Shared electric vehicle scheduling system and method based on big data analysis

Abstract

The invention relates to the field of electric vehicle scheduling, and provides a shared electric vehicle scheduling system and method based on big data analysis, and the method comprises the steps: firstly extracting the historical demand time sequence fluctuation characteristics of each grid region, and then quantifying the road connectivity and spatial proximity between the regions through constructing a grid region spatial topology distribution matrix, then, capturing a demand conduction effect formed by population flow and road topology in adjacent areas through a graph structure, and then introducing a space-time reasoning model to simulate a diffusion path of chained influence of sudden factors such as extreme weather and temporary activities on multi-area demands so as to obtain an electric vehicle demand predicted value of each grid area; and finally, dynamically comparing the difference between the demand thermodynamic diagram and the current vehicle distribution diagram to generate a scheduling strategy. In this way, space-time optimal configuration of vehicle resources of the whole road network can be achieved.

Description

Technical Field

[0001] This application relates to the field of electric vehicle dispatching, and more specifically, to a shared electric vehicle dispatching system and method based on big data analysis. Background Art

[0002] With the acceleration of urbanization and the popularity of the concept of low-carbon travel, shared electric vehicles, as a new type of micro-mobility, are playing an increasingly important role in daily urban travel. However, in actual operation, the problem of mismatch between vehicle supply and demand in space and time has gradually become the main obstacle restricting its service efficiency.

[0003] Traditional methods for processing such supply and demand data usually rely on relatively basic algorithms such as moving average or linear regression. These methods have obvious limitations: they are difficult to capture the complex spatio-temporal correlation features between different regions, and they cannot accurately analyze the unique supply and demand change patterns of various functional areas such as commercial areas and residential areas. More importantly, existing systems often discretize the geographical space into isolated grid cells for analysis, ignoring the dynamic demand transfer effect between adjacent regions due to road connectivity and population flow. For example, during the morning rush hour, there may be a shortage of shared electric vehicles around subway stations, while there are a large number of idle vehicles in neighboring communities. This lack of spatio-temporal collaborative modeling ability not only leads to the inability of dispatching strategies to respond promptly to changes in actual demand, but also makes operating enterprises face the dual challenges of untimely replenishment in hot spots and resource waste in non-hot spots.

[0004] Therefore, an optimized shared electric vehicle dispatching solution is desired. Summary of the Invention

[0005] This application aims at the deficiencies in the prior art and provides a shared electric vehicle dispatching system and method based on big data analysis.

[0006] According to one aspect of the present application, there is provided a shared electric vehicle scheduling method based on big data analysis, which includes: obtaining historical data of the electric vehicle demand in each grid area; calculating the grid area spatial topology distribution matrix of each grid area; performing spatio-temporal collaborative correlation analysis on the historical data of the electric vehicle demand in each grid area and the grid area spatial topology distribution matrix to obtain the electric vehicle demand prediction value of each grid area, including: performing spatio-temporal encoding of the electric vehicle demand in each grid area and the grid area spatial topology distribution matrix to obtain a set of spatio-temporal collaborative correlation features of the electric vehicle demand; performing spatio-temporal inference encoding and decoding of the electric vehicle demand based on structured recursion on the set of spatio-temporal collaborative correlation features of the electric vehicle demand to obtain the electric vehicle demand prediction value of each grid area; generating a demand heat map based on the electric vehicle demand prediction value of each grid area; obtaining the current number of electric vehicles in each grid area, and generating a current vehicle distribution map based on the current number of electric vehicles in each grid area; generating a dynamic scheduling strategy based on the demand heat map and the current vehicle distribution map.

[0007] According to another aspect of the present application, there is provided a shared electric vehicle scheduling system based on big data analysis, which includes: a historical data acquisition module for obtaining historical data of the electric vehicle demand in each grid area; a spatial topology matrix calculation module for calculating the grid area spatial topology distribution matrix of each grid area; a demand prediction module for performing spatio-temporal collaborative correlation analysis on the historical data of the electric vehicle demand in each grid area and the grid area spatial topology distribution matrix to obtain the electric vehicle demand prediction value of each grid area, wherein the demand prediction module is used for: performing spatio-temporal encoding of the electric vehicle demand in each grid area and the grid area spatial topology distribution matrix to obtain a set of spatio-temporal collaborative correlation features of the electric vehicle demand; performing spatio-temporal inference encoding and decoding of the electric vehicle demand based on structured recursion on the set of spatio-temporal collaborative correlation features of the electric vehicle demand to obtain the electric vehicle demand prediction value of each grid area; a demand heat map generation module for generating a demand heat map based on the electric vehicle demand prediction value of each grid area; a current vehicle distribution map generation module for obtaining the current number of electric vehicles in each grid area, and generating a current vehicle distribution map based on the current number of electric vehicles in each grid area; a scheduling strategy generation module for generating a dynamic scheduling strategy based on the demand heat map and the current vehicle distribution map.

[0008] Due to the adoption of the above technical solutions, this application has remarkable technical effects: The shared electric vehicle scheduling system and method based on big data analysis provided by this application first extracts the historical demand time series fluctuation characteristics of each grid area, then quantifies the road connectivity and spatial proximity between regions by constructing a grid area spatial topology distribution matrix, and then captures the demand conduction effect formed by population flow and road topology between adjacent regions through a graph structure. Subsequently, a spatio-temporal reasoning model is introduced to simulate the diffusion path of the chain effect of sudden factors such as extreme weather and temporary activities on the multi-region demand, so as to obtain the electric vehicle demand prediction value of each grid area. Finally, a scheduling strategy is generated by dynamically comparing the differences between the demand heat map and the current vehicle distribution map. In this way, it helps to achieve the spatio-temporal optimal allocation of vehicle resources across the entire road network. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other objects, features, and advantages of the present application will become more obvious. The accompanying drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 It is a flowchart of the shared electric vehicle scheduling method based on big data analysis according to an embodiment of the present application.

[0011] Figure 2 It is a flowchart of step S3 in the shared electric vehicle scheduling method based on big data analysis according to an embodiment of the present application.

[0012] Figure 3 It is a flowchart of step S31 in the shared electric vehicle scheduling method based on big data analysis according to an embodiment of the present application.

[0013] Figure 4 It is a flowchart of step S32 in the shared electric vehicle scheduling method based on big data analysis according to an embodiment of the present application.

[0014] Figure 5 It is a flowchart of step S321 in the shared electric vehicle scheduling method based on big data analysis according to an embodiment of the present application.

[0015] Figure 6 It is a flowchart of step S321-2 in the shared electric vehicle scheduling method based on big data analysis according to an embodiment of the present application.

[0016] Figure 7 It is a system block diagram of the shared electric vehicle scheduling system based on big data analysis according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0018] With the advancement of urbanization and the popularization of the concept of low-carbon travel, shared electric vehicles have become increasingly important in urban daily travel. However, in actual operation, the spatio-temporal mismatch problem between vehicle supply and demand limits its service efficiency. When traditional methods process supply and demand data, they mostly use basic algorithms such as moving average and linear regression, which have obvious limitations. It is difficult to capture the complex spatio-temporal correlation characteristics in different regions, unable to analyze the unique supply and demand change patterns of various functional areas, and ignores the dynamic demand transfer effect caused by road connectivity and population flow between adjacent regions. This results in the scheduling strategy being unable to respond to demand changes in a timely manner, putting the operating enterprises in a double dilemma of untimely replenishment in hot spots and resource waste in non-hot spots.

[0019] To address the above technical problems, the technical concept of the present application is to use data processing and prediction algorithms based on big data analysis and artificial intelligence. Specifically, first, the urban space is divided into grid regions. Based on extracting the temporal fluctuation characteristics of historical demand in each region, the road connectivity and spatial proximity between regions are quantified by constructing a spatial topological distribution matrix, thus transforming the isolated analysis caused by geographical space discretization into a dynamic association network. Then, the demand conduction effect formed by population flow and road topology between adjacent regions is captured through a graph structure (for example, when the demand surges around a subway station, vehicles in adjacent communities can quickly replenish through the main road). Then, a spatio-temporal reasoning model is introduced to simulate the diffusion path of the chain effect of sudden factors such as extreme weather and temporary activities on the demand of multiple regions, so as to obtain the predicted value of the electric vehicle demand in each grid region. Finally, by dynamically comparing the difference between the demand heat map and the current vehicle distribution map, a scheduling strategy is generated. This technical solution effectively solves the problem of "local prediction deviation diffusion" caused by isolated regional analysis in traditional methods, enables the scheduling instructions to adaptively respond to the dynamic characteristics of cross-regional demand transfer, ensures timely replenishment in hot spots while avoiding vehicle idle in communities, and realizes the spatio-temporal optimal allocation of vehicle resources across the entire road network.

