A method and device for intelligently identifying time-space development characteristics of a township, and an electronic device

Abstract

The application relates to a method and device for intelligently identifying time-space development characteristics of a township and electronic equipment, the method comprising: training a target Transformer model using a training sample set and a time-space development state label, the target Transformer model comprising a time sequence mapping module and a space mapping module, the time sequence mapping module mapping index characteristics corresponding to each preset modeling index to a target dimension space to obtain time sequence embedding representation, the space mapping module mapping location characteristics corresponding to a spatial geographic location to the target dimension space to obtain space embedding representation, and the space embedding representation being copied along a time axis using a broadcast mechanism and then loaded onto each time step of the time sequence embedding representation; and determining a time-space development characteristic identification result of a township to be identified based on the trained target Transformer model and second multi-source time-space perception observation data. The method can accurately identify multi-dimensional time-space development characteristics of a township and ensure the accuracy of the identification result.

Description

Technical Field

[0001] This application relates to the field of machine learning technology, and more specifically, to an intelligent identification method, device, and electronic device for the spatiotemporal development characteristics of townships in the field of machine learning technology. Background Technology

[0002] In relevant research and application practices, the technology for identifying the spatiotemporal development characteristics of townships has undergone a series of evolutions. Most of them rely on traditional ledger statistics for static analysis. This identification method is difficult to accurately capture the differentiated spatiotemporal development characteristics of townships, and the identification accuracy is low. Summary of the Invention

[0003] This application provides an intelligent identification method, device, and electronic device for the spatiotemporal development characteristics of townships, which can achieve accurate and dynamic identification of the multidimensional spatiotemporal development characteristics of townships and ensure the accuracy and reliability of the identification results.

[0004] In a first aspect, embodiments of this application provide an intelligent identification method for the spatiotemporal development characteristics of townships, the method comprising: Obtain a training sample set and spatiotemporal development status labels for each training sample. The training sample set is constructed based on the first multi-source spatiotemporal perception observation data of multiple sample towns during historical years. The first multi-source spatiotemporal perception observation data includes the first indicator observation values ​​of multiple preset modeling indicators and the spatial geographical location of the sample towns. Using the training sample set and the spatiotemporal development state labels, the constructed target Transformer model is trained. The target Transformer model includes at least a temporal mapping module and a spatial mapping module. The training process of the target Transformer model includes mapping the indicator features corresponding to each preset modeling indicator to the target dimension space through the temporal mapping module to obtain a temporal embedding representation, and mapping the location features corresponding to the spatial geographic location to the target dimension space through the spatial mapping module to obtain a spatial embedding representation. The spatial embedding representation is then copied along the time axis and loaded onto each time step of the temporal embedding representation using a broadcast mechanism to obtain a fused spatiotemporal embedding representation. Then, the state category prediction value corresponding to the corresponding training sample is determined based on the fused spatiotemporal embedding representation, and the network parameters of the target Transformer model are iteratively optimized using a preset Bayesian optimization algorithm based on the deviation between the state category prediction value and the spatiotemporal development state label. Based on the trained target Transformer model and the second multi-source spatiotemporal perception observation data of the township to be identified, the spatiotemporal development characteristics identification results of the township to be identified are determined.

[0005] Secondly, embodiments of this application provide an intelligent identification device for the spatiotemporal development characteristics of townships, the device comprising: The acquisition unit is used to acquire a training sample set and the spatiotemporal development status label of each training sample. The training sample set is constructed based on the first multi-source spatiotemporal perception observation data of multiple sample towns during the historical year. The first multi-source spatiotemporal perception observation data includes the first indicator observation value of multiple preset modeling indicators and the spatial geographical location of the sample towns. A determination unit is used to train a constructed target Transformer model using the training sample set and the spatiotemporal development state labels. The target Transformer model includes at least a temporal mapping module and a spatial mapping module. The training process of the target Transformer model includes mapping the indicator features corresponding to each preset modeling indicator to the target dimension space through the temporal mapping module to obtain a temporal embedding representation, and mapping the location features corresponding to the spatial geographic location to the target dimension space through the spatial mapping module to obtain a spatial embedding representation. The spatial embedding representation is then copied along the time axis and loaded onto each time step of the temporal embedding representation using a broadcast mechanism to obtain a fused spatiotemporal embedding representation. Then, the state category prediction value corresponding to the corresponding training sample is determined based on the fused spatiotemporal embedding representation, and the network parameters of the target Transformer model are iteratively optimized using a preset Bayesian optimization algorithm based on the deviation between the state category prediction value and the spatiotemporal development state label. The identification unit is used to determine the spatiotemporal development characteristics identification result of the township to be identified based on the trained target Transformer model and the second multi-source spatiotemporal sensing observation data of the township to be identified.

[0006] Thirdly, embodiments of this application provide an electronic device, including: a processor and a memory; The processor is connected to the memory. The aforementioned memory is used to store executable program code; The processor reads the executable program code stored in the memory to run the program corresponding to the executable program code, so as to execute the method provided by the first aspect of the embodiments of this application or any possible implementation of the first aspect.

[0007] Fourthly, embodiments of this application provide a computer storage medium storing a plurality of instructions, which are adapted to be loaded by a processor and executed by the method steps provided in the first aspect of the embodiments of this application or any possible implementation thereof.

[0008] Fifthly, embodiments of this specification provide a computer program product containing instructions that, when run on a computer, cause the computer to perform the method steps provided by the first aspect of the embodiments of this application or any possible implementation thereof.

[0009] The beneficial effects of the technical solutions provided in some embodiments of this application include at least the following: In one or more embodiments of this application, a target Transformer model is constructed, including at least a temporal mapping module and a spatial mapping module. Training samples are built using first multi-source spatiotemporal sensing observation data for model training. This enables the target Transformer model to capture the long-term spatiotemporal evolution of townships while inherently perceiving and processing the spatial implicit constraints brought about by different geographical locations. Subsequently, based on the trained target Transformer model and the second multi-source spatiotemporal sensing observation data of the township to be identified, the spatiotemporal development characteristics of the township to be identified are determined. Thus, in the process of township identification, the deep spatiotemporal correlation features mined by the model are relied upon, while also conforming to the real spatiotemporal development data of the township. This allows for accurate and stable intelligent identification of the spatiotemporal development characteristics of townships, effectively ensuring the accuracy and reliability of the identification results. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A flowchart illustrating an intelligent identification method for the spatiotemporal development characteristics of townships, provided as an exemplary embodiment of this application; Figure 2 A schematic diagram of the module structure of a target Transformer model provided for an exemplary embodiment of this application; Figure 3 A flowchart illustrating another intelligent identification method for the spatiotemporal development characteristics of townships provided as an exemplary embodiment of this application; Figure 4 A schematic diagram of the process framework for an intelligent identification method for the spatiotemporal development characteristics of townships provided as an exemplary embodiment of this application; Figure 5 A flowchart illustrating another intelligent identification method for the spatiotemporal development characteristics of townships provided as an exemplary embodiment of this application; Figure 6A flowchart illustrating another intelligent identification method for the spatiotemporal development characteristics of townships provided as an exemplary embodiment of this application; Figure 7 A schematic diagram of the structure of an intelligent identification device for the spatiotemporal development characteristics of a township, provided as an exemplary embodiment of this application; Figure 8 This is a schematic diagram of the structure of an electronic device provided as an exemplary embodiment of this application. Detailed Implementation

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] This application provides an intelligent identification method, device, electronic device, storage medium, and computer program product for the spatiotemporal development characteristics of townships.

[0014] Please see Figure 1 , Figure 1 This is a flowchart illustrating an intelligent identification method for the spatiotemporal development characteristics of townships, provided as an exemplary embodiment of this application. Figure 1 As shown, the intelligent identification method for the spatiotemporal development characteristics of this township may include the following steps S101-S103, wherein: S101. Obtain the training sample set and the spatiotemporal development status label of each training sample. The training sample set is constructed based on the first multi-source spatiotemporal perception observation data of multiple sample towns during the historical year. The first multi-source spatiotemporal perception observation data includes the first indicator observation value of multiple preset modeling indicators and the spatial geographical location of the sample town.

[0015] The construction of the training samples is based on the primary multi-source spatiotemporal sensing observation data of the sample townships. This primary multi-source spatiotemporal sensing observation data includes spatial geographic location and indicator observation data of the six-dimensional integrated indicator system. Spatial geographic location can refer to the geographic location of the township's administrative center or centroid, including longitude and latitude information. The six-dimensional integrated indicator system is a multi-level structured feature framework constructed for intelligent mining and deep learning modeling of township spatiotemporal features. All types of feature data under this indicator system are acquired and standardized into modeling data sources through technologies such as remote sensing satellite inversion, IoT terminal sensing, geographic information mapping, and full-domain spatiotemporal multi-source acquisition.

