A time-aware urban area representation generation method, system, terminal and storage medium based on dynamic condition modulation

Abstract

The application belongs to the technical field of data processing, and discloses a time-aware urban area representation generation method, system, terminal and storage medium based on dynamic condition modulation, which comprises the following steps: acquiring static urban data and OD flow data of an urban area; performing static feature extraction on the static urban data and fusing the extracted static features to obtain a regional static semantic representation; constructing a regional dynamic state sequence according to the OD flow data and encoding the regional dynamic state sequence to obtain a regional dynamic condition; generating scaling parameters and offset parameters according to the regional dynamic condition, and performing dimension-by-dimension modulation on the regional static semantic representation according to the scaling parameters and offset parameters to obtain a time-aware urban area representation. The regional static semantic representation is modulated from static to dynamic by using the regional dynamic condition, so that the generated time-aware urban area representation can reflect the long-term stable attributes of the region and the dynamic evolution law of the region.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, system, terminal, and computer-readable storage medium for generating time-aware urban area representations based on dynamic condition modulation. Background Technology

[0002] As urbanization continues, the functional structure, physical environment, population activities, and traffic flows within urban areas exhibit complex spatiotemporal characteristics. Tasks such as population estimation, traffic demand forecasting, pollution monitoring, meteorological analysis, vegetation change analysis, and regional functional identification all rely on accurate representation of the urban area's state. Urban area representation refers to processing multi-source urban data related to the urban area as a spatial unit to generate a vectorized representation that reflects the area's comprehensive attributes.

[0003] Existing methods for generating urban regional representations typically focus on learning regional embeddings using static urban data such as points of interest, street view images, remote sensing images, base maps, and road networks. While these methods can reflect the long-term stable functional attributes and spatial environmental characteristics of a region relatively well, they often generate a fixed representation for each region. This makes it difficult to depict the dynamic semantic differences in the same region at different times due to changes in population flow and activity status. In other words, the generated urban regional representations cannot comprehensively reflect both the long-term stable attributes and dynamic evolution patterns of the region.

[0004] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0005] The main objective of this invention is to provide a method, system, terminal, and computer-readable storage medium for generating time-aware urban area representations based on dynamic condition modulation, aiming to solve the problem that the urban area representations generated by traditional urban area representation generation methods cannot synergistically reflect the long-term stable attributes and dynamic evolution patterns of the region.

[0006] To achieve the above-mentioned objective, this invention provides a method for generating time-aware urban area representations based on dynamic condition modulation, the method comprising: The target city area is divided into multiple city regions, and static city data and OD flow data corresponding to each city region are obtained; A static feature extractor is used to extract static features from the static city data to obtain static features of multiple modalities. The static features of multiple modalities are then fused to obtain a regional static semantic representation. Based on the OD flow data, the historical flow status of the urban area is modeled to obtain a regional dynamic state sequence, and the regional dynamic state sequence is encoded to obtain regional dynamic conditions. Based on the dynamic conditions of the region, scaling parameters and offset parameters are generated through a multilayer perceptron network. Then, based on the scaling parameters and offset parameters, the static semantic representation of the region is modulated dimension by dimension to obtain a time-aware urban region representation.

[0007] Optionally, the step of dividing the target city area into multiple city regions and obtaining static city data and OD flow data corresponding to each city region specifically includes: The target city area is divided into multiple non-overlapping city regions: ; in, Indicates the target city range. Indicates the first Each city area Indicates urban area index, This indicates the total number of urban areas; Obtain static urban data of different modalities corresponding to each of the urban areas, wherein the static urban data includes point of interest category distribution, street view images, remote sensing images and open street map base maps; Obtain OD flow data between different urban areas at multiple time slices, wherein the OD flow data includes departure area identifier, arrival area identifier, time slice identifier, and flow intensity; The static urban data is used to describe the long-term stable attributes of the urban area, while the OD flow data is used to describe the activity status of the urban area.

[0008] Optionally, the static feature extractor includes a graph attention network encoder, a street view visual encoder, a remote sensing visual encoder, and a base map encoder, and the static features of multiple modalities include interest point modality features, street view visual features, remote sensing visual features, and base map features; The process involves using a static feature extractor to extract static features from the static city data, obtaining static features of multiple modalities, and then fusing these static features to obtain a regional static semantic representation. Specifically, this includes: The distribution of interest point categories is vectorized to obtain an interest point category distribution vector. Based on this vector, the functional similarity of interest points between any two urban areas is calculated. An interest point functional similarity map is constructed based on this similarity. A pre-trained graph attention network encoder is used to extract static features from the interest point functional similarity map to obtain the modal features of the interest points. ; ; in, Indicates urban area With urban areas Functional similarity of points of interest between them Indicates urban area index, and Representing urban areas and urban areas Interest point category distribution vector This represents the transpose of a vector. The norm of a vector Indicates urban area Point of interest modal features This represents a graph attention network encoder. Similarity graph representing points of interest Functionally similar adjacency matrix; The street view visual features are obtained by extracting static features from the street view image using the pre-trained street view visual encoder: ; in, Indicates urban area Street scene visual features Indicates a street view visual encoder. Indicates urban area Street view images; The remote sensing visual features are obtained by extracting static features from the remote sensing image using the pre-trained remote sensing visual encoder: ; in, Indicates urban area Remote sensing visual features, Indicates a remote sensing visual encoder. Indicates urban area Remote sensing images; The pre-trained basemap encoder is used to extract static features from the open street map basemap to obtain the basemap features: ; in, Indicates urban area Base map features, Indicates the base map encoder. Indicates urban area Open street map base map; The interest point modal features, street view visual features, remote sensing visual features, and base map features are sequentially processed by mapping, stitching, fusion transformation, and normalization to obtain a static semantic representation of the region.

[0009] Optionally, the step of sequentially mapping, stitching, fusion transformation, and normalization of the interest point modal features, street view visual features, remote sensing visual features, and base map features to obtain a static semantic representation of the region specifically includes: The point of interest modal features, the street view visual features, the remote sensing visual features, and the base map features are each mapped to the same dimension to obtain the mapped point of interest modal features, the mapped street view visual features, the mapped remote sensing visual features, and the mapped base map features. The mapped point of interest modal features, mapped street view visual features, mapped remote sensing visual features, and mapped base map features are stitched together to obtain the first fused feature; The first fusion feature is input into the fusion mapping layer for fusion transformation processing to obtain the second fusion feature; The second fused feature is normalized to obtain the region's static semantic representation: ; in, Indicates urban area The region's static semantic representation, This indicates normalization processing. Indicates the fusion mapping layer, This indicates splicing / joining. Representing urban areas The mapped point of interest modal features, mapped street view visual features, mapped remote sensing visual features, and mapped base map features.