[0020] Figure 1 It is a flowchart of a shared electric vehicle scheduling method based on big data analysis according to an embodiment of the present application. As Figure 1As shown in the figure, the shared electric vehicle scheduling method based on big data analysis according to the embodiments of the present application includes: S1, obtaining historical data on the demand for electric vehicles in each grid area; S2, calculating the grid area spatial topology distribution matrix of each grid area; S3, performing spatio-temporal collaborative correlation analysis on the historical data of the demand for electric vehicles in each grid area and the grid area spatial topology distribution matrix to obtain the predicted value of the demand for electric vehicles in each grid area; S4, generating a demand heat map based on the predicted value of the demand for electric vehicles in each grid area; S5, obtaining the current number of electric vehicles in each grid area, and generating a current vehicle distribution map based on the current number of electric vehicles in each grid area; S6, generating a dynamic scheduling strategy based on the demand heat map and the current vehicle distribution map.

[0021] In step S1, historical data on the demand for electric vehicles in each grid area is obtained. It should be understood that the historical data on the demand for electric vehicles in each grid area usually includes vehicle usage records in each grid area in the continuous time dimension, specifically reflected as demand intensity values at different time granularities, such as core indicators like the number of vehicle borrowings and returns per hour or per half hour. This time span usually covers the complete periodic seasonal changes to cover multi-dimensional patterns such as weekdays and holidays, day and night alternation, and special events. In short, by obtaining the historical data on the demand for electric vehicles, a quantitative cognitive system of urban traffic travel patterns can be established. By deeply analyzing this data, the model can identify the demand fluctuation patterns in different regions at specific times, such as the concentrated vehicle usage characteristics in the business district during the morning and evening rush hours on weekdays, the off-peak travel trend in the residential area on weekends, and the tidal vehicle usage pattern around transportation hubs. These patterns provide learning samples for the prediction model, enabling the algorithm to capture hidden spatio-temporal correlations, such as the demand conduction effect between adjacent grids and the demand synchronization of similar functional areas. At the same time, the abnormal fluctuation records (such as sudden demand changes) in the historical data help to enhance the model's adaptability to special scenarios. That is, in the dynamic scheduling scenario, the accurate prediction of future electric vehicle demand can be realized based on the historical data of the demand for electric vehicles, ensuring that the scheduling strategy can effectively respond to the supply-demand mismatch problems reflected in the historical data (such as the shortage of supply around the subway station during peak hours while there are idle vehicles piling up in the adjacent community).

[0022] In step S2, calculate the grid area spatial topology distribution matrix of each grid area. Specifically, in the embodiment of the present application, the values at each position in the non-diagonal positions of the grid area spatial topology distribution matrix are the spatial distances between two grid areas. Correspondingly, considering the real-world scenario, the user's riding behavior is naturally spatially continuous, and dynamic interactions are formed between adjacent areas through road networks, population flows, or complementary functions. For example, the high demand around the subway station often requires vehicles in the adjacent residential areas to be quickly replenished through the main roads. However, due to the lack of quantitative representation of the spatial relationship between regions in the prior art, the system cannot identify such cross-regional demand conduction paths, resulting in the lag of the scheduling strategy and resource misallocation. The essential defect of this traditional fragmented analysis is that it ignores the underlying constraint of the spatial topology structure on demand changes - the spatial distance and road connectivity between regions directly determine the feasibility and efficiency of vehicle scheduling. Therefore, in the technical solution of the present application, calculate the grid area spatial topology distribution matrix of each grid area. In particular, the values at each position in the non-diagonal positions of the grid area spatial topology distribution matrix are the spatial distances between two grid areas. In this way, calculating the grid area spatial topology distribution matrix can associate these isolated grid areas through spatial distances and construct a dynamic association network. This can better capture the complex spatio-temporal association characteristics between different regions, enabling subsequent analysis and models to consider the mutual influence between regions and avoiding the deviation caused by isolated analysis.

[0023] In step S3, perform spatio-temporal collaborative association analysis on the historical data of the electric vehicle demand in each grid area and the grid area spatial topology distribution matrix to obtain the electric vehicle demand prediction value for each grid area. Specifically, Figure 2 The flowchart of step S3 in the shared electric vehicle scheduling method based on big data analysis according to the embodiment of the present application is as follows. As Figure 2 shown, step S3 includes: S31, perform spatio-temporal encoding on the historical data of the electric vehicle demand in each grid area and the grid area spatial topology distribution matrix to obtain a set of spatio-temporal collaborative association characteristics of the electric vehicle demand; S32, perform spatio-temporal reasoning encoding and decoding of the electric vehicle demand based on structured recursion on the set of spatio-temporal collaborative association characteristics of the electric vehicle demand to obtain the electric vehicle demand prediction value for each grid area.

[0024] In step S31, perform spatio-temporal encoding on the historical data of the electric vehicle demand in each grid area and the grid area spatial topology distribution matrix to obtain a set of spatio-temporal collaborative association characteristics of the electric vehicle demand. Specifically, Figure 3 The flowchart of step S31 in the shared electric vehicle scheduling method based on big data analysis according to the embodiment of the present application is as follows. As Figure 3As shown, step S31 includes: S311, performing historical demand feature time series encoding on the historical data of the electric vehicle demand in each grid area to obtain a set of electric vehicle grid area demand time series feature encoding vectors; S312, inputting the set of electric vehicle grid area demand time series feature encoding vectors and the grid area spatial topology distribution matrix into the electric vehicle demand spatio-temporal context encoder based on the graph convolutional neural network model to obtain a set of electric vehicle demand spatio-temporal collaborative association encoding vectors as the set of electric vehicle demand spatio-temporal collaborative association features.

[0025] In step S311, historical demand feature time series encoding is performed on the historical data of the electric vehicle demand in each grid region to obtain a set of electric vehicle grid region demand time series feature encoding vectors. Specifically, in the embodiment of the present application, step S311 includes: performing historical demand feature time series encoding based on a bidirectional gated recurrent unit on the historical data of the electric vehicle demand in each grid region to obtain the set of electric vehicle grid region demand time series feature encoding vectors. Correspondingly, considering that the original historical data of electric vehicle demand is complex and diverse, containing various information at different times and in different grid regions. However, in traditional shared electric vehicle demand prediction, time series data analysis is often limited to simple statistical feature extraction, such as calculating the mean or variance through a sliding window. Although such methods can reflect short-term fluctuation trends, they cannot effectively model the multi-scale time dependence relationships implicit in user travel demands - such as the rigid demand during the morning rush hour on weekdays in commercial areas, the change in travel intensity on weekends in residential areas, and the surge in demand around scenic spots during holidays and other non-linear time series patterns. A deeper problem is that existing methods usually treat time series in different regions as independent variables, ignoring the differential impact of regional functional attributes (such as transportation hubs, office clusters) on the form of time series fluctuations. For example, the vehicle usage demand in the area around a subway station is often strongly correlated with the commuting schedule, and its time series will exhibit steep peak characteristics, while the vehicle usage demand in the area around a park may show a gentle weekend periodic fluctuation. This decoupled analysis of time series features and regional attributes makes it difficult for the prediction model to capture the demand mutation rules in specific scenarios. Based on this, the present application performs historical demand feature time series encoding on the historical data of the electric vehicle demand in each grid region to obtain a set of electric vehicle grid region demand time series feature encoding vectors. In particular, in a specific example of the present application, historical demand feature time series encoding based on a bidirectional gated recurrent unit is performed on the historical data of the electric vehicle demand in each grid region to obtain the set of electric vehicle grid region demand time series feature encoding vectors. That is, the bidirectional gated recurrent unit can process both forward and backward time series information simultaneously, adaptively capture features at different time steps through a gating mechanism, and effectively mine the complex time dependence relationships in historical data. For example, it can consider both the influence of past moments on the current moment and the influence of the current moment on future moments, thereby more comprehensively describing the variation law of electric vehicle demand over time.