[0016] This six-dimensional integrated indicator system includes, but is not limited to, six major technical dimensions: climate remote sensing monitoring indicators, ecological environment remote sensing interpretation indicators, transportation facility Internet of Things sensing indicators, urban and rural spatial topological relationship correlation indicators, natural resource endowment surveying and mapping indicators, and urban and rural business spatiotemporal structure monitoring indicators.Optionally, these six technical dimensions further include the following 18 quantitative modeling features that can be used for model training (i.e., the aforementioned preset modeling indicators, hereinafter referred to as indicators). Among them, the climate remote sensing monitoring indicators are standard quantitative features obtained through time-series inversion of meteorological remote sensing satellites, used to extract the temporal evolution and spatiotemporal distribution characteristics of climate elements at the township scale. These include modeling indicators such as the time-series characteristics of average temperature, the time-series characteristics of average precipitation, the remote sensing interpretation characteristics of drought index, and the time-series monitoring characteristics of relative humidity. The ecological environment remote sensing interpretation indicators are ecological background features extracted based on remote sensing image interpretation and geospatial overlay analysis, used to characterize the spatial differentiation and dynamic evolution of township ecological patterns. These include modeling indicators such as the remote sensing interpretation characteristics of the Ecological Fragility Index (EFI) and the spatiotemporal evolution characteristics of Land Use Dynamic Degree (LUD). The transportation facility Internet of Things sensing indicators are spatial facility features collected based on road network sensing terminals and logistics monitoring equipment, used to characterize the spatial layout and external connectivity topology of township transportation road networks. These include the spatial distribution characteristics of road network density and the accessibility of logistics (Cargo Transportation). Modeling indicators include Connectivity (CTC) perception characteristics and Transportation Capacity (TTC) temporal monitoring characteristics; Urban-Rural Spatial Topological Relationship Indicators are urban-rural coupling characteristics constructed based on geographic information topological analysis and spatial adjacency relationships, used to explore spatial linkages and element agglomeration distribution characteristics between towns and urban areas, including spatiotemporal evolution characteristics of Urbanization Rate (UR), spatial distribution characteristics of Population Density (POP), and spatial coverage characteristics of public education facilities; Natural Resource Endowment Mapping Indicators are baseline quantitative characteristics obtained through on-site land spatial mapping and resource carrying capacity spatial simulation modeling, used to quantify the spatial carrying capacity and temporal change characteristics of the natural resource base of towns, including spatial measurement characteristics of Population Carrying Capacity (PC), spatial assessment characteristics of Ecological Carrying Capacity (EC), and spatial assessment characteristics of Water Resources Carrying Capacity (PC). Modeling indicators such as the temporal spatial characteristics of Capacity (WRCC); monitoring indicators for the temporal and spatial structure of urban and rural business formats are used to depict the temporal evolution trend of the spatial layout and structure of business formats across the entire township. These include modeling indicators such as the temporal evolution characteristics of regional comprehensive volume, the spatial distribution characteristics of the Industrial Structure Index (ISI), and the temporal evolution characteristics based on the proportion of business formats.

[0017] Each training sample includes multiple standardized first-source spatiotemporal sensing observation data. This standardization process is used to eliminate dimensional differences between different indicators. These first-source spatiotemporal sensing observation data are presented in the form of feature sequences, including the indicator features (which can be viewed as temporal features changing over time) corresponding to the aforementioned preset modeling indicator data (the aforementioned first indicator observation values), and the location features corresponding to spatial geographic locations. For example, the set of indicator features (a three-dimensional temporal feature sequence) for N sample towns, T time steps (e.g., 8 years), and D indicators per time step (e.g., the aforementioned 18 indicators) can be denoted as X∈RN×T×D, and the location features can be denoted as... The spatiotemporal development status label is a discrete status label used for model sample classification and annotation. It identifies whether a sample town belongs to a preset target development status category at a subsequent time point. This label can be automatically set based on the historical spatiotemporal development status archive data of the sample town. The label value is set to two discrete values: 1 and 0. 1 represents that the sample town belongs to the preset target development status category at the next time point, and 0 represents that the sample town does not belong to the preset target development status category at the next time point.

[0018] It should be noted that the sources of the first multi-source spatiotemporal sensing observation data include township statistical yearbooks, monitoring data from natural resources departments, data filed by transportation departments, monitoring reports from ecological and environmental departments, publicly available data from urban and rural construction departments, and population census data. Generally, the data can be obtained from formal channels such as designated government data open platforms, geographic information public service websites, and official release channels of industry authorities. Invalid and abnormal data are strictly screened and removed to ensure the integrity, accuracy, and standardization of the sample data.

[0019] S102. Using the training sample set and spatiotemporal development state labels, train the constructed target Transformer model. The target Transformer model includes at least a temporal mapping module and a spatial mapping module. The training process of the target Transformer model includes mapping the indicator features corresponding to each preset modeling indicator to the target dimension space through the temporal mapping module to obtain a temporal embedding representation, and mapping the location features corresponding to the spatial geographic location to the target dimension space through the spatial mapping module to obtain a spatial embedding representation. The spatial embedding representation is copied along the time axis and loaded onto each time step of the temporal embedding representation using a broadcast mechanism to obtain a fused spatiotemporal embedding representation. Then, the state category prediction value corresponding to the corresponding training sample is determined based on the fused spatiotemporal embedding representation. Based on the deviation between the state category prediction value and the spatiotemporal development state label, the network parameters of the target Transformer model are iteratively optimized using a preset Bayesian optimization algorithm.

[0020] The target Transformer model is a deep learning model improved on the Transformer architecture. It uses a nonlinear spatial projection and broadcast fusion mechanism to explicitly encode geographic coordinates into each time step of the temporal feature sequence. This enables the model to capture the long-term spatiotemporal evolution law while inherently perceiving and analyzing the spatial implicit constraints brought about by different geographical locations, significantly improving the model's ability to mine deep spatiotemporal correlation features under complex terrain and differentiated spatial patterns.

[0021] Specifically, please see Figure 2 , Figure 2 This is a schematic diagram of the module structure of a target Transformer model provided as an exemplary embodiment of this application. The target Transformer model includes a temporal mapping module, a spatial mapping module, a positional encoding injection module, a self-attention encoding module, and a prediction output module. The spatial mapping module includes at least two linear transformation layers and a nonlinear activation layer. The spatial mapping module is used to map two-dimensional spatial location features to a spatial embedding representation of the same dimension as the temporal embedding representation, and to copy the spatial embedding representation along the time dimension, so that the temporal embedding representation at each time step contains the spatial location constraint information of the corresponding township unit.

[0022] The feature sequence (i.e., time-series feature sequence) of the indicator features corresponding to each preset modeling indicator in the training sample as a function of time is denoted as X. seq X seq ∈ (that is, X) seq For one A two-dimensional tensor with rows and columns D), the location feature of the i-th sample township is denoted as S. i , The length of the time window (e.g., 1 year). This refers to the total number of indicators (e.g., the 18 indicators mentioned above). The spatiotemporal development status label is set based on the historical evaluation results of the sample townships and can be denoted as... , , indicates whether the sample township belongs to the preset target development state category at the next time point. 1 means that the sample township belongs to the preset target development state category at the next time point, and 0 means that the sample township does not belong to the preset target development state category at the next time point.

[0023] During model training, the training samples are first input into the target Transformer model. The processing flow of the target Transformer model for each training sample is as follows: First, the time-series mapping module maps the time-series feature sequences X from the training samples. seq Mapping to a high-dimensional latent space d via a linear layer model(i.e., the target dimension space mentioned above), to obtain the temporal embedding representation (denoted as ). Its calculation expression is: ,in, , The input time-series features (i.e., the feature characteristics corresponding to each preset modeling index) tensor (B is the batch size). It is a learnable linear projection weight matrix; The hidden layer dimension of the model; This is a scaling factor used to maintain numerical stability.

[0024] At the same time, the spatial mapping module will map the positional features in the training samples. (i.e., two-dimensional latitude and longitude geographic coordinates) are mapped to a spatial embedding representation (denoted as ). Its calculation expression is: ,in, It is the dimension of the intermediate layer of spatial mapping; This is the hyperbolic tangent activation function, used to introduce nonlinear features; , The geographic coordinate matrix input to the current training batch (from global features) (Sampling obtained from B, where B is the batch size and 2 is the latitude and longitude dimension). , The weights and biases are for the first-level linear transformation; , The weights and biases for the second-level linear transformation are used to map the dimension to... (That is, the target dimension space mentioned above). Next, the spatial mapping module uses a broadcast mechanism to embed the spatial representation. Loaded into time-series embedded representation At each time step, spatiotemporal feature fusion is achieved. This operation enables the model to explicitly perceive the geographical location constraints of the town when processing the state at each time point. The corresponding operation can be expressed as the formula: ;in, To represent the temporal features that incorporate geographic location information, This indicates that the dimension is The spatial vector (i.e., the spatial embedding representation) is copied along the time axis. Next, make its dimension become so as to and Perform addition operations.

[0025] Next, the position encoding injection module superimposes sine / cosine position codes to preserve sequence timing information. This operation can be expressed by the following formula: ; ; +PE; in, Indicates the position index of the time step (i.e., from 0 to 1). ); Indicates the index of the hidden layer dimension. and These correspond to even and odd dimension positions, respectively. This is the initial input to the encoder (self-attention encoding module) in the Transformer.

