Method and system for analyzing use data of rented vehicle

Abstract

The invention provides a use data analysis method and system for rental vehicles, and the method comprises the steps: carrying out the space-time marking of historical vehicle use data distributed at different rental points, constructing a space-time data matrix and a space weight matrix, and generating a historical vehicle use data set; identifying a spatial clustering mode and spatial heterogeneity features based on a historical vehicle use data set, and constructing a spatial dependency network of the use behaviors of the rented vehicles; according to the spatial dependency network, parametric modeling is carried out on spatial autocorrelation in the historical vehicle use data, and a spatial autocorrelation correction use model is established; and based on the spatial autocorrelation correction usage model, predicting vehicle demands of the rental sites, and generating a vehicle allocation strategy considering mutual influence between the rental sites. According to the invention, by introducing the space metering method, the space autocorrelation in the use data of the rented vehicle is effectively identified and corrected, and the vehicle demand prediction precision and the deployment decision reasonability are improved.

Description

Technical Field

[0001] The present invention relates to the field of data analysis technology, and in particular to a method and system for analyzing usage data of rental vehicles. Background Art

[0002] With the rapid development of shared mobility, rental vehicles are playing an increasingly important role in urban transportation. Traditional methods for analyzing rental vehicle usage data primarily rely on the independence assumption, treating each rental location as an isolated entity and employing classic statistical methods such as linear regression or time series forecasting for analysis and prediction. These methods typically rely on historical data from a single rental location and ignore the potential spatial interactions between rental locations.

[0003] However, in practice, vehicle usage behavior between adjacent rental locations often exhibits significant spatial correlation. Due to the geographic distribution of rental locations, the usage of one location can impact surrounding locations, forming a complex spatial dependency structure. Traditional methods fail to account for this spatial autocorrelation, leading to biased parameter estimation and reduced prediction accuracy. This, in turn, impacts the accuracy of vehicle dispatch decisions and limits the resource utilization efficiency of the overall service network. Summary of the Invention

[0004] The main purpose of this invention is to solve the technical problems of parameter estimation bias and low prediction accuracy caused by the existing rental vehicle usage data analysis method ignoring the spatial autocorrelation between rental points; A first aspect of the present invention provides a method for analyzing usage data of a rental vehicle, the method comprising: Performing spatiotemporal labeling on historical vehicle usage data distributed across different rental locations to construct a spatiotemporal data matrix for vehicle usage, and calculating a spatial weight matrix between rental locations. Based on the spatiotemporal data matrix and the spatial weight matrix, a historical vehicle usage dataset is generated. Identifying spatial clustering patterns and spatial heterogeneity characteristics in the historical vehicle usage data based on the historical vehicle usage dataset, and constructing a spatial dependency network of rental vehicle usage behavior using the spatial clustering patterns and the spatial heterogeneity characteristics; Based on the spatial dependency network, parameterizing the spatial autocorrelation in the historical vehicle usage data to establish a spatial autocorrelation correction usage model; Based on the spatial autocorrelation correction usage model, vehicle demand for rental points is predicted, and a vehicle deployment strategy that takes into account the mutual influence between rental points is generated based on the vehicle demand.

[0005] Optionally, in a first implementation of the first aspect of the present invention, performing spatiotemporal labeling on historical vehicle usage data distributed across different rental points, constructing a spatiotemporal data matrix of vehicle usage, and calculating a spatial weight matrix between rental points, generating a historical vehicle usage dataset based on the spatiotemporal data matrix and the spatial weight matrix includes: The GPS trajectory data of rental vehicles distributed at different rental points are sampled in segments according to time granularity, and each sampling point is associated with the nearest rental point to form a vehicle-rental point spatiotemporal association record; Extracting features from the spatiotemporal association records of the vehicle and rental point, calculating the vehicle usage frequency, average usage duration, peak usage period, and idle rate of each rental point in different time windows, and constructing a spatiotemporal data matrix containing multiple rental points and multiple time windows; Based on the geographic coordinate information of the rental points, the Euclidean distance between each rental point is calculated, and the Euclidean distance is converted into a spatial weight reflecting spatial proximity using the inverse distance weighting method to construct a spatial weight matrix between the rental points; Performing algebraic operations on the spatiotemporal data matrix and the spatial weight matrix, calculating a weighted vehicle usage index of each rental point affected by adjacent rental points, and combining the original usage index with the weighted vehicle usage index to generate a historical vehicle usage dataset.

[0006] Optionally, in a second implementation of the first aspect of the present invention, identifying spatial clustering patterns and spatial heterogeneity characteristics in the historical vehicle usage data based on the historical vehicle usage dataset, and constructing a spatial dependency network of rental vehicle usage behavior using the spatial clustering patterns and the spatial heterogeneity characteristics includes: Calculating a global Moran's index value for the historical vehicle usage data set, and determining whether there is statistically significant spatial autocorrelation in the overall rental network based on the global Moran's index value using a Monte Carlo simulation method; If yes, then calculate the local spatial autocorrelation index of each rental point based on the historical vehicle usage dataset, perform a significance test on the local spatial autocorrelation index, and identify statistically significant high-value cluster areas, low-value cluster areas, and spatial outliers; Based on the distribution characteristics of the high-value clustering areas, low-value clustering areas, and spatial outliers, the spatial influence coefficients between rental points are calculated, and a spatial flow matrix representing the mutual influence of vehicle usage behaviors between rental points is constructed; The spatial flow matrix is converted into a directed weighted network structure to form a spatial dependency relationship network of rental vehicle usage behavior.

[0007] Optionally, in a third implementation of the first aspect of the present invention, calculating a global Moran's index value for the historical vehicle usage dataset, and determining whether there is statistically significant spatial autocorrelation in the overall rental network based on the global Moran's index value using a Monte Carlo simulation method includes: Standardizing the vehicle usage index in the historical vehicle usage dataset to obtain a standardized usage index; Calculating a global Moran's index value based on the standardized usage index and the spatial weight matrix using a spatial statistical algorithm; Randomly permuting the standardized usage index using a Monte Carlo simulation method, recalculating the global Moran's index value after each permutation, and repeating the random permutation a preset number of times to form a zero distribution of the global Moran's index; Based on the null distribution of the global Moran's index, the p-value of the actual observation value is calculated. When the p-value is less than the preset significance level, it is determined that there is statistically significant spatial autocorrelation in the overall rental network. When the p-value is greater than or equal to the preset significance level, it is determined that there is no spatial autocorrelation and subsequent analysis is terminated.