[0026] In step S312, the set of the electric vehicle grid area demand time-series feature encoding vectors and the grid area spatial topological distribution matrix are input into the electric vehicle demand spatio-temporal context encoder based on the graph convolutional neural network model to obtain the set of electric vehicle demand spatio-temporal collaborative association encoding vectors as the set of the electric vehicle demand spatio-temporal collaborative association features. Accordingly, considering that the demand for shared electric vehicles has obvious spatio-temporal characteristics, neither simple time-series information nor spatial information is sufficient to comprehensively describe its demand changes. The set of the electric vehicle grid area demand time-series feature encoding vectors reflects the variation law of the demand in the time dimension, while the grid area spatial topological distribution matrix describes the spatial relationship between each grid area. For example, the vehicle shortage phenomenon during the morning rush hour at the subway station is essentially a spatio-temporal coupling process in which the commuting flow migrates directionally from the residential area to the transportation hub via the road network topological structure. Therefore, in order to capture the spatio-temporal collaborative association features of the shared electric vehicle demand, in this application, the set of the electric vehicle grid area demand time-series feature encoding vectors and the grid area spatial topological distribution matrix are input into the electric vehicle demand spatio-temporal context encoder based on the graph convolutional neural network model to obtain the set of electric vehicle demand spatio-temporal collaborative association encoding vectors as the set of the electric vehicle demand spatio-temporal collaborative association features. Specifically, first, the electric vehicle grid area demand time-series feature encoding vectors representing the time-series fluctuation law of each area are mapped into graph node features, and at the same time, the edge weights between nodes are defined by the grid area spatial topological distribution matrix to construct a dynamic spatio-temporal graph structure. Under this framework, the graph convolutional neural network model synchronously mines two association modes through a multi-layer message passing mechanism: in the spatial dimension, according to the road connection distance, the feature information of adjacent areas is dynamically aggregated (for example, the subway station node absorbs the available vehicle features of the surrounding community nodes); in the time dimension, the cross-area demand evolution law is transmitted through the node feature update (for example, the evening rush hour feature of the business district node spreads to the adjacent dining area nodes). This two-way interaction enables the feature vector of each node to contain both its own time-series pattern and the spatio-temporal state of adjacent areas, forming a collaborative representation with context awareness ability.

[0027] In step S32, based on structured recursion, the electric vehicle demand spatio-temporal inference encoding and decoding are performed on the set of the electric vehicle demand spatio-temporal collaborative association features to obtain the electric vehicle demand prediction values of each grid area. Specifically, Figure 4 is a flowchart of step S32 in the shared electric vehicle scheduling method based on big data analysis according to an embodiment of the present application. As Figure 4 shown, the step S32 includes: S321, performing electric vehicle demand spatio-temporal inference based on structured recursion on the set of the electric vehicle demand spatio-temporal collaborative association encoding vectors to obtain the electric vehicle demand spatio-temporal state inference encoding vectors; S322, performing feature decoding on the electric vehicle demand spatio-temporal state inference encoding vectors to obtain the electric vehicle demand prediction values of each grid area.

[0028] In step S321, perform spatio-temporal reasoning of electric vehicle demand based on structured recursion on the set of spatio-temporal collaborative association coding vectors of the electric vehicle demand to obtain a spatio-temporal state inference coding vector of the electric vehicle demand. Specifically, Figure 5 is a flowchart of step S321 in the shared electric vehicle scheduling method based on big data analysis according to an embodiment of the present application. As Figure 5 shown, the step S321 includes: S321-1, performing recurrent neural network sequence coding on the set of spatio-temporal collaborative association coding vectors of the electric vehicle demand to obtain a set of initial coding vectors for sequence-transmitted spatio-temporal collaborative association of the electric vehicle demand; S321-2, calculating the spatio-temporal collaborative association feature message passing spatio-temporal collaborative constraint factors of each sequence-transmitted spatio-temporal collaborative association initial coding vector in the set of initial coding vectors for sequence-transmitted spatio-temporal collaborative association of the electric vehicle demand; S321-3, based on the spatio-temporal collaborative association feature message passing spatio-temporal collaborative constraint factors, performing message passing structure modulation on each sequence-transmitted spatio-temporal collaborative association initial coding vector to obtain a set of sequence-transmitted spatio-temporal collaborative association feature structure modulation coding vectors; S321-4, calculating the position-wise summation of the set of sequence-transmitted spatio-temporal collaborative association feature structure modulation coding vectors to obtain the spatio-temporal state inference coding vector of the electric vehicle demand.

[0029] It should be understood that although the set of spatio-temporal collaborative association coding vectors of the electric vehicle demand has integrated spatio-temporal information, there may still be deeper spatio-temporal relationships that have not been fully explored. Traditionally, simple feature concatenation or weighted fusion methods are often used to process spatio-temporal association information. Although this method can achieve shallow feature interaction, it is difficult to capture the causal conduction mechanism of spatio-temporal dynamics in complex scenarios. Therefore, in order to capture the dynamic change rules of electric vehicle demand between grid regions at different time periods and their mutual influence mechanism, so as to more comprehensively understand the spatio-temporal characteristics of electric vehicle demand, the present application performs spatio-temporal reasoning of electric vehicle demand based on structured recursion on the set of spatio-temporal collaborative association coding vectors of the electric vehicle demand to obtain a spatio-temporal state inference coding vector of the electric vehicle demand. In this way, the obtained inference vector is a highly generalized and abstract representation of the spatio-temporal state of the electric vehicle demand. It synthesizes the previously extracted spatio-temporal collaborative association features and further extracts the core information about the electric vehicle demand through the processing of the spatio-temporal reasoning model.

[0030] Specifically, first, map the spatio-temporal collaborative coding vector into dynamic evolution nodes, use a recurrent neural network to extract the inherent temporal law of the demand fluctuation of each node (such as the periodic strengthening of the demand during the morning rush hour in the business district), synchronously calculate the confidence weight factor in the time dimension - quantify the reliability of the node historical data (such as the impact of abnormal demand on the prediction credibility during consecutive rainy days), and evaluate the structural importance of the node in the network topology with the spatial topology confidence scale factor (such as the radiation ability of a transportation hub node in the road network). Subsequently, generate spatio-temporal collaborative constraint weights through a fusion mechanism to dynamically modulate the message propagation intensity between nodes: for nodes with high spatio-temporal confidence (such as subway station nodes during stable weather periods), enhance the propagation range of their demand characteristics in the road network; for low-confidence nodes (such as temporarily traffic-controlled areas), suppress the diffusion of their abnormal fluctuation characteristics.

[0031] Specifically, in the embodiment of the present application, the step S321-1 includes: performing recurrent neural network sequence coding on the set of electric vehicle demand spatio-temporal collaborative association coding vectors to obtain a set of sequence-transmitted initial coding vectors of electric vehicle demand spatio-temporal collaboration, which can be expressed by the following formula: ; where is the set of electric vehicle demand spatio-temporal collaborative association coding vectors, and are respectively the 1st, 2nd, th, and th electric vehicle demand spatio-temporal collaborative association coding vectors in the set of electric vehicle demand spatio-temporal collaborative association coding vectors, is the sequence coding of based on the RNN structure, and are respectively the 1st, 2nd, th, and th sequence-transmitted initial coding vectors of electric vehicle demand spatio-temporal collaboration in the set of sequence-transmitted initial coding vectors of electric vehicle demand spatio-temporal collaboration.

[0032] It should be understood that it is difficult for traditional methods to capture the dynamic causal conduction mechanism of electric vehicle demand in complex scenarios through simple feature concatenation or weighted fusion. By performing recurrent neural network sequence coding processing, the information of historical time steps can be retained through recurrent connections, thereby modeling the dynamic dependencies in time series data. That is, the RNN performs preliminary temporal feature extraction on the set of electric vehicle demand spatio-temporal collaborative association coding vectors, and the generated set of sequence-transmitted initial coding vectors of electric vehicle demand spatio-temporal collaboration not only retains the inherent temporal law of demand fluctuations (such as the periodic strengthening of demand during the morning rush hour in the business district), but also provides a basic representation for integrating spatio-temporal structure information in subsequent steps, effectively solving the limitations of shallow feature interaction.