[0026] Next, The input is processed by the self-attention encoding module, which uses a multi-head attention mechanism to capture long-range temporal dependencies. The relevant operations can be expressed as the formula: , Here, Q, K, and V represent the query, key, and value, respectively, forming the core triplet of the self-attention mechanism, which is obtained through a linear transformation. The dimension of the attention head is used to scale the dot product result, preventing the gradient of the activation function from vanishing due to excessively large values.

[0027] Finally, the prediction output module takes the output data of the last time step of the self-attention encoding module and processes it through a fully connected layer to obtain the final log odds (i.e., the prediction result of whether a township belongs to the preset target development state category at the next time node). The relevant operation can be expressed as the formula: ,in, The log-odds ratio is the final output of the model (i.e., the predicted state category value mentioned above). This predicted value is a continuous probability value between 0 and 1. The closer the value is to 1, the higher the probability that the town will belong to the preset target development state category at the next time point; the closer the value is to 0, the lower the probability that the town will belong to the preset target development state category at the next time point. These are the weights and biases of the output layer, respectively.

[0028] Next, the deviation between the predicted state categories and the actual spatiotemporal development state labels is calculated using a loss function, and the result of the loss function calculation is used as the optimization objective to adjust the model parameters. Parameters include the number of self-attention heads, hidden layer dimension, temporal / spatial location embedding dimension, regularization coefficient (such as Dropout rate), learning rate, etc.

[0029] Optionally, during model training, a pre-defined Bayesian optimization algorithm can be introduced to automatically search for the optimal hyperparameter combination within a pre-defined space, iteratively adjusting the core network parameters of the target Transformer model. The pre-defined Bayesian optimization algorithm can employ a tree-structured Parzen Estimator (TPE). The boundary conditions of the hyperparameter search space can be set based on the operational scenario of assessing the spatiotemporal development status of townships, avoiding model training failure due to parameter values ​​exceeding reasonable ranges.

[0030] It should be noted that during the model training phase, this embodiment integrates a Bayesian optimization framework to establish an adaptive hyperparameter optimization mechanism for small sample scenarios. This mechanism can adaptively match hyperparameter search strategies to the characteristics of small sample data in the township spatiotemporal development status identification task, achieving automated and efficient hyperparameter optimization during model training. This effectively avoids the shortcomings of traditional manual parameter tuning, such as low efficiency, reliance on experience, and susceptibility to local optima, significantly improving the robustness and automated training level of the target Transformer model. Simultaneously, to address the problem of scarce positive samples with spatiotemporal development characteristics of the preset target development status category in the basic sample set, and the model fitting bias caused by the imbalance of positive and negative samples, this embodiment introduces a dynamic sampling (hybrid resampling) strategy to adaptively process the basic sample set. Specifically, feature enhancement is applied to positive samples to expand the effective sample size, and context-aware downsampling is performed on negative samples to remove redundant and noisy samples. This achieves dynamic balance in the distribution of training sample categories, ensuring that the model can fully learn and fit the core spatiotemporal feature differences of different types of samples, further improving the model's prediction accuracy and generalization ability.

[0031] S103. Based on the trained target Transformer model and the second multi-source spatiotemporal perception observation data of the township to be identified, determine the spatiotemporal development characteristics identification results of the township to be identified.

[0032] The second multi-source spatiotemporal sensing observation data includes the second indicator observation values ​​of the aforementioned preset modeling indicators, which can objectively reflect the actual situation of various development elements of the township to be evaluated. First, a preset attribution analysis algorithm (such as the SHAP algorithm) can be used to extract the nonlinear contribution of high-dimensional features to the spatiotemporal development status of the township from the trained target Transformer model. Then, combined with the second multi-source spatiotemporal sensing observation data of the township to be identified, a comprehensive judgment is made to obtain the final spatiotemporal development feature identification result.

[0033] As described above, the intelligent identification method for the spatiotemporal development characteristics of townships provided in this embodiment constructs a target Transformer model that includes at least a temporal mapping module and a spatial mapping module. Training samples are constructed using first-source multi-source spatiotemporal sensing observation data for model training. This allows the target Transformer model to capture the long-term spatiotemporal evolution patterns of townships while inherently perceiving and processing the implicit spatial constraints imposed by different geographical locations. Subsequently, based on the trained target Transformer model and the second-source multi-source spatiotemporal sensing observation data of the township to be identified, the spatiotemporal development characteristics identification result of the township is determined. Thus, in the process of township identification, it relies on the deep spatiotemporal correlation features mined by the model and closely matches the real spatiotemporal development data of the township, enabling accurate and stable intelligent identification of township spatiotemporal development characteristics and effectively ensuring the accuracy and reliability of the identification results.

[0034] For the intelligent identification method of township spatiotemporal development characteristics based on the above embodiments, please refer to [link to relevant documentation]. Figure 3 and Figure 4 , Figure 3 A flowchart illustrating another intelligent identification method for the spatiotemporal development characteristics of townships provided as an exemplary embodiment of this application. Figure 4 This is a schematic flowchart illustrating an exemplary embodiment of the present application of a method for intelligently identifying the spatiotemporal development characteristics of townships. The method specifically includes the following steps S201-S205, wherein: S201. Obtain the training sample set and the spatiotemporal development status label of each training sample. The training sample set is constructed based on the first multi-source spatiotemporal perception observation data of multiple sample towns during the historical year. The first multi-source spatiotemporal perception observation data includes the first indicator observation value of multiple preset modeling indicators and the spatial geographical location of the sample town.

[0035] The steps S201 and S101 are the same as those described above, and will not be repeated here.

[0036] S202. Using the training sample set and spatiotemporal development state labels, train the constructed target Transformer model. The target Transformer model includes at least a temporal mapping module and a spatial mapping module. The training process of the target Transformer model includes mapping the indicator features corresponding to each preset modeling indicator to the target dimension space through the temporal mapping module to obtain a temporal embedding representation, and mapping the location features corresponding to the spatial geographic location to the target dimension space through the spatial mapping module to obtain a spatial embedding representation. The spatial embedding representation is copied along the time axis and loaded onto each time step of the temporal embedding representation using a broadcast mechanism to obtain a fused spatiotemporal embedding representation. Then, the state category prediction value corresponding to the corresponding training sample is determined based on the fused spatiotemporal embedding representation. Based on the deviation between the state category prediction value and the spatiotemporal development state label, the network parameters of the target Transformer model are iteratively optimized using a preset Bayesian optimization algorithm.

[0037] The steps S202 and S102 are the same as those described above, and will not be repeated here.

[0038] S203. Based on the training sample set, the trained target Transformer model, and the preset attribution algorithm, determine the model-driven weights of each preset modeling index.

[0039] The pre-defined attribution analysis algorithm can employ the SHAP algorithm, with the contribution value being the SHAP value. The model-driven weights utilize the spatiotemporal development status categories of townships at subsequent time points as binary classification supervision, inversely calculating the nonlinear contribution of each indicator feature (SHAP Shapley value, which can be considered empirical weights). A deep learning model (target Transformer model) is trained using the future spatiotemporal development status discrimination as a proxy supervision task. The pre-defined attribution analysis algorithm (e.g., SHAP) extracts the nonlinear contribution of high-dimensional features to the spatiotemporal development status of townships from the trained target Transformer model and transforms it into empirical weights with clear physical meaning. This approach successfully transforms discrete supervision annotation information into continuous quantitative discrimination criteria, ensuring that the subsequently generated spatiotemporal development feature representation values ​​have a rigorous data inference foundation. The model-driven weights are set based on the theoretical framework of a six-dimensional integrated indicator system (climate remote sensing monitoring indicators, ecological environment remote sensing interpretation indicators, transportation facility Internet of Things sensing indicators, urban and rural spatial topological relationship correlation indicators, natural resource endowment surveying and mapping indicators, and urban and rural business spatiotemporal structure monitoring indicators), ensuring that the allocation of indicator weights conforms to the inherent laws of township spatial evolution.

[0040] S204. Based on the second indicator observation values ​​of each preset modeling indicator for the township to be identified during the target year, determine the data-driven weights of each preset modeling indicator and the extreme value matrix corresponding to all second indicator observation values.

[0041] Among them, the data-driven weights are statistical weights based on information entropy, which can be considered as objective weights driven by data. They can accurately reflect the differences in the contribution of each preset modeling indicator to the characteristic differentiation of the spatiotemporal development status of townships in the actual data. Specifically, the information entropy value of each preset modeling indicator can be calculated first to measure the dispersion and effective feature information of the indicator data. Then, the data-driven weights of each preset modeling indicator are obtained by normalizing the information entropy value to ensure that the sum of the weights is 1. Generally, the smaller the information entropy value of an indicator, the greater the dispersion of the indicator and the stronger its effective differentiation ability of the spatiotemporal development differences of townships, and the larger the weight ratio it is assigned. Conversely, the larger the information entropy value of an indicator, the higher the homogeneity of the indicator data and the less effective differentiation information it has, and the smaller the weight ratio it is assigned.