[0008] Optionally, in a fourth implementation of the first aspect of the present invention, parameterizing the spatial autocorrelation in the historical vehicle usage data based on the spatial dependency network to establish a spatial autocorrelation correction usage model includes: Based on the structural characteristics of the spatial dependency network, a corresponding spatial econometric model type is selected; According to the selected spatial econometric model type, the usage indicators in the historical vehicle usage dataset are used as dependent variables, the external factors affecting vehicle usage are used as independent variables, and the spatial weights of the spatial dependency network are combined to construct the corresponding model equations; Applying the maximum likelihood estimation method to solve the model equations to obtain the spatial autocorrelation coefficient, regression coefficient and standard error, and evaluating the statistical significance of the spatial autocorrelation coefficient through a likelihood ratio test; The spatial autocorrelation coefficient, the regression coefficient and the standard error are substituted into the model equation group to construct a spatial autocorrelation correction model.

[0009] Optionally, in a fifth implementation of the first aspect of the present invention, applying the maximum likelihood estimation method to solve the model equations to obtain the spatial autocorrelation coefficient, the regression coefficient, and the standard error, and evaluating the statistical significance of the spatial autocorrelation coefficient through a likelihood ratio test includes: Constructing a log-likelihood function corresponding to the model equations, and applying a numerical optimization algorithm to the log-likelihood function to iteratively solve for the optimal estimate of the spatial autocorrelation coefficient, the regression coefficient vector, and the error variance, while obtaining the corresponding maximum log-likelihood value; Obtaining a spatial autocorrelation coefficient, a regression coefficient vector, and a standard error of the error variance based on a second-order derivative matrix of the log-likelihood function at the optimal estimate; Constructing a log-likelihood function of a constraint model that does not include spatial autocorrelation terms, solving parameter estimates of the log-likelihood function of the constraint model using a numerical optimization algorithm, and obtaining a maximum log-likelihood value of the constraint model; A likelihood ratio statistic is calculated according to the maximum log-likelihood value of the model equation group and the maximum log-likelihood value of the constraint model, and whether the spatial autocorrelation coefficient is statistically significant is determined based on the likelihood ratio statistic.

[0010] Optionally, in a sixth implementation of the first aspect of the present invention, predicting vehicle demand at rental points based on the spatial autocorrelation correction usage model, and generating a vehicle allocation strategy that considers mutual influence between rental points based on the vehicle demand includes: Inputting the usage status data and external environmental factor data of the current rental point into the spatial autocorrelation correction usage model to generate a vehicle demand forecast value considering spatial correlation; Identifying rental points with imbalanced vehicle supply and demand based on the vehicle demand forecast values, and calculating the vehicle allocation demand between the rental points in combination with the spatial dependency network; Based on the vehicle allocation demand, a multi-objective optimization problem is constructed with the goals of minimizing the total allocation cost and maximizing the demand satisfaction rate, with the distance cost between rental points and the influence intensity in the spatial dependency network as constraints; A heuristic algorithm is used to solve the multi-objective optimization problem and generate a vehicle deployment strategy including deployment path, deployment quantity and priority.

[0011] A second aspect of the present invention provides a rental vehicle usage data analysis system, the rental vehicle usage data analysis system comprising: A data processing module is used to perform spatiotemporal labeling on historical vehicle usage data distributed across different rental locations, construct a spatiotemporal data matrix for vehicle usage, and calculate a spatial weight matrix between rental locations. Based on the spatiotemporal data matrix and the spatial weight matrix, a historical vehicle usage dataset is generated. a pattern recognition module for identifying spatial clustering patterns and spatial heterogeneity characteristics in the historical vehicle usage data based on the historical vehicle usage dataset, and constructing a spatial dependency network of rental vehicle usage behavior using the spatial clustering patterns and the spatial heterogeneity characteristics; a model building module for parameterizing spatial autocorrelation in historical vehicle usage data based on the spatial dependency network and establishing a spatial autocorrelation correction usage model; A deployment strategy module is used to predict the vehicle demand of the rental points based on the spatial autocorrelation correction usage model, and generate a vehicle deployment strategy that takes into account the mutual influence between the rental points based on the vehicle demand.

[0012] The above-mentioned rental vehicle usage data analysis method and system performs spatiotemporal labeling on historical vehicle usage data distributed across different rental locations, constructing a spatiotemporal data matrix and a spatial weight matrix to generate a historical vehicle usage dataset. Based on the historical vehicle usage dataset, spatial clustering patterns and spatial heterogeneity characteristics are identified to construct a spatial dependency network of rental vehicle usage behavior. Based on the spatial dependency network, spatial autocorrelations in the historical vehicle usage data are parameterized and modeled to establish a spatial autocorrelation-corrected usage model. Based on the spatial autocorrelation-corrected usage model, vehicle demand at rental locations is predicted, generating a vehicle allocation strategy that takes into account the mutual influence between rental locations. By introducing spatial metrology methods, the present invention effectively identifies and corrects spatial autocorrelations in rental vehicle usage data, improving the accuracy of vehicle demand predictions and the rationality of vehicle allocation decisions.

[0013] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or understood by practicing the present invention. The purposes and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0014] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a schematic diagram of a first embodiment of a method for analyzing usage data of a rental vehicle according to an embodiment of the present invention; Figure 2 Schematic diagram of an embodiment of a rental vehicle usage data analysis system in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0017] The terms "including," "having," and any variations thereof, as used in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device comprising a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to the process, method, product, or device.