[0033] Specifically, Figure 6 is a flowchart of step S321-2 in the shared electric vehicle scheduling method based on big data analysis according to an embodiment of the present application. As Figure 6 shown, the step S321-2 includes: S321-21, calculating the electric vehicle demand spatio-temporal collaborative correlation feature time dimension confidence weight factor of each sequence transfer electric vehicle demand spatio-temporal collaborative correlation initial coding vector in the set of sequence transfer electric vehicle demand spatio-temporal collaborative correlation initial coding vectors; S321-22, calculating the electric vehicle demand spatio-temporal collaborative correlation feature space topology confidence scale factor of each sequence transfer electric vehicle demand spatio-temporal collaborative correlation initial coding vector in the set of sequence transfer electric vehicle demand spatio-temporal collaborative correlation initial coding vectors; S321-23, constructing the electric vehicle demand spatio-temporal collaborative correlation feature message passing spatio-temporal collaborative constraint factor of each sequence transfer electric vehicle demand spatio-temporal collaborative correlation initial coding vector based on the electric vehicle demand spatio-temporal collaborative correlation feature time dimension confidence weight factor and the electric vehicle demand spatio-temporal collaborative correlation feature space topology confidence scale factor of each sequence transfer electric vehicle demand spatio-temporal collaborative correlation initial coding vector.

[0034] More specifically, in the embodiment of the present application, the step S321-21 includes: calculating the electric vehicle demand spatio-temporal collaborative correlation feature time dimension confidence weight factor of each sequence transfer electric vehicle demand spatio-temporal collaborative correlation initial coding vector in the set of sequence transfer electric vehicle demand spatio-temporal collaborative correlation initial coding vectors, which can be expressed by the following formula: ; where is the corresponding learnable weight matrix, is the corresponding learnable weight matrix, and are respectively trainable weighted hyperparameters, is the hyperbolic tangent activation function, is matrix multiplication, is the time series scoring weight vector, is the corresponding electric vehicle demand spatio-temporal collaborative correlation feature energy score, is the normalization function, is the corresponding electric vehicle demand spatio-temporal collaborative correlation feature time dimension confidence weight factor.

[0035] It should be understood that the reliability of information at different time steps varies. For example, abnormal demands caused by continuous rainy weather may reduce the prediction credibility of historical data. To dynamically evaluate the information quality of each time step, it is necessary to calculate the confidence weight factor of the spatio-temporal collaborative association feature in the time dimension of the electric vehicle demand. That is, by adaptively weighting to suppress noise interference (such as abnormal fluctuations in temporary traffic control), enhancing the weight of stable periodic features (such as commuting demands), purifying the time-series information, avoiding the negative impact of low-quality historical data on subsequent propagation, and improving the robustness of the model to time-series dynamics.

[0036] More specifically, in the embodiment of the present application, the steps S321-22 include: calculating the spatio-temporal collaborative association feature space topology confidence scale factor of each sequence transfer electric vehicle demand spatio-temporal collaborative association initial coding vector in the set of sequence transfer electric vehicle demand spatio-temporal collaborative association initial coding vectors, which can be expressed by the following formula: ; It should be understood that is the transposed vector of is the square of the F-norm calculation, is the corresponding spatio-temporal collaborative association feature space similarity score value of the electric vehicle demand, is the exponential function value with the natural constant as the base, is the corresponding spatio-temporal collaborative association feature space topology confidence scale factor of the electric vehicle demand.

[0037] It should be understood that due to its strong radiation ability, transportation hub nodes have a significant impact on the demands of surrounding areas. To evaluate the structural importance of each node in the network topology, it is necessary to calculate the spatio-temporal collaborative association feature space topology confidence scale factor of the electric vehicle demand in the present application. Specifically, by explicitly modeling the spatial distribution relationship of nodes (such as road network connectivity), this factor can quantify the information propagation potential of nodes and assign higher weights to key nodes (such as the demand fluctuations of subway station nodes need to be globally perceived more). Its calculation process integrates the spatial role characteristics of nodes (such as centrality, connectivity), enabling the model to identify differences in structural importance, such as suppressing abnormal propagation in temporarily controlled areas and enhancing the feature diffusion range of hub nodes. That is, this step converts spatial structure information into a quantifiable confidence scale, which can provide a regulatory basis for subsequent message passing in the spatial dimension.

[0038] More specifically, in the embodiments of the present application, the step S321-23 includes: performing weighted fusion based on the sigmoid function on the electric vehicle demand spatio-temporal collaborative correlation feature time dimension confidence weight factor and the electric vehicle demand spatio-temporal collaborative correlation feature space topology confidence scale factor of each sequence-transmitted initial coding vector of electric vehicle demand spatio-temporal collaboration to obtain the electric vehicle demand spatio-temporal collaborative correlation feature message passing spatio-temporal collaborative constraint factor of each sequence-transmitted initial coding vector of electric vehicle demand spatio-temporal collaboration. The above process can be expressed by the formula: ; where and are respectively and 's contribution adjustment parameters, is function, is 's corresponding electric vehicle demand spatio-temporal collaborative correlation feature message passing spatio-temporal collaborative constraint factor.

[0039] It should be understood that the confidence evaluation of a single dimension cannot comprehensively reflect the spatio-temporal characteristics of nodes, and spatio-temporal constraints need to be integrated through a fusion strategy. Specifically, a non-linear gating mechanism is used to dynamically balance the temporal reliability (such as weather stability) and spatial importance (such as hub radiation ability), and simulate the synergistic effect of spatio-temporal factors. That is, the generated electric vehicle demand spatio-temporal collaborative correlation feature message passing spatio-temporal collaborative constraint factor not only suppresses the abnormal propagation of low spatio-temporal confidence nodes (such as temporarily controlled areas), but also enhances the cross-spatio-temporal information interaction of high-confidence nodes (such as hub nodes during stable periods), realizing a "logical AND" type of joint regulation, and can provide a refined weight basis for structured message passing.

[0040] Preferably, in another embodiment of the present application, based on the electric vehicle demand spatio-temporal collaborative correlation feature time dimension confidence weight factor and the electric vehicle demand spatio-temporal collaborative correlation feature space topology confidence scale factor of each sequence-transmitted initial coding vector of electric vehicle demand spatio-temporal collaboration, constructing the electric vehicle demand spatio-temporal collaborative correlation feature message passing spatio-temporal collaborative constraint factor of each sequence-transmitted initial coding vector of electric vehicle demand spatio-temporal collaboration includes: performing feature coupling based on spatio-temporal manifold curvature compensation on the electric vehicle demand spatio-temporal collaborative correlation feature time dimension confidence weight factor and the electric vehicle demand spatio-temporal collaborative correlation feature space topology confidence scale factor of each sequence-transmitted initial coding vector of electric vehicle demand spatio-temporal collaboration to obtain the electric vehicle demand spatio-temporal collaborative correlation feature message passing spatio-temporal collaborative constraint factor of each sequence-transmitted initial coding vector of electric vehicle demand spatio-temporal collaboration. The specific processing process of this process is as follows: In the electric vehicle demand spatio-temporal collaborative correlation feature time dimension confidence weight factor and the electric vehicle demand spatio-temporal collaborative correlation feature space topology confidence scale factor When expressing the information distribution intensity of the initial coding vectors of the spatio-temporal collaborative association of electric vehicle demands in each sequence in the time and space dimensions respectively, if the time and space dimensions are regarded as a global spatio-temporal single-degree-of-freedom modeling paradigm, it is necessary to avoid the negative curvature characterization of the joint spatio-temporal manifold caused by the negative correlation effect of the single-modal attention mechanism. This phenomenon will lead to the degradation of the fusion space expression ability under non-planar conditions.

[0041] Based on this, first, through the initial coding vectors of the spatio-temporal collaborative association of electric vehicle demands passed by each sequence The confidence weight factor of the spatio-temporal collaborative association feature of electric vehicle demands corresponding to the time dimension And the confidence scale factor of the spatio-topological collaborative association feature of electric vehicle demands , a constant curvature space mapping And a hypersphere approximation model : .

