructing to obtain multi-scale passenger flow local fragment features, constructing to obtain vacation position features, constructing to obtain passenger flow input features, sequentially carrying out single-flow multi-scale self-attention calculation, cross-flow co-scale interaction calculation and multi-scale fusion calculation on the inbound passenger flow input features and the outbound passenger flow input features to obtain inbound passenger flow fusion features, constructing query vectors according to features of a target time period, and obtaining an inbound passenger flow predicted value and an outbound passenger flow predicted value according to future query vectors and the inbound passenger flow fusion features. The prediction method can effectively solve the problems of dual specificity of holiday influence, time lag correlation of the incoming and outgoing passenger flows and the like in the cross-border port incoming and outgoing scene, and obtain more accurate passenger flow prediction values.

Inventors

ZHAO JUANJUAN
ZHAO YANZHEN
ZHANG FAN
XIE WEIHAO
YE KEJIANG

Assignees

中国科学院深圳先进技术研究院
中华人民共和国深圳湾出入境边防检查站

Dates

Publication Date: 20260508
Application Date: 20251217

Claims (10)

1. A cross-border port entry and exit passenger flow prediction method, the method comprising: The construction stage of the input features comprises the steps of carrying out parameterization processing on the holiday features of the cross-border area to obtain time sequence features, constructing to obtain multi-scale passenger flow local segment features, and constructing to obtain holiday position features; The double-flow encoder processing stage comprises the steps of sequentially carrying out single-flow multi-scale self-attention calculation and cross-flow co-scale interaction calculation on an inbound passenger flow input feature and an outbound passenger flow input feature respectively to obtain an inbound passenger flow interaction feature and an outbound passenger flow interaction feature; and the future time perception decoder processing stage is used for constructing a query vector according to the characteristics of the target time period, and obtaining an inbound passenger flow predicted value and an outbound passenger flow predicted value according to the future query vector and the inbound passenger flow fusion characteristics.
2. The cross-border port entry and exit passenger flow prediction method according to claim 1, wherein parameterizing the holiday characteristics of the cross-border area to obtain the time sequence characteristics comprises: And constructing a holiday characteristic system according to the intra-day time information, the week parameter, the holiday length parameter and the holiday sequence parameter of the cross-border area to serve as a time sequence characteristic.
3. The cross-border port entry and exit passenger flow prediction method according to claim 1, wherein the constructing to obtain the multi-scale passenger flow local segment features comprises: Constructing short-scale passenger flow local fragment features, medium-scale passenger flow local fragment features and long-scale passenger flow local fragment features of the passenger flow sequences and the external features of the plurality of historical time slices according to different time lengths; and respectively carrying out reversible real force normalization treatment on the short-scale passenger flow local fragment characteristics, the medium-scale passenger flow local fragment characteristics and the long-scale passenger flow local fragment characteristics.
4. The cross-border port entry and exit passenger flow prediction method according to claim 3, wherein the single-flow multi-scale self-attention calculation and cross-flow co-scale interaction calculation are sequentially performed on the entry passenger flow input feature and the exit passenger flow input feature respectively to obtain an entry passenger flow interaction feature and an exit passenger flow interaction feature, and the method comprises the following steps: Sequentially performing self-attention computation on the inbound traffic input features of each scale to obtain inbound traffic self-attention features of each scale, and sequentially performing self-attention computation on the outbound traffic input features of each scale to obtain outbound traffic self-attention features of each scale; And sequentially carrying out cross attention calculation on the inbound passenger flow self attention characteristic and the outbound passenger flow self attention characteristic of the same scale to obtain inbound passenger flow interaction characteristics and outbound passenger flow interaction characteristics of all scales.
5. The cross-border port entry and exit passenger flow prediction method according to claim 4, wherein performing multi-scale fusion calculation on the entry passenger flow interaction feature and the exit passenger flow interaction feature to obtain an entry passenger flow fusion feature comprises: And adopting a fusion strategy to fuse the inbound passenger flow interaction characteristics and the outbound passenger flow interaction characteristics of each scale, so as to obtain the inbound passenger flow fusion characteristics.
6. The cross-border port entry and exit passenger flow prediction method of claim 2, wherein constructing the query vector from the characteristics of the target time period comprises: and embedding a plurality of target time periods and external features in the future into a unified dimension to obtain a query vector, wherein the external features comprise holiday parameters, week information and weather forecast.
7. The cross-border port outbound customer flow prediction method of claim 6, wherein obtaining an inbound customer flow prediction value and an outbound customer flow prediction value from the future query vector and the outbound customer flow fusion feature comprises: performing double-layer interaction mechanism fusion on the future query vector and the inbound/outbound passenger flow fusion characteristic to obtain a history characteristic; performing cross-attention fusion on the future query vector and the history feature to obtain decoder output features; and inputting the output characteristics of the decoder into a linear projection layer to obtain an incoming passenger flow predicted value and an outgoing passenger flow predicted value.
8. A cross-border port entry and exit passenger flow prediction apparatus, the apparatus comprising: The system comprises an input feature construction unit, a passenger flow input feature generation unit, a passenger flow input unit and a passenger flow input unit, wherein the input feature construction unit is configured to carry out parameterization processing on the holiday features of a cross-border area to obtain time sequence features, construct multi-scale passenger flow local segment features and construct the holiday position features; The double-flow encoder processing unit is configured to sequentially perform single-flow multi-scale self-attention computation and cross-flow co-scale interaction computation on the inbound passenger flow input feature and the outbound passenger flow input feature respectively to obtain inbound passenger flow interaction feature and outbound passenger flow interaction feature; And the future time perception decoder processing unit is configured to construct a query vector according to the characteristics of the target time period, and obtain an inbound passenger flow predicted value and an outbound passenger flow predicted value according to the future query vector and the outbound passenger flow fusion characteristics.
9. A computer readable storage medium, characterized in that the computer readable storage medium stores a cross-border port entry and exit passenger flow prediction program, which when executed by a processor, implements the cross-border port entry and exit passenger flow prediction method of any one of claims 1 to 7.
10. A computer device comprising a computer readable storage medium, a processor and a cross-port entry and exit passenger flow prediction program stored in the computer readable storage medium, which when executed by the processor implements the cross-port entry and exit passenger flow prediction method of any one of claims 1 to 7.