[0018] To facilitate understanding of this embodiment, a method for analyzing usage data of a rental vehicle disclosed in an embodiment of the present invention is first described in detail. Figure 1 As shown, this method includes the following steps: 101. Perform spatiotemporal labeling on historical vehicle usage data distributed at different rental points to construct a spatiotemporal data matrix of vehicle usage, and calculate a spatial weight matrix between rental points. Based on the spatiotemporal data matrix and the spatial weight matrix, a historical vehicle usage dataset is generated. In one embodiment of the present invention, performing spatiotemporal tagging on historical vehicle usage data distributed across different rental points, constructing a spatiotemporal data matrix of vehicle usage, and calculating a spatial weight matrix between rental points, and generating a historical vehicle usage dataset based on the spatiotemporal data matrix and the spatial weight matrix includes: sampling GPS trajectory data of rental vehicles distributed across different rental points in segments according to time granularity, and associating each sampling point with the nearest rental point to form a spatiotemporal association record between the vehicle and the rental point; performing feature extraction on the spatiotemporal association record between the vehicle and the rental point, calculating the vehicle usage frequency, average usage duration, peak usage period, and idle rate for each rental point within different time windows, and constructing a spatiotemporal data matrix comprising multiple rental points and multiple time windows; calculating the Euclidean distance between each rental point based on the geographic coordinate information of the rental point, and converting the Euclidean distance into a spatial weight reflecting spatial proximity using an inverse distance weighting method to construct a spatial weight matrix between the rental points; performing algebraic operations on the spatiotemporal data matrix and the spatial weight matrix to calculate a weighted vehicle usage index for each rental point affected by adjacent rental points, and combining the original usage index with the weighted vehicle usage index to generate the historical vehicle usage dataset.

[0019] Specifically, when sampling GPS trajectory data for rental vehicles distributed across different rental locations at a time granularity, historical GPS trajectory data, including vehicle ID, timestamp, and latitude and longitude coordinates, is first extracted from the rental system database. The sampling process uses a fixed interval method, specifically setting location information to be extracted every 5 minutes. This sampling frequency ensures the continuity of vehicle movement trajectories while reducing the amount of redundant data. Each GPS sampling point is then associated with the nearest rental point using the K-nearest neighbor algorithm. The geographic distance between the sampling point and each rental point is calculated, and the rental point with the minimum distance is selected as the associated point. The resulting vehicle-rental point spatiotemporal association record consists of a four-tuple structure: vehicle ID, timestamp, geographic coordinates, and associated rental point ID. For example, vehicle A123, located at coordinates (116.404, 39.915) at a certain moment, is associated with rental point S001, which is 300 meters away. This spatiotemporal labeling process converts discrete GPS trajectory data into structured data associated with the topological structure of the rental network, effectively establishing a spatial mapping relationship between vehicle usage behavior and the rental point network, and laying a data foundation for identifying the spatial dependencies between rental points.

[0020] Specifically, feature extraction for spatiotemporal associations between vehicle and rental point records begins with defining multiple time window granularities: hourly (24 time periods / day), daily (7 days / week), and weekly (4 weeks / month). For each rental point within each time window, the following four metrics are calculated: vehicle usage frequency (number of rental transactions completed per unit time), average usage duration (average duration of all rental trips), peak usage period (the three-hour period with the highest usage frequency within a 24-hour period), and idle rate (the proportion of time a vehicle is available but not used). For example, rental point S001 has a usage frequency of 8.5 times / hour between 5:00 PM and 8:00 PM on weekdays, an average usage duration of 45 minutes, and an idle rate of 15%. These calculations construct a three-dimensional spatiotemporal data matrix consisting of rental point ID, time window, and feature dimension. This multidimensional feature extraction fully exploits the spatiotemporal patterns of rental behavior, not only reflecting the usage intensity and cyclical changes of the rental point itself but also providing multi-faceted observational data for subsequent analysis of the interactions between rental points.

[0021] Specifically, when calculating the spatial weight matrix based on the geographic coordinates of the rental points, the straight-line distance between any two rental points is first calculated using the Euclidean distance formula. The inverse distance weighting method is then used to convert these distances into spatial weights. The closer the distance, the greater the impact between rental points, and the higher the weight. During this conversion, when two points are not the same, the weight is inversely proportional to the square of the distance; when the two points are the same, the weight is set to 0. The weight decay exponent of 2 is based on empirical evidence that the impact intensity in urban traffic environments is inversely proportional to the square of the distance. The weight matrix is then row-normalized so that the sum of the weights in each row is 1, ensuring that the overall impact intensity across different rental points is comparable. The resulting spatial weight matrix is a square matrix, where the matrix elements represent the spatial impact intensity of one rental point on another. For example, if rental point S002 is 1 kilometer away from S001, S002's initial weight on S001 is larger. After row normalization, it might be 0.3, indicating that 30% of the total impact on S001 comes from S002. This spatial weight calculation method based on geographic distance accurately quantifies the strength of spatial dependencies between rental points and provides a mathematical expression for simulating the spatial conduction effect of vehicle usage behavior.

[0022] Specifically, when performing algebraic operations on the spatiotemporal data matrix and the spatial weight matrix, matrix multiplication is performed to multiply the spatial weight matrix by the spatiotemporal data matrix. This operation is essentially a spatial lag process. By multiplying the raw usage index value of each rental point by the corresponding spatial weight and performing a weighted summation by rental point, a weighted usage index influenced by surrounding rental points is obtained. For example, the weighted usage frequency of rental point A is equal to the usage frequency of all neighboring rental points multiplied by the sum of their spatial weights for A. The weighted index calculated here reflects the spatial spillover effect, that is, the degree to which the usage of a rental point is influenced by the status of surrounding rental points. Subsequently, the raw usage index of each rental point is combined with the weighted usage index to form an expanded feature set. This combination uses feature concatenation, expanding the original four indicators to eight indicators (the original four + the weighted four) for each time window of each rental point. The resulting historical vehicle usage dataset fully preserves the characteristics of the rental points and their spatial correlation characteristics, enabling the data to reflect both independent usage patterns and spatial dependencies, providing a comprehensive feature representation for subsequent spatial clustering analysis.