[0010] Optionally, the step of modeling the historical flow status of the urban area based on the OD flow data to obtain a regional dynamic state sequence, and encoding the regional dynamic state sequence to obtain regional dynamic conditions, specifically includes: A flow matrix is ​​constructed based on the OD flow data, and flow statistics are calculated based on the flow matrix, wherein the flow statistics include outbound flow, arrival flow, total interaction flow, and net flow. ; ; ; ; in, Indicates urban area In time slice Departure flow, Indicates urban area In time slice The arrival flow, Indicates urban area In time slice Total interaction traffic Indicates urban area In time slice Net flow Indicates time slice The flow matrix of the first line, number Column elements; The flow statistics are concatenated to obtain the dynamic state vector: ; in, Indicates urban area In time slice The dynamic state vector; Use a length of The dynamic state vector is acquired through a continuous time window to obtain the regional dynamic state sequence: ; in, Indicates urban area In time slice The regional dynamic state sequence, Indicates urban area In time slice The dynamic state vector, Indicates urban area In time slice The dynamic state vector, It is a constant; The dynamic state sequence of the region is encoded using a time-series encoder to obtain the dynamic conditions of the region: ; in, Indicates urban area In time slice Regional dynamic conditions, This indicates a timing encoder.

[0011] Optionally, the multilayer perceptron network includes a first multilayer perceptron network for generating scaling parameters and a second multilayer perceptron network for generating offset parameters; The process involves generating scaling and offset parameters through a multilayer perceptron network based on the dynamic conditions of the region, and then modulating the static semantic representation of the region dimension-by-dimensional based on the scaling and offset parameters to obtain a time-aware urban region representation. Specifically, this includes: The dynamic conditions of the region are input into the first multilayer perceptron network and the second multilayer perceptron network, respectively. The first multilayer perceptron network and the second multilayer perceptron network output scaling parameters and offset parameters, respectively, wherein the dimensions of the scaling parameters and the offset parameters are consistent with the dimensions of the static semantic representation of the region. ; ; in, and Representing urban areas In time slice The scaling and offset parameters, and These represent the first multilayer perceptron network and the second multilayer perceptron network, respectively. The scaling parameter is used to perform element-wise multiplication on the static semantic representation of the region to obtain a scaled static semantic representation of the region. The offset parameter is then added to the scaled static semantic representation of the region to obtain a dynamically modulated intermediate representation. Finally, the dynamically modulated intermediate representation is normalized to obtain a time-aware urban region representation. ; in, Indicates urban area In time slice Time-aware urban area representation. This indicates element-wise multiplication.

[0012] Optionally, the step of generating scaling and offset parameters through a multilayer perceptron network based on the dynamic conditions of the region, and performing dimension-wise modulation on the static semantic representation of the region based on the scaling and offset parameters to obtain a time-aware urban region representation, further includes: The time-perceived urban area representation of the urban area in the time slice is used as the anchor point. The time-perceived urban area representation of the urban area in the previous time slice and the next time slice is used as the positive sample. The time-perceived urban area representation of other urban areas or other non-adjacent time slices in the training batch is used as the negative sample. A dynamic contrastive learning loss function is constructed. By minimizing this loss function, the region dynamic conditional coding module and the static-to-dynamic modulation module are jointly optimized to reduce the semantic similarity between the time-aware urban region representations in the positive samples and increase the semantic similarity between the time-aware urban region representations in the negative samples. ; in, This represents the dynamic contrastive learning loss function. Indicates urban area In time slice Positive samples, This represents the time-aware urban region representation in the positive samples. Indicates negative samples Time-aware urban area representation in [the context of] ... Represents the semantic similarity function. Indicates the temperature coefficient; The regional dynamic condition encoding module is used to encode the regional dynamic state sequence to obtain regional dynamic conditions. The static-to-dynamic modulation module is used to generate scaling parameters and offset parameters through a multilayer perceptron network according to the regional dynamic conditions, and to perform dimension-by-dimensional modulation on the regional static semantic representation according to the scaling parameters and the offset parameters to obtain a time-aware urban regional representation.

[0013] To achieve the above-mentioned objectives, the present invention also provides a time-aware urban area representation generation system based on dynamic condition modulation, the time-aware urban area representation generation system based on dynamic condition modulation comprising: The data acquisition module is used to divide the target city area into multiple city regions and acquire static city data and OD flow data corresponding to each city region. The feature extraction and fusion module is used to extract static features from the static city data using a static feature extractor to obtain static features of multiple modalities, and to fuse the static features of multiple modalities to obtain a regional static semantic representation. The modeling and coding module is used to model the historical flow status of the urban area based on the OD flow data, obtain a regional dynamic state sequence, and encode the regional dynamic state sequence to obtain regional dynamic conditions. The static-to-dynamic modulation module is used to generate scaling parameters and offset parameters through a multilayer perceptron network according to the dynamic conditions of the region, and to perform dimension-by-dimensional modulation on the static semantic representation of the region according to the scaling parameters and the offset parameters to obtain a time-aware urban region representation.

[0014] To achieve the above-mentioned objectives, the present invention also provides a terminal, the terminal comprising: a memory, a processor, and a time-aware urban area representation generation program based on dynamic condition modulation stored in the memory and executable on the processor, wherein when the time-aware urban area representation generation program based on dynamic condition modulation is executed by the processor, it implements the steps of the time-aware urban area representation generation method based on dynamic condition modulation as described above.

[0015] To achieve the above-mentioned objectives, the present invention also provides a computer-readable storage medium storing a time-aware urban area representation generation program based on dynamic condition modulation. When the time-aware urban area representation generation program based on dynamic condition modulation is executed by a processor, it implements the steps of the time-aware urban area representation generation method based on dynamic condition modulation as described above.