[0042] The target year period can be determined according to needs, such as selecting the most recent 5 years as the target year period. The extreme value matrix is ​​a matrix obtained by standardizing the extreme values ​​of all the preset modeling indicators of the township to be identified (i.e., the second indicator observation value mentioned above). The matrix dimension corresponds to the total number of indicators (e.g., 18). Its core function is to unify the dimensions of each preset modeling indicator, providing a standardized data benchmark for the subsequent accurate calculation of the spatiotemporal development characteristics of the township to be identified.

[0043] S205. Based on the model-driven weights, data-driven weights, and extreme value matrix, generate the spatiotemporal development characteristics identification results for the township to be identified.

[0044] The spatiotemporal development characteristic identification results include at least the spatiotemporal development characteristic representation values ​​of the township to be identified for each year during the target period. These spatiotemporal development characteristic representation values ​​are in exponential form, comprehensively reflecting the multidimensional spatiotemporal development characteristics of the township at corresponding time points. The larger the value, the more significant the township's spatiotemporal development characteristics and the better the spatial element coupling state in the corresponding year. Specifically, the fusion weights of each preset modeling indicator can be determined first by combining model-driven weights and data-driven weights, and then the spatiotemporal development characteristic representation values ​​can be calculated based on the fusion weights and the extreme value matrix. The spatiotemporal development characteristic representation values ​​take into account both the objective discrete distribution characteristics inherent in the observed data (reflected by data-driven weights) and the deep spatiotemporal feature correlation evolution laws mined by the model (reflected by model-driven weights). Simultaneously, the standardization effect of the extreme value matrix eliminates the differences in indicator dimensions, achieving accurate quantification and horizontal and vertical comparability of the feature contribution of different time points and different modeling indicators, effectively improving the stability and accuracy of the township spatiotemporal development characteristic identification results.

[0045] Optionally, the spatiotemporal development characteristic identification result also includes the spatiotemporal development characteristic evolution trajectory of the township to be identified during the target year. This trajectory is constructed based on the changing trends of the spatiotemporal development characteristic representation values ​​for each year. This trajectory includes multiple time-series evolution intervals, each corresponding to the township's spatiotemporal characteristic evolution state within the corresponding time period (e.g., steady improvement, rapid growth, stable fluctuations), clearly presenting the time-series evolution pattern of the township's multidimensional spatiotemporal development characteristics. Specifically, by fitting the time-series change curves of the spatiotemporal development characteristic representation values ​​for each year, the time nodes where the curve slope changes significantly are identified. These nodes are used as time-series segmentation inflection points, and the spatiotemporal development evolution intervals are divided based on these inflection points, achieving a refined decomposition of the township's stage-specific spatiotemporal development characteristics.

[0046] For the intelligent identification method of township spatiotemporal development characteristics based on the above embodiments, please refer to [link to relevant documentation]. Figure 5 , Figure 5 This is a flowchart illustrating another intelligent identification method for the spatiotemporal development characteristics of townships, provided as an exemplary embodiment of this application. The intelligent identification method for the spatiotemporal development characteristics of townships specifically includes the following steps S301-S305, wherein: S301. Standard deviation standardization is performed on the first indicator observation value and spatial geographical location of multiple preset modeling indicators for each sample township during the historical year to obtain the corresponding indicator features and location features.

[0047] First, the original data matrix is ​​constructed by selecting the observation data (first indicator observation values) of D pre-set modeling indicators (e.g., the aforementioned 18 indicators) from N sample townships over the past T years (e.g., consecutive 8 years). Simultaneously, the spatial geographical locations of the administrative centers or centroids of the sample townships are extracted. Second, the indicator observation data and spatial geographical locations are standardized to obtain standardized indicator features and location features, facilitating the construction of subsequent training samples. For some indicator observation data exhibiting long-tail distribution characteristics (i.e., a few samples with extremely high values ​​and most samples with low values), a logarithmic transformation is used for smoothing to correct skewness and reduce the interference of extreme values ​​on the identification of spatiotemporal development states. The formula is: ;in, This represents the d-th dimension original index observation value of the i-th township in year t (i.e., the first index observation value mentioned above); Let represent the logarithmic transformation of the d-th dimension index observation value of the i-th township in year t; the constant 1 is used to avoid the logarithmic function being undefined when the input value is 0.

[0048] Next, Z-Score standardization was used to process all indicator observations to eliminate dimensional differences between the pre-defined modeling indicators. The formula is as follows: ;in, This represents the standardized characteristics of the indicator. and The first The mean and standard deviation of the dimension index observations across all samples and time steps.

[0049] At the same time, the Z-Score normalization method was also used independently to process the spatial geographic location to obtain standardized location features. This is to ensure that the range of latitude and longitude values ​​matches the temporal characteristics.

[0050] S302. Based on the sliding window sampling strategy, construct a basic sample set for each sample township according to the characteristics of the indicator and the location characteristics. Different basic samples in the basic sample set correspond to different time intervals.

[0051] The basic samples in the basic sample set are based on the above-mentioned index characteristics (which can be called time-series characteristics that change over time). and location features Built-in. Input window length can be set. (Year), Forecast Window For the time series feature data of the i-th township head (8 years), the sliding window sampling technique is used to divide it into multiple short time series (1 year) time series feature data (which can be denoted as sequences). ), and each Geographical features of the i-th township respectively By combining them, we obtain the basic sample ( ), D represents the total number of indicators.

[0052] S303. The basic sample set is processed based on a hybrid resampling strategy to obtain a training sample set.

[0053] Given the scarcity of sample townships belonging to the preset target development state category, the basic sample set may suffer from a severe imbalance between positive and negative samples. Positive samples (samples generated from data of sample townships belonging to the preset target development state category) account for a significantly lower proportion than negative samples (samples generated from data of sample townships not belonging to the preset target development state category). Directly using the original samples to train the evaluation model could easily lead to a bias in predicting negative samples, hindering the effective learning of positive sample feature patterns and ultimately affecting the accuracy of identifying the spatiotemporal development characteristics of townships. Therefore, this embodiment introduces a hybrid resampling strategy, combining positive sample enhancement with context-aware downsampling of negative samples to achieve a balanced sample distribution, providing a high-quality data foundation for subsequent model training.

[0054] Furthermore, step S303 above may specifically include: The positive samples in the base sample set are augmented, and the negative samples in the base sample set are downsampled in a context-aware manner.

[0055] The core objective of positive sample augmentation is to increase the number of positive samples and enhance their diversity without altering their core features (such as key indicator features). The core objective of context-aware downsampling for negative samples is to remove redundant and noisy samples while preserving their diversity and representativeness, thus avoiding the loss of feature information caused by simple random downsampling.

[0056] Specifically, for positive samples, Gaussian noise can be introduced. Enhanced samples are generated to increase the number of positive samples. For negative samples, instead of random discarding, a dynamic retention probability is set based on the historical background of the township / town in the sample. Optimize the training set distribution. For example, samples with a spatiotemporal development status label of 0 at the current moment (indicating that they do not belong to the preset target development status category at the next time point) are considered "difficult negative samples" if the sample township belonged to the preset target development status category in other historical years, and their probabilities are retained. If the sample township has never belonged to the preset target development status category, it is treated as a regular negative sample, and its probability is retained. .

[0057] S304. Using the training sample set and the spatiotemporal development state labels of each training sample, the constructed target Transformer model is trained. The training process of the target Transformer model includes mapping the indicator features corresponding to each preset modeling indicator to the target dimension space through the temporal mapping module to obtain the temporal embedding representation, and mapping the location features corresponding to the spatial geographic location to the target dimension space through the spatial mapping module to obtain the spatial embedding representation. The spatial embedding representation is copied along the time axis and loaded onto each time step of the temporal embedding representation using a broadcast mechanism to obtain the fused spatiotemporal embedding representation. Then, the state category prediction value corresponding to the corresponding training sample is determined according to the fused spatiotemporal embedding representation. Based on the deviation between the state category prediction value and the spatiotemporal development state label, the network parameters of the target Transformer model are iteratively optimized using a preset Bayesian optimization algorithm.

[0058] The steps S304 and S102 are the same as those described above, and will not be repeated here.

[0059] S305. Based on the trained target Transformer model and the second multi-source spatiotemporal perception observation data of the township to be identified, determine the spatiotemporal development characteristics identification result of the township to be identified.

[0060] The steps S305 and S103 are the same as those described above, and will not be repeated here.

[0061] For the intelligent identification method of township spatiotemporal development characteristics based on the above embodiments, please refer to [link to relevant documentation]. Figure 6 , Figure 6 This is a flowchart illustrating another intelligent identification method for the spatiotemporal development characteristics of townships, provided as an exemplary embodiment of this application. The intelligent identification method for the spatiotemporal development characteristics of townships specifically includes the following steps S401-S411, wherein: S401. Obtain the training sample set and the spatiotemporal development status label of each training sample. The training sample set is constructed based on the first multi-source spatiotemporal perception observation data of multiple sample towns during the historical year. The first multi-source spatiotemporal perception observation data includes the first indicator observation value of multiple preset modeling indicators and the spatial geographical location of the sample town.

[0062] The steps S401 and S101 are the same as those described above, and will not be repeated here.