[0023] 102. Identifying spatial clustering patterns and spatial heterogeneity characteristics in the historical vehicle usage data based on the historical vehicle usage dataset, and constructing a spatial dependency network of rental vehicle usage behavior using the spatial clustering patterns and the spatial heterogeneity characteristics; In one embodiment of the present invention, identifying spatial clustering patterns and spatial heterogeneity characteristics in historical vehicle usage data based on the historical vehicle usage dataset, and constructing a spatial dependency network of rental vehicle usage behavior using the spatial clustering patterns and spatial heterogeneity characteristics includes: calculating a global Moran's index value for the historical vehicle usage dataset, and determining whether there is statistically significant spatial autocorrelation in the overall rental network based on the global Moran's index value using a Monte Carlo simulation method; if so, calculating a local spatial autocorrelation index for each rental point based on the historical vehicle usage dataset, and performing a significance test on the local spatial autocorrelation index to identify statistically significant high-value clustering areas, low-value clustering areas, and spatial outliers; calculating spatial influence coefficients between rental points based on the distribution characteristics of the high-value clustering areas, low-value clustering areas, and spatial outliers, and constructing a spatial flow matrix that characterizes the mutual influence of vehicle usage behavior between rental points; and converting the spatial flow matrix into a directed weighted network structure to form a spatial dependency network of rental vehicle usage behavior.

[0024] Specifically, when calculating the global Moran's index for a historical vehicle usage dataset, key usage metrics, such as vehicle usage frequency, are first extracted from the dataset as the target variable for spatial analysis. The global Moran's index is a statistic that measures overall spatial autocorrelation. Values close to 1 indicate clustering of similar values, close to -1 indicate clustering of dissimilar values, and close to 0 indicate random distribution. The calculation uses the spatial weight matrix generated above to normalize the usage metrics for each rental point. The spatial covariance of the metrics is then calculated based on spatial proximity. A Monte Carlo simulation is then performed to randomly permute the usage metrics for each rental point. This process is repeated 999 times, with the Moran's index recalculated each time to form a null hypothesis distribution. The observed values are compared with the null distribution, and a p-value is calculated. A p-value less than 0.05 indicates statistically significant spatial autocorrelation in the rental network. This step requires spatial statistical analysis software, such as GeoDa or the spdep package in R. For example, the calculation of vehicle usage frequency at 58 rental points in a certain area of Beijing resulted in a global Moran index value of 0.68 and a p-value of 0.001, which clearly showed that the usage frequency showed a significant clustered distribution in space rather than a random distribution, thus confirming the necessity of further spatial analysis.

[0025] Specifically, after confirming the presence of global spatial autocorrelation, local spatial autocorrelation indices are calculated for each rental site based on the historical vehicle usage dataset. These indices primarily include the Local Moran's Index (LISA) and the Getis-Ord Gi statistic. The LISA reflects the degree of similarity between a single rental site and its neighbors, while the Gi statistic identifies spatial clusters of high or low values. For each rental site, the calculation considers the product of its usage index value and the weighted average of the surrounding rental sites. These local indices are then tested for significance. A permutation test calculates a p-value for each point, typically with a p-value threshold of 0.05, to determine which rental sites exhibit statistically significant local spatial correlation. Based on the significance test results and the indices' signs, rental sites are classified into four types: high-high clusters (high-value sites are surrounded by high-value sites), low-low clusters (low-value sites are surrounded by low-value sites), high-low outliers (high-value sites are surrounded by low-value sites), and low-high outliers (low-value sites are surrounded by high-value sites). For example, a cluster of rental points in a city's commercial district may form high-high clustering, indicating that the overall rental demand in the area is strong; while isolated rental points located in residential areas may show low-low clustering, reflecting a generally low frequency of use.

[0026] Specifically, the spatial influence coefficients between rental points are calculated based on the distribution characteristics of identified high-value clusters, low-value clusters, and spatial outliers. Unlike the previous spatial weight matrix based on physical distance, the spatial flow matrix constructed in this step takes into account the actual correlation between rental point usage behavior. Specifically, pairs of rental points within the same cluster type (e.g., points belonging to both high-high clusters) are assigned a higher positive influence coefficient. Pairs of rental points between different types of areas (e.g., points between high-high clusters and low-low clusters) are assigned a lower or negative influence coefficient. For boundary areas or between outliers and points of other types, the influence coefficient is dynamically calculated based on the relative magnitude of their local indicators. Furthermore, the actual vehicle flow frequency between rental points is considered, capturing direct interactions by analyzing historical vehicle movement records from one rental point to another. The resulting spatial flow matrix represents the mutual influence patterns of vehicle usage behavior between rental points. The magnitude of the matrix elements reflects the strength of the influence, while the sign indicates the direction of the influence (positive values indicate positive correlation, negative values indicate negative correlation).

[0027] Specifically, when converting the spatial flow matrix into a directed weighted network structure, each nonzero element in the matrix is interpreted as an edge in the network. The element value is the edge weight, and the row and column indices of the element correspond to the starting and ending rental points of the edge, respectively. To improve the interpretability of the network, a threshold is typically set to retain only influence relationships with an absolute value greater than the threshold, filtering out weak connections. The converted network is then visualized using a graph visualization tool (such as Gephi or NetworkX). Nodes represent rental points, node size reflects usage frequency, node color corresponds to clustering type (e.g., red for high-high clustering, blue for low-low clustering), edge thickness indicates influence strength, and edge color indicates influence direction (e.g., green for positive influence, gray for negative influence). The resulting spatial dependency network of rental vehicle usage behavior intuitively illustrates the complex interaction patterns between rental points, identifying key hubs (high-indegree nodes), influence sources (high-outdegree nodes), and relatively independent points (low-connectivity nodes). For example, in actual applications, rental point A in a certain business district may be a typical source of influence, and changes in its usage patterns will transmit and affect multiple surrounding rental points; while rental point B located at a transportation transfer hub may be a key hub point, which is affected by multiple factors while also transmitting influences outward.