[0016] In this invention, the target city area is divided into multiple city regions, and static city data and OD (Origin-Destination) flow data corresponding to each city region are acquired. A static feature extractor is used to extract static features from the static city data to obtain static features of multiple modalities. These static features are then fused to obtain a static semantic representation of the region. Based on the OD flow data, the historical flow state of the city region is modeled to obtain a dynamic state sequence of the region. This dynamic state sequence is then encoded to obtain dynamic conditions for the region. Based on these dynamic conditions, scaling parameters and offset parameters are generated using a multilayer perceptron network. The static semantic representation of the region is then modulated dimension-wise based on these scaling parameters and offset parameters to obtain a time-aware urban region representation. This invention combines multi-source static urban data with OD flow data to construct a time-aware urban regional representation framework based on "static semantic foundation - dynamic condition modulation". This framework enables the resulting time-aware regional representation (i.e., time-aware urban regional representation) to synergistically reflect the long-term stable attributes and dynamic evolution patterns of the region. It solves the problem that traditional urban regional representations can only generate fixed regional representations and are difficult to depict the dynamic state changes of the same region under different time slices. Through static-to-dynamic modulation based on dynamic conditions and combined with dynamic comparative learning of adjacent time slices, the resulting time-aware regional representation maintains both regional static semantic consistency and local temporal continuity, thereby providing a unified representation for dynamic urban tasks such as public transport demand forecasting and taxi demand forecasting. Attached Figure Description

[0017] Figure 1 This is a flowchart of a preferred embodiment of the time-aware urban area representation generation method based on dynamic condition modulation of the present invention; Figure 2 This is another flowchart of a preferred embodiment of the time-aware urban area representation generation method based on dynamic condition modulation of the present invention; Figure 3 This is a structural diagram of a preferred embodiment of the time-aware urban area representation generation system based on dynamic condition modulation of the present invention; Figure 4 This is a structural diagram of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] As urbanization continues, the functional structure, physical environment, population activities, and traffic flows within urban areas exhibit complex spatiotemporal characteristics. Tasks such as population estimation, traffic demand forecasting, pollution monitoring, meteorological analysis, vegetation change analysis, and regional functional identification all rely on accurate representation of the urban area's state. Urban area representation refers to processing multi-source urban data related to the urban area as a spatial unit to generate a vectorized representation that reflects the area's comprehensive attributes.

[0020] Existing methods for generating urban regional representations typically focus on learning regional embeddings using static urban data such as points of interest, street view images, remote sensing images, base maps, and road networks. While these methods can effectively reflect the long-term stable functional attributes and spatial environmental characteristics of a region, they often generate a fixed representation for each region, making it difficult to depict the dynamic semantic differences in the same region at different times due to changes in population flow and activity patterns. For example, a residential area may be dominated by departures during the morning rush hour, while at night it may be dominated by arrivals; a commercial area may also exhibit different activity intensities and interaction patterns on weekdays and holidays.

[0021] On the other hand, some spatiotemporal prediction methods directly model tasks based on historical traffic flow or time-series observation data, which can capture dynamic changes. However, they are usually trained for specific prediction tasks and are difficult to form a transferable and reusable time-aware regional representation. At the same time, if only OD flow data is relied upon, the long-term stable functional and physical semantics of the region are easily ignored, making it difficult for the model to explain the relationship between dynamic states and inherent regional attributes.

[0022] In summary, existing urban regional representation generation methods fail to adequately reflect both the long-term stable attributes and dynamic evolution patterns of a region.

[0023] To address the aforementioned technical problems, this invention provides a time-aware urban region representation generation method based on dynamic condition modulation. The method divides the target city area into multiple urban regions and acquires static urban data and OD flow data corresponding to each urban region. A static feature extractor is used to extract static features from the static urban data, obtaining static features of multiple modalities. These static features are then fused to obtain a static semantic representation of the region. Based on the OD flow data, the historical flow state of the urban region is modeled to obtain a dynamic state sequence of the region. This dynamic state sequence is then encoded to obtain dynamic conditions for the region. Based on these dynamic conditions, scaling parameters and offset parameters are generated using a multilayer perceptron network. The static semantic representation of the region is then modulated dimension-wise based on these scaling parameters and offset parameters to obtain a time-aware urban region representation. This invention combines multi-source static urban data with OD flow data to construct a time-aware urban regional representation framework based on "static semantic foundation—dynamic condition modulation." This framework enables the resulting time-aware regional representation to synergistically reflect the long-term stable attributes and dynamic evolution patterns of the region. It solves the problem that traditional urban regional representations can only generate fixed regional representations and are unable to depict the dynamic state changes of the same region under different time slices. Through static-to-dynamic modulation based on dynamic conditions, and combined with dynamic comparative learning of adjacent time slices, the resulting time-aware regional representation maintains both the consistency of static semantics and local temporal continuity, thereby providing a unified representation for dynamic urban tasks such as public transport demand forecasting and taxi demand forecasting.

[0024] The application content will be further explained below with reference to the accompanying drawings and the description of the embodiments.

[0025] A preferred embodiment of the present invention is a time-aware urban area representation generation method based on dynamic condition modulation, such as... Figure 1 and Figure 2 As shown, it specifically includes: S1. Divide the target city area into multiple city regions and obtain the static city data and OD flow data corresponding to each city region.

[0026] In one implementation of this embodiment, dividing the target city area into multiple city regions and obtaining static city data and OD flow data corresponding to each city region specifically includes: The target city area is divided into multiple non-overlapping city regions: ; in, Indicates the target city range (city or city cluster). Indicates the first Each city area Indicates urban area index, This indicates the total number of urban areas; Obtain static urban data of different modalities corresponding to each of the urban areas, wherein the static urban data includes point of interest category distribution, street view images, remote sensing images and open street map base maps; Obtain OD flow data between different urban areas at multiple time slices, wherein the OD flow data includes departure area identifier, arrival area identifier, time slice identifier, and flow intensity; The static urban data is used to describe the long-term stable attributes of the urban area, while the OD flow data is used to describe the activity status of the urban area.

[0027] This invention provides a time-aware urban area representation generation method based on dynamic condition modulation, including: urban area division and data acquisition; static urban data feature extraction and construction of regional static semantic representation; construction of regional dynamic state sequence based on OD flow data; regional dynamic condition encoding; static-to-dynamic modulation based on regional dynamic conditions; dynamic comparative learning based on adjacent time slices; time-aware urban area representation output and effectiveness evaluation.