[0042] Subsequently, the spatio-temporal collaborative constraint factor of the message passing of the spatio-temporal collaborative association feature of electric vehicle demands is calculated and determined by the following formula : ; When is satisfied, the orthogonal attraction field of the flat spatio-temporal coupling in the spatio-temporal dimension can approach flatness in the sense of differential geometry, that is, by compensating for the generation of single-modal negative curvature, the quasi-Euclidean property on the fusion manifold is realized, thereby enhancing the geometric consistency of the spatio-temporal collaborative constraint of message passing.

[0043] Specifically, in the embodiment of the present application, the step S321-3 includes: based on the spatio-temporal collaborative constraint factor of the message passing of the spatio-temporal collaborative association feature of electric vehicle demands, modulating the message passing structure of the initial coding vectors of the spatio-temporal collaborative association of electric vehicle demands passed by each sequence to obtain a set of structurally modulated coding vectors of the spatio-temporal collaborative association feature of electric vehicle demands passed by the sequence. The above process can be expressed by the formula: ; where is the th structurally modulated coding vector of the spatio-temporal collaborative association feature of electric vehicle demands passed by the sequence in the set of structurally modulated coding vectors of the spatio-temporal collaborative association feature of electric vehicle demands passed by the sequence.

[0044] It should be understood that traditional weighted fusion is difficult to dynamically adjust the propagation path. It is necessary to directionally modulate the message flow through the spatio-temporal collaborative constraint factor of the spatio-temporal collaborative correlation feature message of the electric vehicle demand. Specifically, a per-node multiplication operation is used to control the information propagation intensity, simulating the selective information diffusion mechanism of biological systems. That is, the generated sequence passes the structural modulation coding vector of the spatio-temporal collaborative correlation feature of the electric vehicle demand, which not only retains the key spatio-temporal patterns (such as the spatial conduction chain during morning and evening rush hours), but also filters out local noise, and finally can form a structural feature that can represent the global spatio-temporal state.

[0045] Specifically, in the embodiment of the present application, the step S321-4 includes: calculating the position-wise sum of the set of the structural modulation coding vectors of the sequence passing the spatio-temporal collaborative correlation feature of the electric vehicle demand to obtain the spatio-temporal state inference coding vector of the electric vehicle demand. The above process can be expressed by the formula: ; where is the number of vectors in the set of the structural modulation coding vectors of the sequence passing the spatio-temporal collaborative correlation feature of the electric vehicle demand, is the spatio-temporal state inference coding vector of the electric vehicle demand.

[0046] It should be understood that the set of the structural modulation coding vectors of the sequence passing the spatio-temporal collaborative correlation feature of the electric vehicle demand needs to be compressed into a static vector to support downstream inference tasks. Based on this, a position-wise sum operation is performed on the original set to aggregate the time series information, and then a static spatio-temporal state inference coding of the electric vehicle demand is generated. Specifically, by summing the modulated features along the time dimension, the model can compress the dynamic time series changes (such as the intensity fluctuations of daily peaks within a week) into an overall trend expression (such as the weekly average demand distribution). This operation is essentially a time pooling strategy, which retains the statistical characteristics of the time series (such as the duration of demand peaks), while eliminating the interference of short-term fluctuations. The generated spatio-temporal state inference coding vector of the electric vehicle demand is highly abstract, integrating the global state information under spatio-temporal collaboration constraints (such as the steady-state mode of demand conduction between regions), and can provide a robust spatio-temporal feature representation for downstream tasks (such as demand prediction).

[0047] In step S322, the electric vehicle demand spatiotemporal state reasoning coding vector is feature decoded to obtain the electric vehicle demand forecast value of each grid area. Specifically, in the embodiment of the present application, the step S322 includes: inputting the electric vehicle demand spatiotemporal state reasoning coding vector into the demand forecaster based on the decoder to obtain the electric vehicle demand forecast value of each grid area. It should be understood that the electric vehicle demand spatiotemporal state reasoning coding vector is a highly abstract representation of the electric vehicle demand spatiotemporal state obtained after the previous multiple steps, which contains a large amount of complex spatiotemporal feature information and reasoning results, but this information exists in the form of a coding vector and cannot be directly used for actual scheduling decisions. The demand forecaster based on the decoder can decode these coding vectors and convert the information contained therein into specific and understandable electric vehicle demand forecast values ​​for each grid area. For example, the information about demand trends, inter-regional associations, etc. in the coding vector is converted into a specific quantity forecast, so that the operator can intuitively understand the future electric vehicle demand situation in each grid area. In a specific embodiment of the present application, the demand forecaster based on the decoder can use a recurrent neural network structure to realize the feature decoding of the electric vehicle demand spatiotemporal state reasoning coding vector. Specifically, the encoding vector containing spatiotemporal collaborative correlation information is first used as the initial input of the decoder to capture the dynamic dependency of electric vehicle demand data in the time dimension. As the time step progresses, the decoder iteratively processes the historical decoding results and the current input encoding vector through the cyclic unit, and gradually integrates the temporal characteristics in the time series, such as the demand fluctuation law and periodic changes in different periods. At each time step, the hidden state of the decoder is updated according to the current input and the hidden state of the previous time step, so as to continuously accumulate and update the temporal information about demand changes. Finally, the hidden state output by the decoder is mapped to a specific numerical space through a fully connected layer, and the corresponding electric vehicle demand forecast value is generated for each grid area. This decoding process based on the recurrent neural network structure can effectively utilize the implicit spatiotemporal correlation characteristics in the spatiotemporal state reasoning encoding vector, while considering the continuity of the time series, combined with the impact of the spatial topological structure on demand changes, to achieve accurate prediction of future demand in each grid area.

[0048] In step S4, a demand heat map is generated based on the electric vehicle demand prediction values of the respective grid regions. Correspondingly, considering that demand prediction results are usually presented in the form of discrete numerical values, this data expression method is difficult to intuitively reflect the change of demand density gradient in the urban spatial dimension, and even less able to reveal potential regional demand conduction paths. For example, when multiple adjacent grids simultaneously show moderate demand prediction values, the discrete numerical table cannot effectively identify whether these regions form a continuous demand corridor (such as a cycling demand belt extending along the subway line), or are just isolated point-like demand distributions. The limitations of this data expression form lead to dispatch decision-makers often relying on experience to judge the boundaries of demand aggregation areas, which easily causes misallocation of dispatch resources in the transition zone - it may occur that vehicles are overly concentrated in a single grid with the highest prediction value, while ignoring the collaborative replenishment opportunities in surrounding areas with slightly lower demand. Therefore, in the technical solution of this application, a demand heat map is generated based on the electric vehicle demand prediction values of the respective grid regions. The heat map represents the magnitude of the demand prediction value through the shade of color, and can convert a large amount of digital information into an intuitive visual graph, enabling relevant personnel to immediately see which regions have high demand, which regions have low demand, and the spatial distribution characteristics of the demand. For example, on a city map, the red area represents the high-demand area, and the blue area represents the low-demand area. Through this visualization method, the areas with concentrated demand can be quickly located.

[0049] The following is a detailed elaboration of a specific implementation process of "generating a demand heat map based on the electric vehicle demand prediction values of the respective grid regions": In the color gradient system construction link, a mapping rule between demand intensity and color attributes needs to be established. A continuous color gradient scheme is adopted, usually choosing a transition from cold colors to warm colors (such as from blue to red) to intuitively reflect the spatial difference in demand intensity. The color gradient range needs to cover the entire value range of the demand prediction value. For example, the low-demand interval of 0 - 20 vehicles is mapped to light blue, the medium-demand interval of 21 - 80 vehicles is mapped to yellow, and the high-demand interval of 81 - 150 vehicles is mapped to red. At the same time, according to the maximum and minimum values of the demand prediction value, the color gradient is divided into multiple continuous levels (such as 10 levels) to ensure a smooth color transition between adjacent demand intervals and avoid a sense of visual discontinuity. Each color level corresponds to a specific demand numerical interval, and the association between the demand value and the color depth is achieved through a linear or non-linear mapping method.

[0050] In the process of matching grid demand values with colors, first, the predicted electric vehicle demand values in each grid area are standardized and mapped to the value range of the color gradient. For example, if the range of the predicted demand value is 0 - 150 vehicles, corresponding to 0 (blue) to 255 (red) of the RGB color values, then the corresponding relationship between the demand value and the RGB components is established according to the linear ratio, so that each demand value corresponds to a unique color value. For abnormal demand values beyond the normal range (such as demand mutations caused by extreme events), the extreme values of the color gradient (the darkest red or the lightest blue) are uniformly used for representation to maintain the stability and readability of the color system.