[0028] Furthermore, the calculation of a global Moran's index value for the historical vehicle usage data set and the determination of whether there is statistically significant spatial autocorrelation in the overall rental network based on the global Moran's index value by a Monte Carlo simulation method include: standardizing the vehicle usage index in the historical vehicle usage data set to obtain a standardized usage index; calculating a global Moran's index value based on the standardized usage index and the spatial weight matrix using a spatial statistical algorithm; randomly permuting the standardized usage index by a Monte Carlo simulation method, recalculating the global Moran's index value after each permutation, repeating the random permutation a preset number of times to form a null distribution of the global Moran's index; calculating a p-value of an actual observation value based on the null distribution of the global Moran's index; when the p-value is less than a preset significance level, determining that there is statistically significant spatial autocorrelation in the overall rental network; and when the p-value is greater than or equal to the preset significance level, determining that there is no spatial autocorrelation and terminating subsequent analysis.

[0029] Specifically, when normalizing vehicle usage metrics in a historical vehicle usage dataset, key metrics for analysis, such as vehicle usage frequency and average usage duration, are first selected from the dataset. Normalization employs the Z-score method: For each rental location i, the mean μ for that metric across all rental locations is subtracted, and the result is divided by the standard deviation σ to obtain the normalized metric value z_i. This normalization eliminates dimensional differences between metrics, transforming the data into a distribution with a mean of 0 and a standard deviation of 1. For example, if the original usage frequency for a rental location is 8 times per hour, while the average for all rental locations is 5 times per hour with a standard deviation of 2 times per hour, the normalized usage frequency for that location is 1.5, indicating that its usage frequency is 1.5 standard deviations above the average. Normalization is crucial for spatial autocorrelation analysis because it eliminates the influence of data units and absolute values, ensuring that the analysis results reflect only the patterns of relative positional relationships and relative value differences between rental locations, rather than the influence of the original data scale.

[0030] Specifically, when calculating the global Moran's index using a spatial statistical algorithm, the previously generated spatial weight matrix and the standardized usage index are combined. The calculation process considers the multiplication and weighting of the usage index values between each pair of rental locations. In implementation, a programmatic approach is used to calculate the product of the index value of each rental location with that of all other locations, weighted by the spatial weight as a coefficient, and finally summed and normalized to obtain the global Moran's index value I. For example, in a certain city's rental system, the global Moran's index for vehicle usage frequency at 60 rental locations was calculated to be 0.65, which is close to 1, indicating that rental locations with similar usage frequencies tend to cluster spatially. The index ranges from [-1 to 1]: positive values indicate spatial clustering of similar values (high-valued locations are surrounded by high-valued locations, and low-valued locations are surrounded by low-valued locations), negative values indicate spatial clustering of dissimilar values (high-valued locations are surrounded by low-valued locations), and values close to zero indicate random distribution. The global Moran's index provides an overall assessment of the spatial patterns of the entire rental network and serves as a preliminary basis for determining whether spatial autocorrelation exists.

[0031] Specifically, when randomly permuting the standardized usage index using a Monte Carlo simulation method, the spatial location and spatial weight matrix of the rental points remain unchanged, and only the index values at each point are randomly redistributed. Specifically, a sampling method without replacement is used to randomly assign the original standardized index values to each rental point. The global Moran's index is then recalculated based on this random assignment. This process is repeated a preset number of times, typically 999 times, to form the null distribution of the global Moran's index. Each simulation represents the "absence of spatial autocorrelation" scenario, where the index values are completely randomly distributed in space. These 999 randomly generated index values constitute an empirical distribution, representing the range of possible Moran's index values under the null hypothesis (absence of spatial autocorrelation). For example, after 999 random permutations of a city's rental network, the resulting null distribution of the Moran's index has a mean close to 0 and a standard deviation of approximately 0.1, reflecting the expected performance of the index under the assumption of spatial randomness.

[0032] Specifically, based on the null distribution of the global Moran's index, the p-value for the observed value is calculated by counting the number of simulations that generated an index value greater than or equal to the observed value and dividing it by the total number of simulations plus 1 (i.e., 999 + 1 = 1000). The p-value measures the probability of observing the current or more extreme Moran's index value, assuming the null hypothesis holds. The default significance level is typically 0.05, representing a 5% statistical significance threshold. A calculated p-value less than 0.05 indicates that the observed spatial pattern is unlikely to occur by chance, and therefore rejects the null hypothesis of "no spatial autocorrelation." Statistically significant spatial autocorrelation is determined in the rental network, allowing for further local spatial analysis. Conversely, if the p-value is greater than or equal to 0.05, the null hypothesis is accepted, concluding that rental vehicle usage data is spatially randomly distributed, and no further spatial dependency analysis is necessary. For example, the Moran's index calculated from actual data is 0.65, but only 3 out of 999 random simulations produced values greater than or equal to 0.65. The p-value is 4 / 1000=0.004, which is much smaller than 0.05, clearly indicating the existence of significant spatial autocorrelation.

[0033] 103. Analyze the spatial dependency network based on the usage data of rental vehicles, perform parameterized modeling on the spatial autocorrelation in the historical vehicle usage data, and establish a spatial autocorrelation correction usage model; In one embodiment of the present invention, parameterizing the spatial autocorrelation in the historical vehicle usage data based on the spatial dependency network and establishing a spatial autocorrelation correction usage model includes: selecting a corresponding spatial econometric model type based on the structural characteristics of the spatial dependency network; according to the selected spatial econometric model type, using the usage indicators in the historical vehicle usage data set as dependent variables, using the external factor data affecting vehicle usage as independent variables, and combining the spatial weights of the spatial dependency network to construct a corresponding model equation group; applying the maximum likelihood estimation method to solve the model equation group to obtain the spatial autocorrelation coefficient, regression coefficient and standard error, and evaluating the statistical significance of the spatial autocorrelation coefficient through a likelihood ratio test; substituting the spatial autocorrelation coefficient, the regression coefficient and standard error into the model equation group to construct a spatial autocorrelation correction usage model.