[0063] S402. Using the training sample set and the spatiotemporal development state label, train the constructed target Transformer model. The target Transformer model includes at least a temporal mapping module and a spatial mapping module. The training process of the target Transformer model includes mapping the indicator features corresponding to each preset modeling indicator to the target dimension space through the temporal mapping module to obtain a temporal embedding representation, and mapping the location features corresponding to the spatial geographic location to the target dimension space through the spatial mapping module to obtain a spatial embedding representation. The spatial embedding representation is then copied along the time axis and loaded onto each time step of the temporal embedding representation using a broadcast mechanism to obtain a fused spatiotemporal embedding representation. Then, the state category prediction value corresponding to the corresponding training sample is determined based on the fused spatiotemporal embedding representation. Based on the deviation between the state category prediction value and the spatiotemporal development state label, the network parameters of the target Transformer model are iteratively optimized using a preset Bayesian optimization algorithm.

[0064] The steps S402 and S102 are the same as those described above, and will not be repeated here.

[0065] S403. Determine the comparison sample group corresponding to each training sample, and input the training sample and the comparison sample group into the trained target Transformer model for processing.

[0066] The comparison sample group is obtained by removing indicators from individual training samples. Specifically, one indicator feature (i.e., the first indicator observation value mentioned above) is removed sequentially from the training sample. Each removal of a different indicator feature yields a corresponding comparison sample. All comparison samples obtained from the training sample in this way constitute the comparison sample group corresponding to the training sample. For example, if the training sample contains indicator features corresponding to 18 preset modeling indicators, removing one indicator feature sequentially from the training sample can construct 17 comparison samples. These 17 comparison samples together form the comparison sample group corresponding to the training sample, and each comparison sample contains indicator features corresponding to the remaining 17 preset modeling indicators. Each training sample and its comparison sample group are input into the trained target Transformer model for processing to output a predicted value (model output result) indicating whether a township belongs to a preset target development state category.

[0067] S404. Perform attribution analysis on the output of the target Transformer model using a preset attribution analysis algorithm to determine the contribution of each preset modeling index to the output.

[0068] The pre-defined attribution analysis algorithm can employ the SHAP algorithm, with the contribution value being the SHAP value. The SHAP algorithm determines the positive or negative contribution of the removed indicators to the model's prediction results based on the differences between the model outputs of the training samples and the comparison samples. Specifically, it calculates the SHAP value by comparing the predicted values ​​of the training samples with the predicted values ​​of the comparison samples after removing a certain pre-defined modeling indicator. The magnitude and direction of the SHAP value quantify the impact of the indicator's absence on the model's prediction results, clearly demonstrating the importance of different indicators in the model's predictions. This provides reliable technical support for subsequent analysis of key influencing factors in identifying the spatiotemporal development characteristics of townships.

[0069] S405. Determine the model-driven weights of each preset modeling indicator based on the contribution value.

[0070] In some embodiments, step S405 specifically includes: Take the absolute value of each contribution value and calculate the average value of the absolute values ​​of the same indicator to obtain the average value corresponding to each preset modeling indicator; Normalize all average values ​​to obtain the model-driven weights corresponding to each preset modeling index.

[0071] This process involves taking the absolute value of the SHAP values ​​for all samples, averaging them, and then normalizing them. This step "compresses" the complex nonlinear interaction effects captured by deep learning into dimensionless weight coefficients that reflect the global importance of features, achieving the best balance between model fitting accuracy and index interpretability.

[0072] ; in, is the normalized SHAP weight of the a-th indicator (i.e., the model-driven weight mentioned above). It is the SHAP value (contribution value) of the a-th feature in the m-th sample (training sample or comparison sample). The magnitude of the value represents the degree of influence, and the sign ( () represents a positive or negative effect; M is the total number of samples, and D is the total number of indicators.

[0073] S406. Determine the directional information of each preset modeling index, including positive and negative directions.

[0074] Among them, positive indicators are those whose larger values ​​have a more significant positive effect on the spatiotemporal development status of townships and a higher probability that the township belongs to the preset target development status category. Examples include the temporal evolution characteristics of regional comprehensive volume, the spatial distribution characteristics of road network density, and the spatiotemporal evolution characteristics of urbanization rate among the preset modeling indicators mentioned above. Negative indicators are those whose larger values ​​have a more significant negative impact on the spatiotemporal development status of townships and a lower probability that the township belongs to the preset target development status category. Examples include the remote sensing interpretation characteristics of drought index and the remote sensing interpretation characteristics of ecological vulnerability index among the preset modeling indicators mentioned above.

[0075] S407. Based on this directional information, the second indicator observation values ​​of each preset modeling indicator of the township to be identified during the target year are subjected to directional extreme value standardization processing to obtain the extreme value matrix.

[0076] Specifically, mapping formulas can be constructed for positive indicators (such as the spatiotemporal evolution characteristics of urbanization rate) and negative indicators (such as remote sensing interpretation characteristics of ecological vulnerability) to perform extreme value standardization processing, so as to unify the values ​​of all indicator features to the [0,1] interval, eliminate the influence of differences in the dimensions of different indicators, and finally obtain a standardized extreme value matrix. This provides standardized and comparable feature data support for the subsequent synthesis of representation values ​​(indices) of the spatiotemporal development characteristics of townships.

[0077] S408. Based on the entropy method and the extreme value matrix, determine the data-driven weights of each preset modeling index.

[0078] In order to mitigate the risk of overfitting in deep learning models due to sample bias and limitations in small-sample scenarios, and to improve the objectivity and reliability of indicator weight assignment, specifically, based on the aforementioned standardized extreme value matrix... First, the information entropy of the *a*-th pre-defined modeling indicator is calculated to measure the dispersion of the indicator's value and the amount of effective feature information. Then, based on this information entropy, the difference coefficient of the corresponding pre-defined modeling indicator is calculated. The larger the difference coefficient, the stronger the indicator's ability to effectively distinguish the spatial and temporal development differences of townships and the more effective information it contains. Finally, the difference coefficients of all indicators are normalized to obtain the data-driven weights of each pre-defined modeling indicator. This weight can objectively reflect the differences in information content of each preset modeling indicator, providing scientific and reasonable objective weight support for the subsequent synthesis of township spatiotemporal development characteristic representation values.

[0079] S409. Determine the fusion weight of each preset modeling index based on the model-driven weight and the data-driven weight.

[0080] The model-driven weights and data-driven weights can be fused using a weighted summation method. This ensures that the final fused weights balance the "empirical rationality" of machine learning and the "data rationality" of statistics, achieving a complementary advantage of the two weights and improving the scientific rigor and reliability of subsequent calculations of spatiotemporal development characteristic values. The specific weighted summation formula is as follows: ; in, This is the fusion coefficient (ranging from [0,1]), used to adjust the fusion ratio of model-driven weights and data-driven weights; The fusion weight of the a-th indicator, The data-driven weight (objective weight) for the a-th indicator calculated based on the entropy method is given above. These are the model-driven weights (empirical weights) calculated based on the SHAP algorithm.

[0081] It is important to emphasize that, addressing the core need for identifying the spatiotemporal development characteristics of townships, this embodiment employs a dual-track strategy of "separating and decoupling model training and index calculation." The core objective is to balance the accuracy of model predictions with the objectivity and comparability of index evaluation. To achieve this goal, this embodiment utilizes an "intensity-direction separation" mechanism to precisely transfer the nonlinear correlation patterns of indicators mined by the target Transformer model to a linear evaluation system, ultimately constructing an aggregated index (representation value) of township spatiotemporal development characteristics that is physically meaningful, highly interpretable, and easily comparable across different regions. Specifically, the township spatiotemporal development state category is used as a binary classification supervision task. This task is used to reverse-engineer the nonlinear contribution of each preset modeling indicator, i.e., the SHAP value (model-driven weight). Simultaneously, based on the data distribution characteristics of the extreme value normalization matrix, the data-driven weights (objective weights) at the objective statistical level are calculated using the entropy method. Subsequently, these two weights are dynamically fused, which not only preserves the nonlinear characteristics learned by the model, but also relies on statistical methods to ensure the objectivity of the weight values. This effectively improves the problem of lack of objective reference and large subjective bias in the traditional quantitative measurement method of weight assignment, and enhances the scientificity and stability of weight configuration.

[0082] S410. Based on the fusion weights and extreme value matrix, determine the time series sequence of the spatiotemporal development characteristics of the township to be identified over time. The time series sequence of spatiotemporal development characteristics includes the spatiotemporal development characteristic representation values ​​corresponding to each year during the target year.

[0083] In some embodiments, step S410 specifically includes: Group all extreme values ​​corresponding to the same year in the extreme value matrix into one group to obtain the extreme value groups corresponding to each year during the target year period; The extreme values ​​in the extreme value group are weighted and summed according to the fusion weight of each indicator, and the sum is used as the spatiotemporal development characteristic value of the corresponding year. The spatiotemporal development characteristic values ​​are combined into a sequence according to the chronological order to obtain the time series sequence of the spatiotemporal development characteristics of the township to be identified over time.

[0084] After calculating the fusion weights, the extreme value matrix can be grouped. Specifically, the extreme value matrix Ynorm obtained in the above steps has rows corresponding to the years within the target period and columns corresponding to the indicators. Each element in the matrix is ​​the standardized extreme value of the corresponding year and indicator. During grouping, all column elements (i.e., the extreme values ​​of all indicators under that year) in the same row (i.e., the same year) of the matrix are extracted and grouped into a set. This set is the extreme value group corresponding to that year, providing basic data for subsequent calculation of spatiotemporal development characteristic representation values ​​by year.