[0051] The visualization implementation of the geographic information system is a key link. First, load the base map of the electronic map, accurately locate the geographic coordinates of all grid areas, and ensure that the grid boundaries match the actual geographic elements (such as roads, blocks). Fill the color attributes of each grid (determined based on the predicted demand value) into the corresponding geographic area through GIS tools. During the rendering process, the overlaps or gaps between grids need to be processed to ensure the continuity of color display and visual smoothness. To enhance the visualization effect, the transparency can be reduced for high-demand areas and increased for low-demand areas to form a visual hierarchy. At the same time, specific numerical values and area names are marked for the grids with the top-ranked demand values to facilitate quick positioning of hot spots. The legend and interactive functions are configured to provide auxiliary interpretation tools for the heat map. A color-demand comparison table is attached beside the heat map to clarify the range of demand values corresponding to each color interval, helping users establish an intuitive association between colors and demand intensity.

[0052] In the final output link, the generated demand heat map needs to be presented in a suitable format, supporting forms such as pictures (such as PNG, JPG) or dynamic layers (such as interactive layers in a GIS system), and adapting to the display requirements of computers, mobile devices, or large-screen monitoring systems. At the same time, a real-time update mechanism is established. According to the update frequency of demand prediction (such as every 15 minutes), the color mapping is automatically recalculated and the heat map is refreshed to ensure that the visualization result is synchronized with the latest demand prediction data, meeting the real-time requirements of dynamic scheduling scenarios.

[0053] In step S5, the number of current electric vehicles in each grid area is obtained, and a current vehicle distribution map is generated based on the number of current electric vehicles in each grid area. It should be understood that shared electric vehicles are in a dynamic state of use and parking in the city, and their quantity and location may change at any time. Obtaining the number of current electric vehicles in each grid area can enable real-time understanding of the vehicle distribution in the city, providing accurate basic data for subsequent dispatching decisions. For example, through real-time monitoring, operators can know which areas have more vehicles and which areas have fewer vehicles, avoiding dispatching errors caused by information lag. Subsequently, considering the number of current electric vehicles in each grid area presented in digital form, although it can accurately reflect the quantity information, it is difficult to intuitively show the spatial distribution of vehicles. The current vehicle distribution map, through a graphical method, converts the vehicle quantity information into a visual map display, using intuitive elements such as different colors, icons, or densities to represent the vehicle quantity differences in each grid area, enabling relevant personnel to quickly and intuitively understand the vehicle distribution pattern in the city, such as which areas are vehicle-dense and which areas are vehicle-sparse.

[0054] Specifically, in a specific embodiment of the present application, first, a real-time positioning data acquisition system needs to be constructed. The position information of each vehicle is obtained in real time through in-vehicle positioning devices deployed on shared electric vehicles. This information includes accurate longitude and latitude coordinates and timestamps, and is uploaded to the operation management server at a minute-level frequency (such as once per minute). The server-side preprocesses the original data, filters abnormal coordinates (such as drift data beyond the urban geographical range) through a data cleaning algorithm, eliminates duplicate records (consecutive positioning data of the same vehicle in a short period), and marks offline vehicles with long-unupdated positions, ensuring the accuracy and effectiveness of the data entering the subsequent processing link.

[0055] In the grid area spatial matching stage, the preprocessed vehicle position data needs to be spatially associated with the predefined grid areas. Each grid area is composed of clear geographical boundaries (such as a rectangular or polygonal area defined by longitude and latitude ranges). Each electric vehicle's belonging grid is located one by one through a spatial geometry algorithm (such as the ray method to determine whether a point is inside a polygon). This process needs to ensure that the grid division criteria are completely consistent with those in the demand prediction link (such as the same grid size, shape, and geographical coverage) to achieve the unification of supply and demand data in the spatial dimension. For vehicles parked near the grid boundary, they are assigned to the nearest grid area according to the principle of the minimum Euclidean distance, avoiding statistical deviations caused by ambiguous positions.

[0056] After completing the spatial matching, enter the dynamic vehicle count phase. The server groups and aggregates the real-time location data by grid area and calculates the number of electric vehicles parked in each grid through a time window mechanism (e.g., valid location data within the last 5 minutes before the current time). For vehicles in a riding state (e.g., moving vehicles with a speed greater than 5 km / h), they are automatically excluded by the motion state detection module, and only vehicles in a stationary parked state (speed is 0 or below a specific threshold and persists for a certain period) are included in the current grid statistics. Meanwhile, an abnormal data monitoring mechanism is established. Vehicles whose positions have not been updated for 30 consecutive minutes are marked as "data abnormal" and are temporarily not included in the current statistics. They will be re-included in the calculation after the position information returns to normal to ensure that the statistical results reflect the real-time parked state.

[0057] The realization of GIS visualization is a key step in converting statistical data into intuitive graphics. First, load the electronic map base map with the same coordinate system as the required heat map to ensure the precise alignment of spatial positions. Then, based on the geographical boundaries of the grid area and the statistically obtained number of vehicles, generate a visualization layer through the GIS rendering engine. The visualization methods can adopt graduated color mapping or density distribution rendering: For graduated color mapping, a correspondence between the number of vehicles and the color gradient needs to be established. For example, map the low-demand interval of 0 - 20 vehicles to light blue, the medium-demand interval of 21 - 80 vehicles to yellow, and the high-demand interval of 81 - 150 vehicles to red. Each color interval corresponds to a specific number range, and the color depth increases with the increase in the number; Density distribution rendering is achieved by overlaying a semi-transparent point layer on the map. The size or transparency of the points is dynamically adjusted according to the number of vehicles in the area, forming a continuous density distribution visual effect to intuitively display the degree of vehicle aggregation.

[0058] To enhance the visualization effect and information communication efficiency, the top 15% of high-inventory grids in terms of the number of vehicles can be prominently marked, showing the specific number of vehicles and the area name on the map (such as "Industrial Park B, current number of vehicles: 92"). Meanwhile, add a legend beside the visualization interface to clarify the vehicle number intervals corresponding to different colors or density symbols (such as "Red: 81 - 150 vehicles; Yellow: 21 - 80 vehicles; Blue: 0 - 20 vehicles") to help users quickly establish the correspondence between visual symbols and data values.

[0059] In the data output and application stage, the current vehicle distribution map needs to support multi-terminal adaptation and be presented in the form of static pictures, dynamic interactive layers, or API interfaces (for the dispatching system to directly call spatial data). For dynamic dispatching requirements, a real-time data synchronization mechanism needs to be established to ensure that the update frequency of the vehicle distribution map is the same as that of the demand heat map (e.g., refreshed every 15 minutes). Through spatial coordinate matching technology, accurate overlay display of the two can be achieved, facilitating operators to intuitively compare the differences in supply and demand distribution. At the same time, encrypt the vehicle location data, follow data security specifications, prevent the leakage of user privacy or illegal data calls, and ensure the security and compliance of system operation.

[0060] In step S6, based on the demand heat map and the current vehicle distribution map, a dynamic dispatching strategy is generated. Accordingly, considering that the demand heat map reflects the expected demand degree of electric vehicles in each grid area, while the current vehicle distribution map shows the actual distribution of electric vehicles. Making dispatching decisions solely based on the demand heat map or the current vehicle distribution map may be one-sided. Combining the two can comprehensively and accurately grasp the supply and demand relationship in the market, not only understanding which areas have high demand but also knowing whether the vehicle supply in these areas is sufficient, thus providing a complete information basis for formulating reasonable dispatching strategies. In this way, through reasonable vehicle dispatching, the supply of electric vehicles in each grid area is matched with the demand. For areas where demand is higher than vehicle supply, dispatch vehicles from areas with vehicle surpluses for supplementation; for areas where vehicle supply is more than demand, dispatch the excess vehicles to other areas with demand. In this way, the vehicle usage needs of users are maximally met, the waiting time of users is reduced, the utilization rate of vehicles is increased, and resource waste is avoided.