[0034] Specifically, when selecting a spatial econometric model based on the structural characteristics of a spatial dependency network, it is necessary to analyze the specific manifestations of spatial autocorrelation. First, the most appropriate spatial econometric model type is determined by examining the network's density, connectivity patterns, and directionality. The main spatial econometric models include the spatial lag model (SLM), the spatial error model (SEM), and the spatial Durbin model (SDM). When the spatial dependency network indicates that the use of a rental location is directly influenced by the use of surrounding rental locations, the spatial lag model is selected. When the spatial correlation is primarily manifested in the error term, meaning that unobserved factors exhibit spatial spillover effects, the spatial error model is selected. When both direct spatial influences and spatial spillover effects of independent variables are present, the spatial Durbin model is selected. For example, an analysis of a city's rental network revealed that nearby locations with high usage rates also tend to have high usage rates, and the correlation is primarily reflected in usage behavior rather than the error structure. Therefore, the spatial lag model was selected as the preferred model type. Model selection also needs to consider the topological characteristics of the network. For example, different model settings may be required for star network structures (the central point affects the surrounding points) and grid structures (uniform influence between points) to ensure that the selected model matches the actual spatial dependency structure.

[0035] Specifically, when constructing the model equation system based on the selected spatial econometric model type, usage indicators (such as frequency or duration of use) in the historical vehicle usage dataset are used as dependent variables, and data on external factors affecting vehicle use are selected as independent variables. These external factors include population density around the rental point, the number of commercial facilities, public transportation accessibility, weather conditions, weekday / weekend signs, etc. At the same time, the spatial weights of the spatial dependency network are combined to construct a complete model equation system. Taking the spatial lag model as an example, its equation system includes the dependent variable, the spatial lag term (the weighted average of the dependent variables of the surrounding rental points), the independent variable set, and the error term. The spatial lag term is obtained by multiplying the spatial weight matrix with the dependent variable vector, reflecting the impact of the surrounding rental points on the current point. During the construction process, the identification conditions of the model must also be considered to ensure that the parameters are estimable. For example, in a certain practical application, the dependent variable is the vehicle utilization rate during peak hours (17:00-19:00), and the independent variables include the number of commercial facilities within 500 meters of the rental point, the distance to the nearest subway station, the surrounding road congestion index, and the weekday / weekend indicator variable. Together with the spatial weight matrix generated by the spatial dependency network, they constitute a complete set of model equations.

[0036] Specifically, when applying the maximum likelihood estimation method to solve the model equations, the log-likelihood function is first constructed. This function expresses the relationship between the probability of observing the current data and the model parameters. A numerical optimization algorithm, such as the Newton-Raphson method or the BFGS algorithm, is then used to iteratively find the parameter combination that maximizes the log-likelihood function. These parameters include the spatial autocorrelation coefficient (indicating the strength of spatial dependence), the regression coefficient (indicating the influence of each variable on the dependent variable), and the variance of the error term. During the optimization process, the algorithm starts with the initial parameter values and continuously adjusts the parameter values based on the gradient information of the log-likelihood function until convergence to a local optimal solution. After the solution is completed, the standard error of each parameter is calculated using the second-order derivative matrix (information matrix) of the log-likelihood function at the optimal solution to assess the statistical reliability of the parameter estimates. Finally, the statistical significance of the spatial autocorrelation coefficient was assessed using a likelihood ratio test. This involves constructing two models: a complete model that includes the spatial term and a constrained model that excludes the spatial term. The difference in the log-likelihood between the two is calculated and multiplied by two. The resulting statistic follows a chi-square distribution. If this statistic is greater than the critical value, the spatial autocorrelation term is considered statistically significant. For example, the estimated spatial autocorrelation coefficient for a city rental network model is 0.45, with a standard error of 0.08. The likelihood ratio test p-value is 0.0002, indicating that the spatial autocorrelation effect is statistically significant.

[0037] Specifically, the spatial autocorrelation coefficient, regression coefficient, and standard error are substituted into the model equations. When constructing a spatial autocorrelation correction model, the parameter values obtained in the previous step need to be entered into the corresponding equations. Taking the spatial lag model as an example, the estimated spatial autocorrelation coefficient ρ is substituted before the spatial lag term in the equation, and each regression coefficient β is substituted before the corresponding independent variable to form a specific prediction equation. This parameterized model equation is no longer an abstract expression containing unknown parameters, but a mathematical tool that can be used for practical prediction. This spatial autocorrelation correction model accurately accounts for the spatial dependencies between rental locations, avoiding the parameter estimation bias of traditional models assuming independence. The model's standard error information is used to construct prediction intervals, providing a reliability assessment of the prediction results. In practical applications, this model can be expressed as a specific numerical equation that intuitively demonstrates the magnitude and direction of spatial effects and the influence of various factors, providing a quantitative framework for understanding the spatial dynamics of rental vehicle use.

[0038] Furthermore, the method of applying the maximum likelihood estimation method to solve the model equation group to obtain the spatial autocorrelation coefficient, regression coefficient and standard error, and evaluating the statistical significance of the spatial autocorrelation coefficient through a likelihood ratio test includes: constructing a log-likelihood function corresponding to the model equation group, and applying a numerical optimization algorithm to the log-likelihood function, solving the optimal estimate of the spatial autocorrelation coefficient, regression coefficient vector and error variance in an iterative manner, and obtaining the corresponding maximum log-likelihood value; obtaining the spatial autocorrelation coefficient, regression coefficient vector and standard error of the error variance based on the second-order derivative matrix of the log-likelihood function at the optimal estimate; constructing a log-likelihood function of a constraint model that does not contain spatial autocorrelation terms, using a numerical optimization algorithm to solve the parameter estimate of the log-likelihood function of the constraint model, and obtaining the maximum log-likelihood value of the constraint model; calculating a likelihood ratio statistic based on the maximum log-likelihood value of the model equation group and the maximum log-likelihood value of the constraint model, and judging whether the spatial autocorrelation coefficient is statistically significant based on the likelihood ratio statistic.