[0028] In this invention, urban area division and data acquisition specifically include: given a target city or urban agglomeration, dividing the target research area into... Each city area is a non-overlapping urban region unit. To obtain static urban data (i.e., static urban multi-source data) and OD flow data that meet the requirements for subsequent processing, the specific content includes: Get city area The corresponding four types of static urban data include point-of-interest (POI) category distribution, street view images, remote sensing images, and OpenStreetMap (OSM) base maps. Among them, POI category distribution is used to describe the functional structure of the region, street view images are used to describe the ground visual environment of the region, remote sensing images are used to describe the aerial spatial morphology of the region, and OpenStreetMap base maps are used to describe the map semantics and built environment structure of the region.

[0029] Acquire OD (Original Departure) flow data between different urban areas within the target study area across multiple discrete time slices. OD flow data refers to the inter-regional flow records from one urban area to another within a specified time slice, used to describe the interaction of population travel or activities between urban areas. OD flow data includes departure area identifier, arrival area identifier, time slice identifier, and flow intensity. For any given time slice... Construct an OD flow matrix based on OD flow data. Specifically, all urban areas are arranged according to their area codes as follows: ,in, Indicates the first urban area. This indicates the second city area. Indicates the first Each city region is represented by a matrix with departure regions as rows and arrival regions as columns, and time slices are defined accordingly. OD flow records with the same departure and arrival areas are summarized to obtain matrix elements. , where matrix elements Indicates time slice From the urban area To urban areas The flow intensity, It represents the set of real numbers.

[0030] Through the above steps, static urban data describing the long-term stable attributes of the region and OD flow data describing the activity status of the region are obtained.

[0031] S2. A static feature extractor is used to extract static features from the static city data to obtain static features of multiple modalities. The static features of multiple modalities are then fused to obtain a regional static semantic representation.

[0032] In one implementation of this embodiment, the static feature extractor includes a graph attention network encoder, a street view visual encoder, a remote sensing visual encoder, and a base map encoder, and the static features of multiple modalities include interest point modal features, street view visual features, remote sensing visual features, and base map features. The process involves using a static feature extractor to extract static features from the static city data, obtaining static features of multiple modalities, and then fusing these static features to obtain a regional static semantic representation. Specifically, this includes: The distribution of interest point categories is vectorized to obtain an interest point category distribution vector. Based on this vector, the functional similarity of interest points between any two urban areas is calculated. An interest point functional similarity map is constructed based on this similarity. A pre-trained graph attention network encoder is used to extract static features from the interest point functional similarity map to obtain the modal features of the interest points. ; ; in, Indicates urban area With urban areas Functional similarity of points of interest between them Indicates urban area index, and Representing urban areas and urban areas Interest point category distribution vector This represents the transpose of a vector. The norm of a vector Indicates urban area Point of interest modal features (GraphAttention Network v2) represents the graph attention network encoder. Similarity graph representing points of interest The functional similarity adjacency matrix (i.e., the functional similarity adjacency matrix constructed from the functional similarity of interest points), where... ; The street view visual features are obtained by extracting static features from the street view image using the pre-trained street view visual encoder: ; in, Indicates urban area Street scene visual features This refers to the street view visual encoder. In this embodiment, the street view visual encoder uses the CLIP (Contrastive Language-Image Pre-training) ViT-L / 14 (Vision Transformer-Large / 14, a large visual Transformer model / 14-slice version) pre-trained encoder. Indicates urban area Street view images; The remote sensing visual features are obtained by extracting static features from the remote sensing image using the pre-trained remote sensing visual encoder: ; in, Indicates urban area Remote sensing visual features, This refers to the remote sensing visual encoder. In this embodiment, the remote sensing visual encoder uses the RemoteCLIP (Remote Sensing Contrastive Language-Image Pre-training) ViT-L / 14 pre-trained encoder. Indicates urban area Remote sensing images; The pre-trained basemap encoder is used to extract static features from the open street map basemap to obtain the basemap features: ; in, Indicates urban area Base map features, This refers to the base map encoder. In this embodiment, the base map encoder is a ViT-B / 16 (VisionTransformer-Base / 16, a 16-slice version of the Visual Transformer base model) encoder trained based on Masked Autoencoder (MAE). Indicates urban area Open street map base map; The interest point modal features, street view visual features, remote sensing visual features, and base map features are sequentially processed by mapping, stitching, fusion transformation, and normalization to obtain a static semantic representation of the region.

[0033] In one implementation of this embodiment, the step of sequentially mapping, stitching, fusion transformation, and normalization of the interest point modal features, the street view visual features, the remote sensing visual features, and the base map features to obtain a static semantic representation of the region specifically includes: The point of interest modal features, the street view visual features, the remote sensing visual features, and the base map features are each mapped to the same dimension to obtain the mapped point of interest modal features, the mapped street view visual features, the mapped remote sensing visual features, and the mapped base map features. The mapped point of interest modal features, mapped street view visual features, mapped remote sensing visual features, and mapped base map features are stitched together to obtain the first fused feature; The first fusion feature is input into the fusion mapping layer for fusion transformation processing to obtain the second fusion feature; The second fused feature is normalized to obtain the region's static semantic representation: ; in, Indicates urban area The region's static semantic representation, This indicates normalization processing. This represents the fusion mapping layer, used to fuse and transform the concatenated multimodal features; This indicates splicing / joining. Representing urban areas The mapped point of interest modal features, mapped street view visual features, mapped remote sensing visual features, and mapped base map features.

[0034] In this invention, the extraction of static urban data features and the construction of regional static semantic representations specifically include: based on the static urban data obtained in the above steps, extracting static features (i.e., static semantic features) of different modalities for each urban region, and constructing a regional static semantic representation, specifically including: For the distribution of interest point categories, GATv2 is used for encoding to obtain the modal features of interest points; for street view images, CLIP ViT-L / 14 pre-trained encoder is used to extract street view visual features; for remote sensing images, RemoteCLIPViT-L / 14 pre-trained encoder is used to extract remote sensing visual features; for OpenStreetMap base maps, ViT-B / 16 encoder trained based on MaskedAutoencoder is used to extract base map features.