[0085] Next, based on the fusion weight Together with the aforementioned extreme value group, the spatiotemporal development characteristic values ​​of the township to be identified in each year of the target year are calculated by weighted summation, thereby obtaining the time series sequence of spatiotemporal development characteristics changing over time. The specific calculation formula is as follows: ; in, Let t be the spatiotemporal development characteristic representation value of the township corresponding to time t (i.e., the t-th year within the target year period); , Here, represents the extreme value of the 'a'-th preset modeling indicator in year t (one of the extreme value groups corresponding to year t) and the fusion weight, respectively; D represents the total number of indicators (such as the six dimensions of indicators mentioned earlier). The spatiotemporal development characteristic representation value calculated by this formula can comprehensively reflect the multidimensional spatiotemporal element coupling state and comprehensive development characteristics of the township to be identified at each time series node. The larger the value, the more reasonable the spatiotemporal element configuration, the more significant the spatiotemporal development coupling characteristics, and the better the comprehensive spatiotemporal development state of the township in the corresponding year.

[0086] It should be noted that although the target Transformer model constructs a nonlinear mapping relationship between indicator features and the spatiotemporal development characteristics of townships, this embodiment specifically adopts a linear weighted form to calculate the spatiotemporal development characteristic representation values ​​in order to ensure that the final spatiotemporal development characteristic representation values ​​have horizontal comparability across townships and years, and to facilitate subsequent differential comparisons and hierarchical classifications of township spatiotemporal development characteristics. In this process, the SHAP value (model-driven weight) acts as the core medium for "linearizing the projection of nonlinear laws," realizing the effective transfer of the importance of nonlinear features learned by the model to a quantifiable linear evaluation system, while taking into account both the model's nonlinear feature fitting ability and the standardization and comparability of the exponential quantification results.

[0087] Furthermore, the Z-Score standardization used in the model training samples and the Min-Max extreme value standardization used in the exponent calculation process (i.e., the processing method in step S407) only have a linear scaling relationship and will not change the feature importance ranking of each preset modeling index. Therefore, the index importance rules learned by the target Transformer model can be directly transferred to the exponent calculation process, effectively ensuring the consistency between the model rules and the exponent calculation, and providing core support for the accurate calculation of spatiotemporal development feature representation values.

[0088] S411. Based on the time series of the spatiotemporal development characteristics and the preset Bayesian segmented regression model, determine the evolution trajectory of the spatiotemporal development characteristics of the township to be identified during the target year.

[0089] Among them, the Bayesian segmented regression model is used to take the spatiotemporal development characteristic representation value as the observation value, and determine the posterior probability of each year as the turning point year within the target year period based on the observation likelihood and prior probability, thereby identifying the temporal segmentation characteristics and abrupt change inflection points of the spatiotemporal development characteristic representation value, and completing the refined and data-driven characterization of the evolution trajectory of township spatiotemporal development characteristics.

[0090] Furthermore, step S411 specifically includes: The parameters of the pre-defined Bayesian piecewise regression model are initialized. The initialized parameters include the prior distribution parameters, which are used to quantify the initial probability of each year during the target year period as a turning point year under the assumption of no mutation. Based on the initialized Bayesian piecewise regression model, the spatiotemporal development feature characterization values ​​in the time series of spatiotemporal development features are used as observation values ​​to determine the observation likelihood of each year as a turning point year within the target year period. Based on the prior distribution parameters and the observed likelihood, the posterior probability of each year as a turning point is determined, and the significance of the posterior probability of each year is tested to determine at least one time inflection point. Based on this inflection point, the development stages of the target year are divided to obtain the spatiotemporal development characteristics evolution trajectory that includes multiple different development stages.

[0091] Considering that townships may experience multiple evolutionary stages during their decades-long spatiotemporal development, the Bayesian piecewise regression model designed in this embodiment adaptively identifies single or multiple structural mutations in the index sequence through a recursive strategy, thereby dividing the development stages and constructing the spatiotemporal development characteristic evolution trajectory. The specific implementation steps of its core logic are as follows: 1) Interval Initialization and Single-Point Assumption: The time interval corresponding to the spatiotemporal development characteristic time series is defined as... (Initially the entire time domain of the target year) Where Ttotal is the total number of years in the target year period. Also, assume the current interval... Memory at a dominant mutation point The temporal evolution of the spatiotemporal development characteristic values ​​differs significantly before and after τ. The two time series data follow different linear fitting rules before and after τ, and the corresponding piecewise linear regression model is expressed as follows: ; in, Let t be the spatiotemporal development characteristic observation value. This could be a potential turning point year; inflection point The linear regression slope and intercept before the inflection point reflect the temporal evolution rate and baseline level of the spatiotemporal development characteristics of townships before the inflection point; inflection point The slope and intercept of the linear regression after the inflection point reflect the temporal evolution rate and baseline level of the spatiotemporal development characteristics of townships after the inflection point. This is a random noise that follows a Gaussian distribution, used to quantify the random error between observed values ​​and model-fitted values.

[0092] 2) Global likelihood scanning and posterior inference: Traversing the current interval For each year t in the equation, we take it as the hypothetical inflection point. (Denoteed as g), least squares linear regression is performed on two independent time series data points before and after τ (t≤τ and t>τ), respectively. The sum of squared residuals (RSS) of the two data points is calculated, and the log-likelihood of the model when the inflection point is assumed to be g is obtained. Then, Bayes' theorem is introduced, and the log-likelihood of all years is normalized using the Softmax function to obtain the posterior probability of each year as the true inflection point. The specific calculation formula is as follows: ; ; ; in: The sum of squared residuals of the piecewise linear regression is the sum of squared residuals when the inflection point is assumed to be g, which reflects the quality of the fit. The total length of the spatiotemporal development characteristics time series (i.e., the total number of years during the target year period). The observed values ​​represent the spatiotemporal development characteristics of year t. These are the linear regression predicted values ​​before and after the inflection point g, respectively; This is the log-likelihood value of the model when the inflection point is assumed to be g. The larger the value, the higher the probability that the year is the inflection point. Indicating in the observation data With the support of [the relevant factor], the year g is the posterior probability of the true inflection point.

[0093] 3) Inflection point determination and significance test: From all hypothesized inflection point years, select the year with the highest posterior probability P(τ=g|HPt). As candidate inflection points, to avoid misjudgment due to noise, the Bayesian Information Criterion (BIC) is introduced to evaluate the candidate inflection points. Perform a significance test, if If the significance condition is met (e.g., the BIC value is below the critical value), it is confirmed as a valid inflection point; otherwise, the branch search is stopped, and it is determined that there is no significant inflection point in the current interval Tcurrent.

[0094] 4) Recursive search: Once a valid inflection point is confirmed... The current time series interval Tcurrent is set to... Divide the sequence into two subsequences, left and right, denoted as and respectively. (Subsequence before the inflection point) and (Subsequence after the inflection point). Then, recursively execute steps 1) to 3) above for these two subsequences, repeating interval initialization, single-point hypothesis, likelihood scan, and inflection point test until the length of the subsequence is less than the preset minimum time span (e.g., less than 3 years, which cannot form an effective development stage), or until an effective inflection point that meets the significance condition cannot be found.

[0095] Through the above-mentioned multi-layer recursive retrieval strategy, this model can adaptively identify single or multiple structural mutations in the time series of spatiotemporal development characteristics, complete the refined automatic segmentation of the time series evolution interval, and finally summarize all effective mutation inflection points that pass the significance test and sort them by year, accurately depicting the multi-stage time series evolution characteristics of township spatiotemporal development characteristics, and realizing the refined, data-driven, and structured construction of the evolution trajectory of spatiotemporal development characteristics.

[0096] It should be noted that this embodiment, by establishing a probabilistic inflection point detection technology based on Bayesian piecewise regression, can not only accurately locate the specific temporal nodes where structural transformations and feature transitions occur in the spatiotemporal development characteristics of townships, but also simultaneously output the confidence level, i.e., the posterior probability, corresponding to the inflection point determination result. This quantifies the confidence level of the inflection point determination, realizing a technological upgrade from the traditional fixed-threshold qualitative discrimination method to a probabilistic, quantifiable, and measurable intelligent recognition method. This probabilistic recognition mechanism can effectively avoid misjudgments and errors caused by accidental fluctuations and random noise interference in time-series data, significantly improving the stability and accuracy of the inflection point identification results for time-series abrupt changes. It provides reliable technical support for the refined and credible analysis of the spatiotemporal evolution trajectory of township development.