Description

Method, device, storage medium and computer equipment for predicting cross-border port entry and exit passenger flow Technical Field The application belongs to the technical field of information, and particularly relates to a cross-border port entrance and exit passenger flow prediction method, a prediction device, a computer readable storage medium and computer equipment. Background The cross-border port is an important node of regional traffic, and accurate prediction of the outbound and inbound passenger flows is important for optimizing resource allocation (such as customs clearance window scheduling and inspection personnel scheduling), improving cross-border travel experience (reducing congestion waiting) and guaranteeing public safety (emergency people flow diversion). Compared with a general traffic scene, the port passenger flow not only shows regularity of intra-day and pericycle, but also is influenced by superposition of external factors such as holidays, weather, large events, economy and the like in different areas, and the external influence factors are relatively more, so that stronger randomness and uncertainty are shown, and the prediction difficulty is remarkably increased. Passenger flow demand prediction has been studied extensively in the traffic and travel fields. In these fields, early studies relied primarily on CNN, which is good at capturing local spatio-temporal features, and RNN/LSTM, which is effective in modeling sequence dependencies, particularly for time series data. To extract features of different temporal granularity, subsequently, RNN/LSTM structures of encoder-Decoder (Encoder-Decoder) are widely used, whose encoder maps input sequences into vector representations, and the Decoder directly generates target sequences, enables end-to-end prediction, and exhibits leading performance therein. However, RNNs are prone to gradient vanishing problems when dealing with long sequences. In recent years, a transducer remarkably improves the modeling capability of long sequence dependence through a self-attention mechanism, and higher precision and robustness are shown in a time sequence prediction task. Some studies introduce external information (such as holidays, weather, and special events) into the predictive model to improve the accuracy of the predictions. But cross-border port passenger flows have different characteristics than conventional traffic predictions. For example, port traffic exhibits both intra-day and peri-periodic regularity, and is affected by trans-regional holidays, while there is a significant time-lag correlation between inbound and outbound traffic. The specific expression is as follows: (1) The double specificity of holiday influence is that transregional superposition and transperiod extension port passenger flow are simultaneously influenced by Chinese and western holidays, dislocation and false release often occur, and asynchronous impact is formed. For example, on the Shenzhen port, a large number of residents go to Hongkong travel in the spring festival, while on the Christmas festival, the residents in the port and the Australia concentrate on shopping, and due to the unsynchronized holidays, alternating peaks of passenger flow occur in different directions. Meanwhile, holiday effects can also extend to front and rear time periods, such as cross-border purchasing tide before national celebration festival and post-festival return peak, and continuous fluctuation of 'front-middle-rear of festival' is often formed. The holiday effect across areas and periods is obviously different from a single-area short-period holiday influence mode in the traditional scene. (2) Time lag correlation of incoming and outgoing traffic, unlike the "time period multidimensional correlation" common in traffic prediction (e.g., real-time coupling of road traffic and speed), there is a significant time difference between the entry and exit of the port. For example, the tourists outside the spring festival are concentrated to enter the environment, a large-scale exit peak is usually formed one to two weeks after the holiday is finished, and the hysteresis mode of 'first in/out and later out/in' breaks through the assumption of 'real-time synchronous linkage' in the traditional prediction. The characteristics make the existing passenger flow prediction model architecture difficult to be directly applied to port passenger flow prediction. Disclosure of Invention The method solves the technical problems of double specificity of holiday influence, time lag correlation of the incoming and outgoing passenger flows and the like in a cross-border port incoming and outgoing scene, so as to improve the accuracy of passenger flow prediction. The application provides a cross-border port entry and exit passenger flow prediction method, which comprises the following steps: The construction stage of the input features comprises the steps of carrying out parameterization processing on the