[0035] Modal features corresponding to point of interest category distribution, street view images, remote sensing images, and OpenStreetMap base maps. , , and Mapping them to the same dimension respectively yields... , , and The four modal features with unified dimensions are concatenated, and a static semantic representation of the region is obtained by fusing the mapping layer and normalization operation.

[0036] Through the above steps, each urban area obtains a static semantic representation that reflects its long-term stable functional attributes, visual environment, and spatial structure characteristics.

[0037] S3. Based on the OD flow data, model the historical flow status of the urban area to obtain a regional dynamic state sequence, and encode the regional dynamic state sequence to obtain regional dynamic conditions.

[0038] In one implementation of this embodiment, the step of modeling the historical flow status of the urban area based on the OD flow data to obtain a regional dynamic state sequence, and encoding the regional dynamic state sequence to obtain regional dynamic conditions, specifically includes: A flow matrix is ​​constructed based on the OD flow data, and flow statistics are calculated based on the flow matrix, wherein the flow statistics include outbound flow, arrival flow, total interaction flow, and net flow. ; ; ; ; in, Indicates urban area In time slice Departure flow, Indicates urban area In time slice The arrival flow, Indicates urban area In time slice Total interaction traffic Indicates urban area In time slice Net flow Indicates time slice The flow matrix of the first line, number Column elements; The flow statistics are concatenated to obtain the dynamic state vector: ; in, Indicates urban area In time slice The dynamic state vector; Use a length of The dynamic state vector is acquired through a continuous time window to obtain the regional dynamic state sequence: ; in, Indicates urban area In time slice The regional dynamic state sequence, Indicates urban area In time slice The dynamic state vector, Indicates urban area In time slice The dynamic state vector, It is a constant; The dynamic state sequence of the region is encoded using a time-series encoder to obtain the dynamic conditions of the region: ; in, Indicates urban area In time slice Regional dynamic conditions, This refers to a timing encoder. In this embodiment, the timing encoder uses a gated recurrent unit (GRU). The GRU encodes a continuous sequence of OD dynamic states in chronological order and uses the last hidden state as the region dynamic condition. .

[0039] In this invention, the construction of regional dynamic state sequences based on OD flow data specifically includes: based on the acquired OD flow data, analyzing the dynamic state sequences of each urban area at a target time slice. Modeling the historical flow patterns in the vicinity to construct a regional dynamic state sequence includes the following: For the region and time slices According to the OD flow matrix Calculate regional-level flow statistics, including outbound flow, arrival flow, total interaction flow, and net flow; then combine these flow statistics to form a regional flow rate. In time slice The dynamic state vector; further select a length of Constructing a region using a continuous time window. The historical dynamic state sequence; through the above steps, the OD dynamic state sequence corresponding to each region in the target time slice is obtained.

[0040] In this invention, the regional dynamic condition encoding specifically includes: generating regional dynamic conditions based on the obtained historical dynamic state sequence using a time-series encoder, specifically including: ... The input timing encoder performs dynamic information aggregation to obtain the region. In time slice Dynamic conditions Regional dynamic conditions Used to represent a region In the target time slice The nearby flow activity state is used as a conditional input for the subsequent static semantic representation of the modulation region.

[0041] S4. Based on the dynamic conditions of the region, scaling parameters and offset parameters are generated through a multilayer perceptron network, and the static semantic representation of the region is modulated dimension by dimension according to the scaling parameters and the offset parameters to obtain a time-aware urban region representation.

[0042] In one implementation of this embodiment, the multilayer perceptron network includes a first multilayer perceptron network for generating scaling parameters and a second multilayer perceptron network for generating offset parameters. The process involves generating scaling and offset parameters through a multilayer perceptron network based on the dynamic conditions of the region, and then modulating the static semantic representation of the region dimension-by-dimensional based on the scaling and offset parameters to obtain a time-aware urban region representation. Specifically, this includes: The dynamic conditions of the region are input into the first multilayer perceptron network and the second multilayer perceptron network, respectively. The first multilayer perceptron network and the second multilayer perceptron network output scaling parameters and offset parameters, respectively, wherein the dimensions of the scaling parameters and the offset parameters are consistent with the dimensions of the static semantic representation of the region. ; ; in, and Representing urban areas In time slice The scaling and offset parameters, and ... The scaling parameter is used to perform element-wise multiplication on the static semantic representation of the region to obtain a scaled static semantic representation of the region. The offset parameter is then added to the scaled static semantic representation of the region to obtain a dynamically modulated intermediate representation. Finally, the dynamically modulated intermediate representation is normalized to obtain a time-aware urban region representation. ; in, Indicates urban area In time slice Time-aware urban area representation. This indicates element-wise multiplication.

[0043] In this invention, the static-to-dynamic modulation based on regional dynamic conditions specifically includes: based on the obtained regional static semantic representation and regional dynamic conditions Generate time-aware urban area representations, including: Regional dynamic conditions Input two multilayer perceptron networks to generate scaling parameters consistent with the dimension of the region's static semantic representation. and offset parameters Then, using scaling parameters Static semantic representation of regions Element-wise multiplication is performed to obtain the scaled static semantic representation of the region. Then, the offset parameter is... The scaled region static semantic representation is added to obtain the dynamically modulated intermediate representation. Finally, the dynamically modulated intermediate representation is normalized to obtain the region. In time slice Time-aware urban area representation.

[0044] Through the above modulation process, the regional dynamic conditions It does not directly replace the static semantic representation of the region. Instead, it controls the expression of each dimension of features in the static semantic representation of the region in the current time slice by scaling parameters and offset parameters, so that the same region generates different time-aware urban region representations in different time slices, while preserving its long-term stable regional static semantics.