[0097] Meanwhile, this embodiment presents the identification results by quantifying the spatiotemporal development characteristic representation values ​​and the evolution trajectory of spatiotemporal development characteristics. This achieves precise quantification of the multidimensional spatiotemporal development characteristics of townships and intuitive representation of their temporal evolution patterns, improving the intuitiveness and readability of the identification results and reducing the difficulty of interpreting the temporal evolution patterns of spatiotemporal development characteristics. Furthermore, this refined and temporally sequenced intelligent identification result can accurately reflect the characteristic differences and stage-specific evolutionary shortcomings of township spatiotemporal development, providing comprehensive, scientific, and practical technical reference for optimizing township spatial structure, rationally allocating spatiotemporal elements, and refining intelligent management and control of land space. This effectively supports the dynamic optimization and orderly evolution of the spatiotemporal development pattern of townships.

[0098] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0099] It should be noted that the scope of protection of the intelligent identification method for the spatiotemporal development characteristics of townships provided in this application is not limited to the above embodiments. Any solution achieved by adding, subtracting, or replacing steps in the prior art based on the principles of this application, according to the execution order of each embodiment, is included within the scope of protection of this application. Furthermore, any reasonable substitution or merging of steps between different embodiments based on the principles of this application is also included within the scope of protection of this application.

[0100] Based on the method described in the above embodiments, this application also provides an intelligent identification device for the spatiotemporal development characteristics of townships, used to execute the steps in the intelligent identification method for the spatiotemporal development characteristics of townships described above. Please refer to... Figure 7 , Figure 7 This is a schematic diagram of the structure of an intelligent identification device for the spatiotemporal development characteristics of a township, provided as an exemplary embodiment of this application. Specifically, the intelligent identification device 500 for the spatiotemporal development characteristics of a township includes an acquisition unit 501, a determination unit 502, and an identification unit 503, wherein: The acquisition unit 501 is used to acquire a training sample set and the spatiotemporal development status label of each training sample. The training sample set is constructed based on the first multi-source spatiotemporal perception observation data of multiple sample towns during the historical year. The first multi-source spatiotemporal perception observation data includes the first indicator observation value of multiple preset modeling indicators and the spatial geographical location of the sample town. The determination unit 502 is used to train the constructed target Transformer model using the training sample set and the spatiotemporal development state label. The target Transformer model includes at least a temporal mapping module and a spatial mapping module. The training process of the target Transformer model includes mapping the indicator features corresponding to each preset modeling indicator to the target dimension space through the temporal mapping module to obtain a temporal embedding representation, and mapping the location features corresponding to the spatial geographic location to the target dimension space through the spatial mapping module to obtain a spatial embedding representation. The spatial embedding representation is copied along the time axis and loaded onto each time step of the temporal embedding representation using a broadcast mechanism to obtain a fused spatiotemporal embedding representation. Then, the state category prediction value corresponding to the corresponding training sample is determined according to the fused spatiotemporal embedding representation. Based on the deviation between the state category prediction value and the spatiotemporal development state label, the network parameters of the target Transformer model are iteratively optimized using a preset Bayesian optimization algorithm. The identification unit 503 is used to determine the spatiotemporal development characteristics identification result of the township to be identified based on the trained target Transformer model and the second multi-source spatiotemporal perception observation data of the township to be identified.

[0101] In some embodiments, the acquisition unit 501 is further configured to: Standard deviation standardization was performed on the first indicator observations and spatial geographic locations of multiple preset modeling indicators for each sample township during the historical year period to obtain the corresponding indicator features and location features. Based on the sliding window sampling strategy, a basic sample set for each sample township is constructed according to the indicator characteristics and the location characteristics. Different basic samples in the basic sample set correspond to different time intervals. The base sample set is processed using a hybrid resampling strategy to obtain the training sample set.

[0102] In some embodiments, the acquisition unit 501 is used for: The positive samples in the base sample set are augmented, and the negative samples in the base sample set are downsampled in a context-aware manner.

[0103] In some embodiments, the second multi-source spatiotemporal sensing observation data includes second index observation values ​​of each preset modeling index, and the identification unit 503 is used for: Based on the training sample set, the trained target Transformer model, and the preset attribution algorithm, determine the model-driving weights of each preset modeling metric. Based on the second indicator observation values ​​of the townships to be identified during the target year, determine the data-driven weights of each preset modeling indicator and the extreme value matrix corresponding to all the second indicator observation values; Based on the model-driven weights, the data-driven weights, and the extreme value matrix, the spatiotemporal development characteristics of the township to be identified are determined.

[0104] In some embodiments, the identification unit 503 is used for: Determine the alignment sample group corresponding to each training sample in the training sample set, and input the training sample and the alignment sample group into the trained target Transformer model for processing. The output of the target Transformer model is analyzed by a preset attribution analysis algorithm to determine the contribution of each preset modeling index to the output. The model-driven weights of each preset modeling indicator are determined based on this contribution value.

[0105] In some embodiments, the identification unit 503 is used for: Take the absolute value of each contribution value and calculate the average value of the absolute values ​​of the same indicator to obtain the average value corresponding to each preset modeling indicator; Normalize all these average values ​​to obtain the model-driven weights corresponding to each preset modeling index.

[0106] In some embodiments, the identification unit 503 is used for: Determine the directional information for each of the preset modeling indicators, including both positive and negative directional information; Based on this directional information, the second indicator observation values ​​of each preset modeling indicator of the township to be identified during the target year are subjected to directional extreme value standardization to obtain the extreme value matrix. Based on the entropy method and the extreme value matrix, the data-driven weights of each preset modeling index are determined.

[0107] In some embodiments, the identification unit 503 is used for: The fusion weights of each preset modeling index are determined based on the model-driven weights and the data-driven weights. Based on the fusion weight and the extreme value matrix, the time series sequence of the spatiotemporal development characteristics of the township to be identified over time is determined. The time series sequence of spatiotemporal development characteristics includes the spatiotemporal development characteristic representation values ​​corresponding to each year within the target year period. Based on the time series of the spatiotemporal development characteristics and the pre-set Bayesian piecewise regression model, the evolution trajectory of the spatiotemporal development characteristics of the township to be identified during the target year is determined.

[0108] In some embodiments, the identification unit 503 is used for: Group all extreme values ​​corresponding to the same year in the extreme value matrix into one group to obtain the extreme value groups corresponding to each year during the target year period; The extreme values ​​in the extreme value group are weighted and summed according to the fusion weight of each indicator, and the sum is used as the spatiotemporal development characteristic value of the corresponding year. The spatiotemporal development characteristic values ​​are combined into a sequence according to the chronological order to obtain the time series sequence of the spatiotemporal development characteristics of the township to be identified over time.

[0109] In some embodiments, the identification unit 503 is used for: The parameters of the preset Bayesian piecewise regression model are initialized. The initialized parameters include the prior distribution parameters, which are used to quantify the initial probability of each year during the target year as a turning point year under the assumption of no mutation. Based on the initialized Bayesian piecewise regression model, the spatiotemporal development feature characterization values ​​in the time series of spatiotemporal development features are used as observation values ​​to determine the observation likelihood of each year as a turning point year within the target year period. Based on the prior distribution parameters and the observed likelihood, the posterior probability of each year as a turning point is determined, and the significance of the posterior probability of each year is tested to determine at least one time inflection point. Based on this inflection point in time, the spatiotemporal development characteristics and evolution trajectory of the township to be identified are determined.

[0110] It should be noted that the division of the unit modules in the above-mentioned intelligent identification device for the spatiotemporal development characteristics of townships is only for illustrative purposes. In other embodiments, the intelligent identification device for the spatiotemporal development characteristics of townships can be divided into different unit modules as needed to complete all or part of the functions of the intelligent identification device for the spatiotemporal development characteristics of townships. The implementation of each unit module in the intelligent identification device for the spatiotemporal development characteristics of townships provided in the embodiments of this specification can be in the form of a computer program. The computer program can run on a terminal or server. The program modules constituted by the computer program can be stored in the memory of the terminal or server. When the computer program is executed by the processor, it implements all or part of the steps of the intelligent identification method for the spatiotemporal development characteristics of townships described in the embodiments of this specification.

[0111] Please see below. Figure 8 , Figure 8 This is a schematic diagram of the structure of an electronic device provided for an exemplary embodiment of this application. For example... Figure 8As shown, the electronic device 600 may include at least one processor 610, at least one network interface 620, user interface 630, memory 640, and at least one communication bus 650.

[0112] The communication bus 650 is used to enable communication between these components.

[0113] The network interface 620 may include, but is not limited to, a low-power Bluetooth module, a near field communication (NFC) module, a wireless Fidelity (Wi-Fi) module, etc.

[0114] The user interface 630 may include a display screen and a camera. Optionally, the user interface 630 may also include a standard wired interface and a wireless interface.

[0115] The processor 610 may include one or more processing cores. The processor 610 connects to various parts within the electronic device 600 using various interfaces and lines, and performs various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 640, and by calling data stored in the memory 640. Optionally, the processor 610 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 610 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 610 and may be implemented as a separate chip.

[0116] The memory 640 may include random access memory (RAM) or read-only memory (ROM). Optionally, the memory 640 may include a non-transitory computer-readable storage medium. The memory 640 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 640 may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for at least one function (such as voiceprint direction recognition, voice separation, personalized data storage, etc.), and instructions for implementing the various method embodiments described above. The data storage area may store data involved in the various method embodiments described above. Optionally, the memory 640 may also be at least one storage device located remotely from the aforementioned processor 610. Figure 8 As shown, the memory 640, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and program instructions.