[0061] The following is a detailed elaboration of a specific implementation process of "generating a dynamic dispatching strategy based on the demand heat map and the current vehicle distribution map": First, perform data alignment and grid matching. Align the grid areas of the two maps in terms of spatial coordinates according to a unified standard (same shape, size, and geographical boundary) to ensure that each grid corresponds to a unique geographical unit in the two maps. Associate the demand prediction value with the current vehicle quantity through a data interface to form a two-dimensional data matrix containing both.

[0062] In the supply and demand difference quantitative analysis stage, calculate the supply and demand difference for each grid, and judge whether the grid is a demand shortage area, a vehicle surplus area, or a supply and demand balance area based on the positive or negative of the difference. The latter two are sorted according to the demand urgency degree (shortage quantity and demand prediction value) and vehicle redundancy degree (redundancy quantity and spatial location) respectively to generate a shortage area list and a surplus area list.

[0063] During the scheduling priority determination process, the shortage areas are classified according to the demand forecast value and the shortage quantity. The areas with high demand and severe shortage are designated as the first level and need to complete the scheduling within a short time. By analogy, the second and third level shortage areas and the corresponding scheduling completion times are set; for the surplus areas, the priority is determined by combining the redundant vehicle quantity and the spatial distance from the shortage areas. The areas with a large redundant quantity and adjacent to the shortage areas are designated as the first level and used as the starting point for priority scheduling.

[0064] When calculating and matching the scheduling quantity, first determine that the required scheduling quantity for each shortage area is the difference between supply and demand, and does not exceed the redundant quantity of the corresponding surplus area. Following the principle of "nearest first", use the spatial topological distribution matrix of the grid area to calculate the shortest road distance between the shortage area and the surplus area, and match them in ascending order of distance. If a single surplus area cannot meet the demand of a shortage area, the demand will be split into multiple adjacent surplus areas.

[0065] In the scheduling path planning stage, for each pair of matched surplus area and shortage area, generate the shortest path considering the actual traffic rules (road capacity, restricted areas, etc.) with the help of the geographic information system, and record the distance and the estimated travel time; when there are multiple scheduling tasks, use the heuristic algorithm to merge the scheduling routes, reduce duplicate transportation, and lower the total transportation cost.

[0066] The dynamic scheduling strategy generation includes multiple aspects. The scheduling instruction list clarifies the departure grid, arrival grid, number of scheduling vehicles, priority level, and the latest completion time of each instruction. The priority level corresponds to the scheduling completion time to ensure that urgent demands are processed first. The transportation resource allocation deploys transportation vehicles according to the quantity and distribution of the scheduling tasks. The vehicle load limit matches the scheduling quantity, and information such as vehicle number, departure time, and estimated arrival time is recorded. The emergency adjustment mechanism reserves a certain proportion of redundant vehicles as emergency reserves to cope with sudden demand changes or transportation delays. At the same time, a real-time monitoring threshold is set, and when the actual scheduling deviation exceeds the threshold, emergency scheduling is triggered.

[0067] In the strategy execution and real-time monitoring link, push the scheduling strategy to the execution end through the operation management system, clarifying the execution order and operation requirements; update the current vehicle distribution map in real time during the scheduling process, and automatically synchronously update the vehicle quantity of relevant grids after each scheduling task is completed; after the scheduling is completed, compare the demand forecast value with the actual vehicle quantity to evaluate the supply-demand matching situation and provide a basis for subsequent strategy optimization.

[0068] During the implementation process, it is necessary to note that the generation frequency of the scheduling strategy is consistent with the update frequency of the two maps to adapt to the dynamic changes in demand. And before generating the strategy, verify the number of schedulable vehicles in the surplus area, deduct the unavailable vehicles, ensure that the scheduling quantity is within the range of actual available resources, and guarantee the feasibility and effectiveness of the strategy.

[0069] In summary, the shared electric vehicle scheduling method based on big data analysis according to the embodiments of the present application is elucidated. First, the historical demand time series fluctuation characteristics of each grid area are extracted. Then, the road connectivity and spatial proximity between regions are quantified by constructing a grid area spatial topology distribution matrix. Next, the demand conduction effect formed by population flow and road topology in adjacent regions is captured through a graph structure. Subsequently, a spatio-temporal reasoning model is introduced to simulate the diffusion path of the chain effect of sudden factors such as extreme weather and temporary activities on the multi-region demand, so as to obtain the electric vehicle demand prediction values of each grid area. Finally, a scheduling strategy is generated by dynamically comparing the differences between the demand heat map and the current vehicle distribution map. In this way, it helps to achieve the spatio-temporal optimal allocation of vehicle resources across the entire road network.

[0070] Figure 7 FIG. is a system block diagram of a shared electric vehicle scheduling system based on big data analysis according to an embodiment of the present application. As Figure 7 shown, the shared electric vehicle scheduling system 100 based on big data analysis according to an embodiment of the present application includes: a historical data acquisition module 110 for acquiring historical data of the electric vehicle demand in each grid area; a spatial topology matrix calculation module 120 for calculating the grid area spatial topology distribution matrix of each grid area; a demand prediction module 130 for performing spatio-temporal collaborative correlation analysis on the historical data of the electric vehicle demand in each grid area and the grid area spatial topology distribution matrix to obtain the electric vehicle demand prediction values of each grid area; a demand heat map generation module 140 for generating a demand heat map based on the electric vehicle demand prediction values of each grid area; a current vehicle distribution map generation module 150 for acquiring the current number of electric vehicles in each grid area and generating a current vehicle distribution map based on the current number of electric vehicles in each grid area; and a scheduling strategy generation module 160 for generating a dynamic scheduling strategy based on the demand heat map and the current vehicle distribution map.

[0071] Here, those skilled in the art can understand that the specific functions and operations of each unit and module in the above-mentioned shared electric vehicle scheduling system 100 based on big data analysis have been introduced in detail in the description of the shared electric vehicle scheduling method based on big data analysis above, and therefore, the repeated description thereof will be omitted. Figures 1 to 6 of the shared electric vehicle scheduling method based on big data analysis above, and therefore, the repeated description thereof will be omitted.

[0072] In summary, the shared electric vehicle scheduling system 100 based on big data analysis according to the embodiments of the present application is elucidated. It first extracts the historical demand time series fluctuation characteristics of each grid area, then quantifies the road connectivity and spatial proximity between regions by constructing a grid area spatial topology distribution matrix, and then captures the demand conduction effect formed by population flow and road topology in adjacent regions through a graph structure. Subsequently, a spatio-temporal inference model is introduced to simulate the diffusion path of the chain effect of sudden factors such as extreme weather and temporary activities on the multi-region demand, so as to obtain the electric vehicle demand prediction value of each grid area. Finally, a scheduling strategy is generated by dynamically comparing the differences between the demand heat map and the current vehicle distribution map. In this way, it helps to achieve the spatio-temporal optimal allocation of vehicle resources across the entire road network.

Claims

1. A shared electric vehicle scheduling method based on big data analysis, characterized in that: include: Obtaining historical data on electric vehicle demand in each grid area; calculating a grid area spatial topological distribution matrix of each grid area; The historical data of electric vehicle demand in each grid area and the spatial topological distribution matrix of the grid area are subjected to spatiotemporal collaborative correlation analysis to obtain the electric vehicle demand forecast value of each grid area, including: spatiotemporal encoding of the electric vehicle demand in each grid area and the spatial topological distribution matrix of the grid area to obtain a set of spatiotemporal collaborative correlation features of the electric vehicle demand; spatiotemporal reasoning encoding and decoding of the electric vehicle demand based on structured recursion to obtain the electric vehicle demand forecast value of each grid area; generating a demand heat map based on the electric vehicle demand forecast value of each grid area; obtaining the current number of electric vehicles in each grid area, and generating a current vehicle distribution map based on the current number of electric vehicles in each grid area; generating a dynamic scheduling strategy based on the demand heat map and the current vehicle distribution map.

2. The method for scheduling shared electric vehicles based on big data analysis according to claim 1 is characterized in that: The value of each non-diagonal position in the grid area spatial topological distribution matrix is ​​the spatial distance between two grid areas.