[0045] In one implementation of this embodiment, the process of generating scaling and offset parameters through a multilayer perceptron network based on the dynamic conditions of the region, and then performing dimension-wise modulation on the static semantic representation of the region based on the scaling and offset parameters to obtain a time-aware urban region representation, further includes: The time-perceived urban area representation of the urban area in the time slice is used as the anchor point. The time-perceived urban area representation of the urban area in the previous time slice and the next time slice is used as the positive sample. The time-perceived urban area representation of other urban areas or other non-adjacent time slices in the training batch is used as the negative sample. A dynamic contrastive learning loss function is constructed. By minimizing this loss function, the region dynamic conditional coding module and the static-to-dynamic modulation module are jointly optimized to reduce the semantic similarity between the time-aware urban region representations in the positive samples and increase the semantic similarity between the time-aware urban region representations in the negative samples. ; in, This represents the dynamic contrastive learning loss function. Indicates urban area In time slice Positive samples (i.e., the same region surrounding the time slice) (the set of positive samples in the nearest time). This represents the time-aware urban area representation in the positive samples (i.e., the positive sample representation). This represents the set of candidate negative samples in the training batch. Indicates negative samples The time-aware urban area representation (i.e., candidate negative sample representation) in the context of time perception. Represents the semantic similarity function. Indicates the temperature coefficient; The regional dynamic condition encoding module is used to encode the regional dynamic state sequence to obtain regional dynamic conditions. The static-to-dynamic modulation module is used to generate scaling parameters and offset parameters through a multilayer perceptron network according to the regional dynamic conditions, and to perform dimension-by-dimensional modulation on the regional static semantic representation according to the scaling parameters and the offset parameters to obtain a time-aware urban regional representation.

[0046] In this invention, dynamic comparative learning based on adjacent time slices specifically includes: a time-aware urban area representation obtained from the "static-to-dynamic modulation based on regional dynamic conditions" step. A dynamic contrastive learning objective is constructed to jointly optimize the regional dynamic conditional coding module and the static-to-dynamic modulation module. Specifically, for the same region... The time-aware urban area representations of the preceding time slice, the target time slice, and the following time slice are obtained respectively. , and .in, As an anchor point representation and As a positive sample representation with local temporal continuity with the anchor point: ; This positive sample relationship is used to constrain the temporally perceived region representations of the same region to remain close in the feature space under adjacent time slices, thereby maintaining the local temporal continuity of the region state. At the same time, the temporally perceived region representations of other regions or other non-adjacent time slices in the training batch are used as candidate negative samples, enabling the model to distinguish region representations of different regions or different dynamic states.

[0047] During training, the learning loss is dynamically compared. Backpropagation to the regional dynamic condition coding module and the static-to-dynamic modulation module ， The dynamic conditions of the region obtained by GRU encoding It can reflect the dynamic continuity between adjacent time slices and make the modulation parameters (i.e., dynamic modulation parameters) reflect the dynamic continuity between adjacent time slices. and Capable of generating time-aware urban area representations with temporal distinctiveness and local continuity. .

[0048] By minimizing the dynamic contrastive learning loss, the time-aware region representations (positive sample representations) of adjacent time slices in the same region are brought closer together, while the time-aware region representations (candidate negative sample representations) of different regions or non-adjacent time slices are distinguished. This enables the time-aware region representations generated in the above steps to simultaneously possess regional static semantic consistency and local temporal continuity, and serve as input to the downstream task model.

[0049] In one implementation of this embodiment, the time-aware urban area representation generation method based on dynamic conditional modulation further includes: time-aware urban area representation output and effectiveness evaluation. The time-aware urban area representation obtained through dynamic contrastive learning is used as a unified representation result of the urban area in a time slice. The representation result is input into a multilayer perceptron (MLP) downstream task model for various urban sensing tasks. The effectiveness of the time-aware area representation is evaluated based on the prediction results of the downstream tasks.

[0050] This invention uses urban areas as the basic analysis unit, acquiring multi-source static urban data and OD flow data corresponding to the area; extracting static features of different modalities through a static feature extractor and fusing them to obtain a static semantic representation of the area; constructing a dynamic state sequence of the area based on OD flow data within a continuous time window and encoding it to obtain dynamic conditions of the area; generating scaling and offset parameters corresponding to the dimensions of the static semantic representation of the area based on the dynamic conditions of the area, and using the scaling and offset parameters to modulate the static semantic representation of the area dimension by dimension to obtain a time-aware urban area representation under the target time slice. This representation can be used for urban computing tasks such as traffic demand forecasting, urban vitality estimation, environmental indicator prediction, and regional function identification.

[0051] In summary, this invention decouples the static semantic representation of urban areas from the dynamic conditions constructed based on OD flow data. It utilizes these dynamic conditions to modulate the static semantic representation from static to dynamic, thereby generating time-aware urban area representations across different time slices. Specifically, this invention combines multi-source static urban data with OD flow data to construct a "static semantic foundation—dynamic condition modulation" time-aware urban area representation framework. This allows the generated time-aware area representation to synergistically represent both the long-term stable attributes and dynamic patterns of the region, solving the problem that traditional urban area representations can only generate fixed regional representations and struggle to depict the dynamic state changes of the same region across different time slices. Furthermore, this invention combines the dynamic conditions obtained from OD data with a modulation mechanism. By generating scaling and offset parameters through dynamic conditions, it modulates the static semantic representation of the region dimensionally. Combined with dynamic comparative learning of adjacent time slices, the resulting time-aware area representation maintains both the consistency of the static semantic representation and local temporal continuity, thus providing a unified representation for dynamic urban tasks such as public transport demand forecasting and taxi demand forecasting.

[0052] In other implementations of this embodiment, the static semantic representation (i.e., regional static semantic representation) construction method of this invention can be extended from simplified multimodal feature concatenation to a multimodal Transformer fusion method, and further incorporates random modal masking and static contrastive learning to enhance the consistency between different static modalities and robustness under missing modal conditions. The dynamic conditions (i.e., regional dynamic conditions) construction method of this invention can further incorporate temporal context information on top of using only OD flow data, and can expand the positive samples of adjacent time slices into a set of positive samples composed of adjacent time slices, time slices with the same period, and time slices with similar OD states, to enhance the ability of time-aware regional representation to characterize periodic urban activities and similar dynamic states.