[0117] exist Figure 8 In the electronic device 600 shown, the user interface 630 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 610 can be used to call the program instructions stored in the memory 640. The electronic device 600 is used to execute the intelligent identification method of the spatiotemporal development characteristics of townships described in any of the above embodiments.

[0118] This application also provides a computer storage medium storing instructions that, when run on a computer or processor, cause the computer or processor to execute one or more steps of any of the above methods. If the constituent modules of the aforementioned intelligent identification device for the spatiotemporal development characteristics of townships are implemented as software functional units and sold or used as independent products, they can be stored in this storage medium.

[0119] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. This computer program product includes one or more computer instructions. When these computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., Digital Versatile Discs (DVDs)), or semiconductor media (e.g., solid-state disks (SSDs)).

[0120] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks. Unless otherwise specified, the technical features of this embodiment and its implementation can be combined arbitrarily.

[0121] The above embodiments are merely descriptions of preferred embodiments of this application and are not intended to limit the scope of this application. Any modifications and improvements made by those skilled in the art to the technical solutions of this application without departing from the spirit of this application should fall within the protection scope defined by the claims of this application.

Claims

1. A method for intelligent identification of the spatiotemporal development characteristics of townships, characterized in that, The method includes: Obtain a training sample set and spatiotemporal development status labels for each training sample. The training sample set is constructed based on the first multi-source spatiotemporal perception observation data of multiple sample towns during historical years. The first multi-source spatiotemporal perception observation data includes the first indicator observation values ​​of multiple preset modeling indicators and the spatial geographical location of the sample towns. Using the training sample set and the spatiotemporal development state labels, the constructed target Transformer model is trained. The target Transformer model includes at least a temporal mapping module and a spatial mapping module. The training process of the target Transformer model includes mapping the indicator features corresponding to each preset modeling indicator to the target dimension space through the temporal mapping module to obtain a temporal embedding representation, and mapping the location features corresponding to the spatial geographic location to the target dimension space through the spatial mapping module to obtain a spatial embedding representation. The spatial embedding representation is then copied along the time axis and loaded onto each time step of the temporal embedding representation using a broadcast mechanism to obtain a fused spatiotemporal embedding representation. Then, the state category prediction value corresponding to the corresponding training sample is determined based on the fused spatiotemporal embedding representation, and the network parameters of the target Transformer model are iteratively optimized using a preset Bayesian optimization algorithm based on the deviation between the state category prediction value and the spatiotemporal development state label. Based on the trained target Transformer model and the second multi-source spatiotemporal perception observation data of the township to be identified, the spatiotemporal development characteristics identification results of the township to be identified are determined.

2. The method according to claim 1, characterized in that, The steps for constructing the training sample set include: Standard deviation standardization was performed on the first indicator observations and spatial geographic locations of multiple preset modeling indicators for each sample township during the historical year period to obtain the corresponding indicator features and location features. Based on the sliding window sampling strategy, a basic sample set for each sample township is constructed according to the indicator features and the location features. Different basic samples in the basic sample set correspond to different time intervals. The base sample set is processed using a hybrid resampling strategy to obtain the training sample set.

3. The method according to claim 2, characterized in that, The processing of the base sample set based on the hybrid resampling strategy includes: The positive samples in the base sample set are enhanced, and the negative samples in the base sample set are subjected to context-aware downsampling.

4. The method according to claim 1, characterized in that, The second multi-source spatiotemporal sensing observation data includes the second indicator observation values ​​of each preset modeling index. The determination of the spatiotemporal development characteristics identification result of the township to be identified, based on the trained target Transformer model and the second multi-source spatiotemporal sensing observation data, includes: Based on the training sample set, the trained target Transformer model, and the preset attribution algorithm, determine the model-driving weights of each preset modeling metric; Based on the observed values ​​of the second indicator of the township to be identified during the target year, determine the data-driven weights of each preset modeling indicator and the extreme value matrix corresponding to all the observed values ​​of the second indicator; Based on the model-driven weights, the data-driven weights, and the extreme value matrix, the spatiotemporal development characteristics identification results of the township to be identified are determined.

5. The method according to claim 4, characterized in that, The step of determining the model-driving weights of each preset modeling metric based on the training sample set, the trained target Transformer model, and the preset attribution algorithm includes: Determine the comparison sample group corresponding to each training sample in the training sample set, and input the training sample and the comparison sample group into the trained target Transformer model for processing; The output of the target Transformer model is analyzed by a preset attribution analysis algorithm to determine the contribution of each preset modeling index to the output. The model-driven weights of each preset modeling indicator are determined based on the contribution value.

6. The method according to claim 4, characterized in that, The step of determining the data-driven weights of each preset modeling indicator and the extreme value matrix corresponding to all the second indicator observations based on the observed values ​​of the township to be identified during the target year includes: Determine the directional information of each of the preset modeling indicators, wherein the directional information includes positive and negative; Based on the directional information, the second indicator observation values ​​of each preset modeling indicator of the township to be identified during the target year are subjected to directional extreme value standardization processing to obtain the extreme value matrix. Based on the entropy method and the extreme value matrix, the data-driven weights of each preset modeling index are determined.

7. The method according to claim 4, characterized in that, The step of determining the spatiotemporal development characteristic identification result of the township to be identified based on the model-driven weights, the data-driven weights, and the extreme value matrix includes: The fusion weights of each preset modeling index are determined based on the model-driven weights and the data-driven weights. Based on the fusion weights and the extreme value matrix, the time series sequence of the spatiotemporal development characteristics of the township to be identified over time is determined. The time series sequence of spatiotemporal development characteristics includes the spatiotemporal development characteristic representation values ​​corresponding to each year within the target year period. Based on the spatiotemporal development characteristics time series and the preset Bayesian piecewise regression model, the spatiotemporal development characteristics evolution trajectory of the township to be identified during the target year is determined.

8. The method according to claim 7, characterized in that, The step of determining the time-series sequence of the spatiotemporal development characteristics of the township to be identified over time based on the fusion weights and the extreme value matrix includes: Group all extreme values ​​corresponding to the same year in the extreme value matrix into one group to obtain the extreme value groups corresponding to each year during the target year period; The extreme values ​​in the extreme value group are weighted and summed according to the fusion weights of each indicator, and the sum is used as the spatiotemporal development characteristic value of the corresponding year. The spatiotemporal development characteristic values ​​are combined into a sequence according to the chronological order to obtain the time series sequence of the spatiotemporal development characteristics of the township to be identified as it changes over time.

9. The method according to claim 7, characterized in that, The determination of the spatiotemporal development characteristic evolution trajectory of the township to be identified during the target year, based on the spatiotemporal development characteristic time series and a preset Bayesian piecewise regression model, includes: The parameters of the preset Bayesian piecewise regression model are initialized. The initialized parameters include prior distribution parameters, which are used to quantify the initial probability of each year during the target year period as a turning point year under the assumption of no mutation. Based on the initialized Bayesian piecewise regression model, the spatiotemporal development feature representation values ​​in the spatiotemporal development feature time series are used as observation values ​​to determine the observation likelihood of each year as a turning point year within the target year period. Based on the prior distribution parameters and the observed likelihood, the posterior probability of each year as a turning point is determined, and the significance of the posterior probability of each year is tested to determine at least one time inflection point. The spatiotemporal development characteristics of the township to be identified are determined based on the aforementioned inflection point in time.

10. An intelligent identification device for the spatiotemporal development characteristics of townships, characterized in that, The device includes: The acquisition unit is used to acquire a training sample set and the spatiotemporal development status label of each training sample. The training sample set is constructed based on the first multi-source spatiotemporal perception observation data of multiple sample towns during the historical year. The first multi-source spatiotemporal perception observation data includes the first indicator observation value of multiple preset modeling indicators and the spatial geographical location of the sample towns. A determination unit is used to train a constructed target Transformer model using the training sample set and the spatiotemporal development state labels. The target Transformer model includes at least a temporal mapping module and a spatial mapping module. The training process of the target Transformer model includes mapping the indicator features corresponding to each preset modeling indicator to the target dimension space through the temporal mapping module to obtain a temporal embedding representation, and mapping the location features corresponding to the spatial geographic location to the target dimension space through the spatial mapping module to obtain a spatial embedding representation. The spatial embedding representation is then copied along the time axis and loaded onto each time step of the temporal embedding representation using a broadcast mechanism to obtain a fused spatiotemporal embedding representation. Then, the state category prediction value corresponding to the corresponding training sample is determined based on the fused spatiotemporal embedding representation, and the network parameters of the target Transformer model are iteratively optimized using a preset Bayesian optimization algorithm based on the deviation between the state category prediction value and the spatiotemporal development state label. The identification unit is used to determine the spatiotemporal development characteristics identification result of the township to be identified based on the trained target Transformer model and the second multi-source spatiotemporal sensing observation data of the township to be identified.

11. An electronic device, characterized in that, include: Memory, used to store executable program code; A processor is configured to call and run the executable program code from the memory, causing the electronic device to perform the intelligent identification method for the spatiotemporal development characteristics of townships as described in any one of claims 1 to 9.

Images