[0039] Specifically, when constructing the log-likelihood function corresponding to the model equations, the corresponding log-likelihood expression is established based on the spatial econometric model type selected above. Taking the spatial lag model as an example, its log-likelihood function consists of three key components: a constant term, a spatial coefficient term (involving the logarithm of the determinant of the spatial weight matrix), and the sum of squared error terms. The construction of the log-likelihood function relies on the distributional assumptions about the model's error terms, which are typically assumed to follow a normal distribution. In practical implementation, numerical computing software such as the spdep package in R or the PySAL library in Python is used to construct the complete log-likelihood function based on the given spatial weight matrix, dependent variable (vehicle usage indicator), and independent variable (influencing factor) data. Subsequently, a numerical optimization algorithm, such as the Newton-Raphson method or the BFGS algorithm, is applied to the function to iteratively search the parameter space. Starting from initial parameter values (typically derived from ordinary least squares estimation results), the optimization process continuously updates the parameter values based on the gradient information of the log-likelihood function until convergence to a local optimal solution. In each iteration, the algorithm calculates the log-likelihood value and its gradient for the current parameter value, determining the next search direction and step size. Ultimately, the algorithm converges to the parameter combination that maximizes the log-likelihood function, obtaining the optimal estimates of the spatial autocorrelation coefficient, the regression coefficient vector, and the error variance. The corresponding maximum log-likelihood value is also recorded, representing the logarithmic probability of observing the current data under the optimal parameters.

[0040] Specifically, when calculating the standard error based on the second-order derivative matrix of the log-likelihood function at the optimal estimate (i.e., the information matrix), it is first necessary to obtain the Hessian matrix, which is a matrix composed of the second-order partial derivatives of the log-likelihood function with respect to each parameter. In numerical implementation, this can be achieved by numerically differentiating the log-likelihood function values near the optimal point, or by using the Hessian matrix approximation calculated within the optimization algorithm. Then, the matrix is inverted to obtain the variance-covariance matrix of the parameter estimates. The elements on the diagonal of this matrix are the variances of the parameter estimates, and taking their square roots gives the standard errors of each parameter. The standard error provides a measure of the accuracy of the parameter estimate, reflecting the possible fluctuation range of the parameter estimate. The standard error of the spatial autocorrelation coefficient is particularly important, as it is directly related to the reliability of the assessment of the strength of spatial dependence. A smaller standard error indicates that the parameter estimate is more accurate and the estimated value is closer to the true value. In analyzing rental vehicle usage patterns, the standard error of the regression coefficient helps determine the significance of the impact of various factors (such as commercial density and transportation accessibility) on usage. The smaller the standard error, the greater the certainty of the factor's influence. The standard error also allows for the calculation of the t-statistic (parameter estimate divided by its standard error), which is used to test the significance of individual parameters.

[0041] Specifically, when constructing a constrained model that excludes spatial autocorrelation, the spatial autocorrelation term is removed from the original model, retaining only the direct influence of the independent variable on the dependent variable. Essentially, this is equivalent to constraining the spatial autocorrelation coefficient to zero and constructing a traditional multivariate linear regression model. The log-likelihood function of the constrained model is simpler and does not include the spatial coefficient term. Numerical optimization algorithms (such as least squares or maximum likelihood) are also used to solve the constrained model, obtaining estimates of the regression coefficients and error variance. Because the constrained model is a special case of the original model (i.e., when the spatial autocorrelation coefficient is zero), its fit is generally inferior to that of the full model. After optimization, the maximum log-likelihood value of the constrained model is recorded. This value indicates how well the model fits the observed data without considering spatial autocorrelation. In the analysis of rental vehicle usage data, the constrained model is equivalent to a traditional usage rate prediction model. It assumes that each rental point is independent of each other and that the usage of a point is unaffected by surrounding points. This does not conform to the spatial dependence observed in actual rental networks, resulting in poor fit and a low maximum log-likelihood value.

[0042] Specifically, the likelihood ratio statistic is calculated based on the maximum log-likelihood values of the full model and the constrained model using the formula LR = 2*(L1-L0), where L1 is the maximum log-likelihood value of the full model and L0 is the maximum log-likelihood value of the constrained model. This statistic measures the difference in goodness-of-fit between the two models and reflects the improvement in model fit after accounting for spatial autocorrelation. When the assumptions are met, the likelihood ratio statistic follows a chi-square distribution with the number of degrees of freedom equal to the number of constraints (in this case, 1, i.e., the spatial autocorrelation coefficient). Based on this distribution, the corresponding p-value is calculated, which represents the probability of observing the current or more extreme likelihood ratio value. When the p-value is less than the preset significance level (usually 0.05), the spatial autocorrelation coefficient is considered statistically significant, indicating that spatial dependence exists in the rental network and that the full spatial autocorrelation model outperforms traditional models that ignore spatial effects. Conversely, if the p-value is greater than or equal to the significance level, the hypothesis of "zero spatial autocorrelation coefficient" cannot be rejected, indicating that spatial effects are likely insignificant. In this case, a simpler non-spatial model can be considered. In practical applications, the results of the significance test of the spatial autocorrelation coefficient directly affect the formulation of subsequent modeling strategies and resource allocation decisions.

[0043] 104. Based on the spatial autocorrelation correction usage model, predict the vehicle demand of the rental points, and generate a vehicle deployment strategy that takes into account the mutual influence between the rental points based on the vehicle demand.

[0044] In one embodiment of the present invention, the method of predicting vehicle demand at rental points based on the spatial autocorrelation correction usage model and generating a vehicle allocation strategy based on the vehicle demand that takes into account the mutual influence between rental points includes: inputting the usage status data and external environmental factor data of the current rental point into the spatial autocorrelation correction usage model to generate a vehicle demand prediction value that takes into account spatial correlation; identifying rental points with an imbalance in vehicle supply and demand based on the vehicle demand prediction value, and calculating the vehicle allocation demand between rental points in combination with the spatial dependency network; constructing a multi-objective optimization problem based on the vehicle allocation demand with the goals of minimizing the total allocation cost and maximizing the demand satisfaction rate, using the distance cost between rental points and the influence intensity in the spatial dependency network as constraints; and using a heuristic algorithm to solve the multi-objective optimization problem to generate a vehicle allocation strategy that includes an allocation path, allocation quantity, and priority.