[0053] Furthermore, based on the aforementioned time-aware urban area representation generation method based on dynamic condition modulation, this invention also provides a time-aware urban area representation generation system based on dynamic condition modulation, wherein a preferred embodiment of the time-aware urban area representation generation system based on dynamic condition modulation is as follows: Figure 3 As shown, it specifically includes: The data acquisition module 01 is used to divide the target city area into multiple city regions and acquire static city data and OD flow data corresponding to each city region. The feature extraction and fusion module 02 is used to extract static features from the static city data using a static feature extractor, obtain static features of multiple modalities, and fuse the static features of multiple modalities to obtain a regional static semantic representation. The modeling and coding module 03 is used to model the historical flow status of the urban area based on the OD flow data to obtain a regional dynamic state sequence, and to encode the regional dynamic state sequence to obtain regional dynamic conditions. The modeling and coding module 03 includes a regional dynamic condition coding module. The static-to-dynamic modulation module 04 is used to generate scaling parameters and offset parameters through a multilayer perceptron network according to the dynamic conditions of the region, and to perform dimension-by-dimensional modulation on the static semantic representation of the region according to the scaling parameters and the offset parameters to obtain a time-aware urban region representation.

[0054] Furthermore, based on the aforementioned method and system for generating time-aware urban area representations based on dynamic condition modulation, this invention also provides a terminal, wherein a preferred embodiment of the terminal is as follows: Figure 4 As shown, it specifically includes a processor 10, a memory 20, and a display 30. Figure 4 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0055] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, or flash card. Furthermore, the memory 20 may include both internal and external storage units. The memory 20 is used to store application software and various types of data installed on the terminal, such as the terminal's program code. The memory 20 may also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores a time-aware urban area representation generation program 40 based on dynamic condition modulation, which can be executed by the processor 10 to implement the steps of the time-aware urban area representation generation method based on dynamic condition modulation in this application.

[0056] In some embodiments, processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in memory 20 or process data, such as executing a time-aware urban area representation generation program 40 based on dynamic condition modulation.

[0057] In some embodiments, display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. Display 30 is used to display information on the terminal and to display a visual user interface.

[0058] In one embodiment, when the processor 10 executes the time-aware urban area representation generation program 40 based on dynamic condition modulation in the memory 20, it implements the steps of the time-aware urban area representation generation method based on dynamic condition modulation as described above.

[0059] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a time-aware urban area representation generation program based on dynamic condition modulation, and when the time-aware urban area representation generation program based on dynamic condition modulation is executed by a processor, it implements the steps of the time-aware urban area representation generation method based on dynamic condition modulation as described above.

[0060] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.

[0061] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The computer-readable storage medium can be a memory, magnetic disk, optical disk, etc.

[0062] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for generating time-aware urban area representations based on dynamic condition modulation, characterized in that, The time-aware urban area representation generation method based on dynamic condition modulation includes: The target city area is divided into multiple city regions, and static city data and OD flow data corresponding to each city region are obtained; A static feature extractor is used to extract static features from the static city data to obtain static features of multiple modalities. The static features of multiple modalities are then fused to obtain a regional static semantic representation. Based on the OD flow data, the historical flow status of the urban area is modeled to obtain a regional dynamic state sequence, and the regional dynamic state sequence is encoded to obtain regional dynamic conditions. Based on the dynamic conditions of the region, scaling parameters and offset parameters are generated through a multilayer perceptron network. Then, based on the scaling parameters and offset parameters, the static semantic representation of the region is modulated dimension by dimension to obtain a time-aware urban region representation.

2. The method for generating time-aware urban area representations based on dynamic condition modulation according to claim 1, characterized in that, The process of dividing the target city area into multiple city regions and obtaining static city data and OD flow data corresponding to each city region specifically includes: The target city area is divided into multiple non-overlapping city regions: ； in, Indicates the target city range. Indicates the first Each city area Indicates urban area index, This indicates the total number of urban areas; Obtain static urban data of different modalities corresponding to each of the urban areas, wherein the static urban data includes point of interest category distribution, street view images, remote sensing images and open street map base maps; Obtain OD flow data between different urban areas at multiple time slices, wherein the OD flow data includes departure area identifier, arrival area identifier, time slice identifier, and flow intensity; The static urban data is used to describe the long-term stable attributes of the urban area, while the OD flow data is used to describe the activity status of the urban area.

3. The method for generating time-aware urban area representations based on dynamic condition modulation according to claim 2, characterized in that, The static feature extractor includes a graph attention network encoder, a street view visual encoder, a remote sensing visual encoder, and a base map encoder. The static features of the various modalities include interest point modality features, street view visual features, remote sensing visual features, and base map features. The process involves using a static feature extractor to extract static features from the static city data, obtaining static features of multiple modalities, and then fusing these static features to obtain a regional static semantic representation. Specifically, this includes: The distribution of interest point categories is vectorized to obtain an interest point category distribution vector. Based on this vector, the functional similarity of interest points between any two urban areas is calculated. An interest point functional similarity map is constructed based on this similarity. A pre-trained graph attention network encoder is used to extract static features from the interest point functional similarity map to obtain the modal features of the interest points. ； ； in, Indicates urban area With urban areas Functional similarity of points of interest between them Indicates urban area index, and Representing urban areas and urban areas Interest point category distribution vector This represents the transpose of a vector. The norm of a vector Indicates urban area Point of interest modal features This represents a graph attention network encoder. Similarity graph representing points of interest Functionally similar adjacency matrix; The street view visual features are obtained by extracting static features from the street view image using the pre-trained street view visual encoder: ； in, Indicates urban area Street scene visual features Indicates a street view visual encoder. Indicates urban area Street view images; The remote sensing visual features are obtained by extracting static features from the remote sensing image using the pre-trained remote sensing visual encoder: ； in, Indicates urban area Remote sensing visual features, Indicates a remote sensing visual encoder. Indicates urban area Remote sensing images; The pre-trained basemap encoder is used to extract static features from the open street map basemap to obtain the basemap features: ； in, Indicates urban area Base map features, Indicates the base map encoder. Indicates urban area Open street map base map; The interest point modal features, street view visual features, remote sensing visual features, and base map features are sequentially processed by mapping, stitching, fusion transformation, and normalization to obtain a static semantic representation of the region.

4. The method for generating time-aware urban area representations based on dynamic condition modulation according to claim 3, characterized in that, The process of sequentially mapping, stitching, fusion transformation, and normalization of the interest point modal features, street view visual features, remote sensing visual features, and base map features to obtain a static semantic representation of the region specifically includes: The point of interest modal features, the street view visual features, the remote sensing visual features, and the base map features are each mapped to the same dimension to obtain the mapped point of interest modal features, the mapped street view visual features, the mapped remote sensing visual features, and the mapped base map features. The mapped point of interest modal features, mapped street view visual features, mapped remote sensing visual features, and mapped base map features are stitched together to obtain the first fused feature; The first fusion feature is input into the fusion mapping layer for fusion transformation processing to obtain the second fusion feature; The second fused feature is normalized to obtain the region's static semantic representation: ； in, Indicates urban area The region's static semantic representation, This indicates normalization processing. Indicates the fusion mapping layer, This indicates splicing / joining. Representing urban areas The mapped point of interest modal features, mapped street view visual features, mapped remote sensing visual features, and mapped base map features.