3. The method for scheduling shared electric vehicles based on big data analysis according to claim 1 is characterized in that: The historical data of electric vehicle demand in each grid area and the spatial topological distribution matrix of the grid area are subjected to spatiotemporal encoding of electric vehicle demand to obtain a set of spatiotemporal collaborative correlation characteristics of electric vehicle demand, including: performing time series encoding of historical demand characteristics on the historical data of electric vehicle demand in each grid area to obtain a set of electric vehicle grid area demand time series feature coding vectors; inputting the set of electric vehicle grid area demand time series feature coding vectors and the grid area spatial topological distribution matrix into an electric vehicle demand spatiotemporal context encoder based on a graph convolutional neural network model to obtain a set of electric vehicle demand spatiotemporal collaborative correlation coding vectors as the set of electric vehicle demand spatiotemporal collaborative correlation characteristics.

4. The method for scheduling shared electric vehicles based on big data analysis according to claim 3 is characterized in that: The historical data of the electric vehicle demand in each grid area is subjected to historical demand characteristic time series coding to obtain a set of electric vehicle grid area demand timing characteristic coding vectors, including: the historical data of the electric vehicle demand in each grid area is subjected to historical demand characteristic time series coding based on a bidirectional gated cyclic unit to obtain a set of electric vehicle grid area demand timing characteristic coding vectors.

5. The method for scheduling shared electric vehicles based on big data analysis according to claim 1 is characterized in that: The set of the electric vehicle demand spatiotemporal collaborative correlation features is subjected to structured recursive electric vehicle demand spatiotemporal reasoning encoding and decoding to obtain the electric vehicle demand forecast value for each grid area, including: performing structured recursive electric vehicle demand spatiotemporal reasoning on the set of the electric vehicle demand spatiotemporal collaborative correlation coding vectors to obtain the electric vehicle demand spatiotemporal state reasoning coding vector; and feature decoding the electric vehicle demand spatiotemporal state reasoning coding vector to obtain the electric vehicle demand forecast value for each grid area.

6. The method for scheduling shared electric vehicles based on big data analysis according to claim 5 is characterized in that: The electric vehicle demand spatiotemporal reasoning based on structured recursion is performed on the set of the electric vehicle demand spatiotemporal collaborative association coding vectors to obtain the electric vehicle demand spatiotemporal state reasoning coding vector, including: performing recurrent neural network sequence encoding on the set of the electric vehicle demand spatiotemporal collaborative association coding vectors to obtain a set of sequence-transmitted electric vehicle demand spatiotemporal collaborative association initial coding vectors; calculating the electric vehicle demand spatiotemporal collaborative association characteristic message transmission spatiotemporal collaborative constraint factor of each sequence-transmitted electric vehicle demand spatiotemporal collaborative association initial coding vector in the set of sequence-transmitted electric vehicle demand spatiotemporal collaborative association initial coding vectors; based on the electric vehicle demand spatiotemporal collaborative association characteristic message transmission spatiotemporal collaborative constraint factor, performing message transmission structure modulation on each sequence-transmitted electric vehicle demand spatiotemporal collaborative association initial coding vector to obtain a set of sequence-transmitted electric vehicle demand spatiotemporal collaborative association characteristic structural modulation coding vectors; calculating the positional sum of the set of sequence-transmitted electric vehicle demand spatiotemporal collaborative association characteristic structural modulation coding vectors to obtain the electric vehicle demand spatiotemporal state reasoning coding vector.

7. The method for scheduling shared electric vehicles based on big data analysis according to claim 6 is characterized in that: The method comprises: calculating the electric vehicle demand spatiotemporal collaborative association characteristic message transmission spatiotemporal collaborative constraint factor of each sequence-transmitted electric vehicle demand spatiotemporal collaborative association initial coding vector in the set of the sequence-transmitted electric vehicle demand spatiotemporal collaborative association initial coding vector, including: calculating the electric vehicle demand spatiotemporal collaborative association characteristic time dimension confidence weight factor of each sequence-transmitted electric vehicle demand spatiotemporal collaborative association initial coding vector in the set of the sequence-transmitted electric vehicle demand spatiotemporal collaborative association initial coding vector; calculating the electric vehicle demand spatiotemporal collaborative association characteristic space topology confidence scaling factor of each sequence-transmitted electric vehicle demand spatiotemporal collaborative association initial coding vector in the set of the sequence-transmitted electric vehicle demand spatiotemporal collaborative association initial coding vector; and calculating the electric vehicle demand spatiotemporal collaborative association characteristic space topology confidence scaling factor of each sequence-transmitted electric vehicle demand spatiotemporal collaborative association initial coding vector based on the sequence-transmitted electric vehicle demand spatiotemporal collaborative association initial coding vector. The electric vehicle demand spatiotemporal collaborative association feature time dimension confidence weight factor and the electric vehicle demand spatiotemporal collaborative association feature space topology confidence scaling factor of the initial coding vector of the electric vehicle demand are used to construct the spatiotemporal collaborative constraint factor of the electric vehicle demand spatiotemporal collaborative association feature message transmitted by each sequence of the electric vehicle demand spatiotemporal collaborative association initial coding vector, including: weighted fusion of the electric vehicle demand spatiotemporal collaborative association feature time dimension confidence weight factor and the electric vehicle demand spatiotemporal collaborative association feature space topology confidence scaling factor of the electric vehicle demand spatiotemporal collaborative association feature message transmitted by each sequence of the electric vehicle demand spatiotemporal collaborative association initial coding vector based on the sigmoid function to obtain the spatiotemporal collaborative constraint factor of the electric vehicle demand spatiotemporal collaborative association feature message transmitted by each sequence of the electric vehicle demand spatiotemporal collaborative association initial coding vector.

8. The method for scheduling shared electric vehicles based on big data analysis according to claim 7 is characterized in that: Based on the electric vehicle demand spatiotemporal collaborative association feature time dimension confidence weight factor and the electric vehicle demand spatiotemporal collaborative association feature space topology confidence scaling factor of each sequence transmitting the electric vehicle demand spatiotemporal collaborative association initial coding vector, the electric vehicle demand spatiotemporal collaborative association feature message transmission spatiotemporal collaborative constraint factor of each sequence transmitting the electric vehicle demand spatiotemporal collaborative association initial coding vector is constructed, including: feature coupling based on space-time manifold curvature compensation is performed on the electric vehicle demand spatiotemporal collaborative association feature time dimension confidence weight factor and the electric vehicle demand spatiotemporal collaborative association feature space topology confidence scaling factor of each sequence transmitting the electric vehicle demand spatiotemporal collaborative association initial coding vector to obtain the electric vehicle demand spatiotemporal collaborative association feature message transmission spatiotemporal collaborative constraint factor of each sequence transmitting the electric vehicle demand spatiotemporal collaborative association initial coding vector.

9. The method for scheduling shared electric vehicles based on big data analysis according to claim 8 is characterized in that: The electric vehicle demand spatiotemporal state reasoning coding vector is feature decoded to obtain the electric vehicle demand forecast value for each grid area, including: inputting the electric vehicle demand spatiotemporal state reasoning coding vector into a decoder-based demand predictor to obtain the electric vehicle demand forecast value for each grid area.

10. A shared electric vehicle dispatching system based on big data analysis, characterized in that: include: A historical data acquisition module is used to obtain historical data on the demand for electric vehicles in each grid area; A spatial topology matrix calculation module is used to calculate the grid area spatial topology distribution matrix of each grid area; a demand prediction module is used to perform spatiotemporal collaborative association analysis on the historical data of electric vehicle demand in each grid area and the grid area spatial topology distribution matrix to obtain the electric vehicle demand forecast value of each grid area, wherein the demand prediction module is used to: perform spatiotemporal encoding of electric vehicle demand on the historical data of electric vehicle demand in each grid area and the grid area spatial topology distribution matrix to obtain a set of spatiotemporal collaborative association features of electric vehicle demand; perform spatiotemporal reasoning encoding and decoding of electric vehicle demand based on structured recursion on the set of spatiotemporal collaborative association features of electric vehicle demand to obtain the electric vehicle demand forecast value of each grid area; a demand heat map generation module is used to generate a demand heat map based on the electric vehicle demand forecast value of each grid area; a current vehicle distribution map generation module is used to obtain the current number of electric vehicles in each grid area, and generate a current vehicle distribution map based on the current number of electric vehicles in each grid area; a scheduling strategy generation module is used to generate a dynamic scheduling strategy based on the demand heat map and the current vehicle distribution map.

Images