5. The time-aware urban area representation generation method based on dynamic condition modulation according to claim 4, characterized in that, The process of modeling the historical flow status of the urban area based on the OD flow data to obtain a regional dynamic state sequence, and encoding the regional dynamic state sequence to obtain regional dynamic conditions, specifically includes: A flow matrix is ​​constructed based on the OD flow data, and flow statistics are calculated based on the flow matrix, wherein the flow statistics include outbound flow, arrival flow, total interaction flow, and net flow. ； ； ； ； in, Indicates urban area In time slice Departure flow, Indicates urban area In time slice The arrival flow, Indicates urban area In time slice Total interaction traffic Indicates urban area In time slice Net flow Indicates time slice The flow matrix of the first line, number Column elements; The flow statistics are concatenated to obtain the dynamic state vector: ； in, Indicates urban area In time slice The dynamic state vector; Use a length of The dynamic state vector is acquired through a continuous time window to obtain the regional dynamic state sequence: ； in, Indicates urban area In time slice The regional dynamic state sequence, Indicates urban area In time slice The dynamic state vector, Indicates urban area In time slice The dynamic state vector, It is a constant; The dynamic state sequence of the region is encoded using a time-series encoder to obtain the dynamic conditions of the region: ； in, Indicates urban area In time slice Regional dynamic conditions, This indicates a timing encoder.

6. The time-aware urban area representation generation method based on dynamic condition modulation according to claim 5, characterized in that, The multilayer perceptron network includes a first multilayer perceptron network for generating scaling parameters and a second multilayer perceptron network for generating offset parameters. The process involves generating scaling and offset parameters through a multilayer perceptron network based on the dynamic conditions of the region, and then modulating the static semantic representation of the region dimension-by-dimensional based on the scaling and offset parameters to obtain a time-aware urban region representation. Specifically, this includes: The dynamic conditions of the region are input into the first multilayer perceptron network and the second multilayer perceptron network, respectively. The first multilayer perceptron network and the second multilayer perceptron network output scaling parameters and offset parameters, respectively, wherein the dimensions of the scaling parameters and the offset parameters are consistent with the dimensions of the static semantic representation of the region. ； ； in, and Representing urban areas In time slice The scaling and offset parameters, and These represent the first multilayer perceptron network and the second multilayer perceptron network, respectively. The scaling parameter is used to perform element-wise multiplication on the static semantic representation of the region to obtain a scaled static semantic representation of the region. The offset parameter is then added to the scaled static semantic representation of the region to obtain a dynamically modulated intermediate representation. Finally, the dynamically modulated intermediate representation is normalized to obtain a time-aware urban region representation. ； in, Indicates urban area In time slice Time-aware urban area representation. This indicates element-wise multiplication.

7. The method for generating time-aware urban area representations based on dynamic condition modulation according to claim 6, characterized in that, The process involves generating scaling and offset parameters through a multilayer perceptron network based on the dynamic conditions of the region, and then modulating the static semantic representation of the region dimension-by-dimensional based on the scaling and offset parameters to obtain a time-aware urban region representation. The process further includes: The time-perceived urban area representation of the urban area in the time slice is used as the anchor point. The time-perceived urban area representation of the urban area in the previous time slice and the next time slice is used as the positive sample. The time-perceived urban area representation of other urban areas or other non-adjacent time slices in the training batch is used as the negative sample. A dynamic contrastive learning loss function is constructed. By minimizing this loss function, the region dynamic conditional coding module and the static-to-dynamic modulation module are jointly optimized to reduce the semantic similarity between the time-aware urban region representations in the positive samples and increase the semantic similarity between the time-aware urban region representations in the negative samples. ； in, This represents the dynamic contrastive learning loss function. Indicates urban area In time slice Positive samples, This represents the time-aware urban area representation in the positive samples. Indicates negative samples Time-aware urban area representation in [the context of] ... Represents the semantic similarity function. Indicates the temperature coefficient; The regional dynamic condition encoding module is used to encode the regional dynamic state sequence to obtain regional dynamic conditions. The static-to-dynamic modulation module is used to generate scaling parameters and offset parameters through a multilayer perceptron network according to the regional dynamic conditions, and to perform dimension-by-dimensional modulation on the regional static semantic representation according to the scaling parameters and the offset parameters to obtain a time-aware urban regional representation.

8. A time-aware urban area representation generation system based on dynamic condition modulation, characterized in that, The time-aware urban area representation generation system based on dynamic condition modulation includes: The data acquisition module is used to divide the target city area into multiple city regions and acquire static city data and OD flow data corresponding to each city region. The feature extraction and fusion module is used to extract static features from the static city data using a static feature extractor to obtain static features of multiple modalities, and to fuse the static features of multiple modalities to obtain a regional static semantic representation. The modeling and coding module is used to model the historical flow status of the urban area based on the OD flow data, obtain a regional dynamic state sequence, and encode the regional dynamic state sequence to obtain regional dynamic conditions. The static-to-dynamic modulation module is used to generate scaling parameters and offset parameters through a multilayer perceptron network according to the dynamic conditions of the region, and to perform dimension-by-dimensional modulation on the static semantic representation of the region according to the scaling parameters and the offset parameters to obtain a time-aware urban region representation.

9. A terminal, characterized in that, The terminal includes: a memory, a processor, and a time-aware urban area representation generation program based on dynamic condition modulation stored in the memory and executable on the processor. When the time-aware urban area representation generation program based on dynamic condition modulation is executed by the processor, it implements the steps of the time-aware urban area representation generation method based on dynamic condition modulation as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a time-aware urban area representation generation program based on dynamic condition modulation, which, when executed by a processor, implements the steps of the time-aware urban area representation generation method based on dynamic condition modulation as described in any one of claims 1-7.

Images