[0045] Specifically, to input current rental point usage data and external environmental factors into the spatial autocorrelation correction model, real-time data must first be collected. This includes the current vehicle inventory level at each rental point, usage frequency over the past 24 hours, and current external environmental factors such as weather conditions, whether it is a weekday, and surrounding traffic congestion index. This real-time data must maintain the same variable type and data structure as the historical data used in the modeling phase, requiring data preprocessing, including standardization and missing value handling. Once the data is prepared, it is fed into the constructed spatial autocorrelation correction model. Given the characteristics of the spatial lag model, the forecast for each rental point must consider both its own characteristics and the influence of surrounding rental points. The model output includes a forecast of vehicle demand for each rental point within the forecast period (e.g., the next 2, 4, or 24 hours). This forecast already incorporates spatial correlation correction, making it more accurate than predictions from traditional standalone models. For example, the model may predict that rental point A located in a commercial district will have a demand for 15 vehicles between 5 and 7 pm. However, considering the influence of surrounding rental points B and C, the actual demand may be adjusted to 17 vehicles. This adjustment reflects the spatial spillover effect of customer flow and selection behavior between rental points.

[0046] Specifically, when identifying rental locations with supply-demand imbalances based on vehicle demand forecasts, the current vehicle inventory at each location is compared with the forecasted demand, and the difference is calculated. A positive difference indicates a vehicle surplus, while a negative difference indicates a vehicle shortage. The larger the absolute value of the difference, the more severe the imbalance. For example, if rental location D currently has 10 vehicles, but the forecasted demand for 15 within the next three hours is 35, there is a shortage of 5 vehicles. Meanwhile, if rental location E has 20 vehicles but the forecasted demand is only 12, there is a surplus of 8 vehicles. After identifying the supply-demand imbalance points, the spatial dependency network is used to calculate the vehicle allocation requirements between rental locations. This calculation not only considers the shortage or surplus at a single location, but also the spatial influence relationships between rental locations. The allocation requirement calculation utilizes network flow analysis, using the connection strengths in the spatial dependency network as flow weights. Potential allocation paths are constructed from points with surpluses to points with shortages, while also considering the timeliness of allocations. In specific implementation, for each shortage point, priority is given to allocating vehicles from rental points with high spatial correlation and surplus vehicles. This can not only meet the needs of the shortage points, but also maintain service balance among spatially related points.

[0047] Specifically, when constructing a multi-objective optimization problem based on vehicle dispatch demand, two primary optimization objectives are set: minimizing total dispatch costs and maximizing the demand fulfillment rate. Dispatching costs include vehicle travel distance costs (proportional to the physical distance between rental points), human resource costs (proportional to the number of dispatch operations), and time costs (proportional to the urgency of the dispatch). The demand fulfillment rate measures the proportion of vehicle shortages that can be met through dispatch. This multi-objective setting reflects the trade-off between cost control and service quality in real-world dispatch decisions. Constraints in the optimization problem include: the distance cost between rental points must not exceed a preset threshold; the dispatch path must follow the actual road network; the number of dispatches must not exceed the available dispatch capacity of surplus points; and influence strength constraints within the spatial dependency network. In particular, the influence strength constraint prioritizes dispatches between rental points with strong spatial correlations. This constraint, based on the reality of frequent customer flow between related points, improves dispatch accuracy. For example, if the connection strength between rental points F and G in the spatial dependency network is high (e.g., 0.8), dispatches from F to G will be given a higher priority.

[0048] Specifically, when using heuristic algorithms to solve multi-objective optimization problems, evolutionary algorithms such as genetic algorithms (GAs) or particle swarm optimization (PSOs) are suitable for handling complex optimization problems with multiple objectives and nonlinear constraints. During algorithm implementation, the solution encoding method is first defined, typically using integer or real numbers to represent the allocation quantity and path selection from each surplus point to each shortage point. An initial population is then set, typically consisting of 50-100 individuals, randomly generating allocation solutions that satisfy basic constraints. An iterative optimization process is then performed. Each iteration includes selection, crossover, mutation, and elite retention. Through continuous evolution, a Pareto-optimal solution set is found—a set of solutions that achieve different trade-offs between total cost and satisfaction rate. The number of iterations is typically set to 100-200, or until the optimal solution has not significantly improved over multiple generations. Finally, the optimal compromise solution from the Pareto solution set is selected as the final allocation strategy. This strategy includes the specific allocation path (e.g., 3 vehicles from point H to point I, 2 vehicles from point J to point K), the number of vehicles to be allocated, and the execution priority. For example, for a dispatch task before a predicted peak period, the algorithm may generate a plan with a total dispatch distance of 25 kilometers and a satisfaction rate of 95%, clearly specifying seven dispatch operations and sorting them by urgency to guide fleet managers to execute them efficiently.

[0049] In this embodiment, historical vehicle usage data distributed across different rental locations is spatiotemporally labeled to construct a spatiotemporal data matrix and a spatial weight matrix, generating a historical vehicle usage dataset. Spatial clustering patterns and spatial heterogeneity characteristics are identified based on the historical vehicle usage dataset, and a spatial dependency network of rental vehicle usage behavior is constructed. Based on the spatial dependency network, spatial autocorrelations in the historical vehicle usage data are parameterized and modeled to establish a spatial autocorrelation-corrected usage model. Based on the spatial autocorrelation-corrected usage model, vehicle demand at rental locations is predicted, generating a vehicle allocation strategy that considers the interactions between rental locations. By introducing spatial metrology methods, this invention effectively identifies and corrects spatial autocorrelations in rental vehicle usage data, improving the accuracy of vehicle demand forecasts and the rationality of vehicle allocation decisions.

[0050] The above describes the method for analyzing the usage data of a rental vehicle in an embodiment of the present invention. The following describes the system for analyzing the usage data of a rental vehicle in an embodiment of the present invention. Figure 2 An embodiment of the rental vehicle usage data analysis system in the embodiment of the present invention includes: Data processing module 201 is used to perform spatiotemporal labeling on historical vehicle usage data distributed across different rental locations, construct a spatiotemporal data matrix of vehicle usage, and calculate a spatial weight matrix between rental locations. Based on the spatiotemporal data matrix and the spatial weight matrix, a historical vehicle usage dataset is generated. a pattern recognition module 202 for identifying spatial clustering patterns and spatial heterogeneity characteristics in the historical vehicle usage data based on the historical vehicle usage dataset, and constructing a spatial dependency network of rental vehicle usage behavior using the spatial clustering patterns and the spatial heterogeneity characteristics; A model building module 203 is used to perform parameterized modeling on the spatial autocorrelation in the historical vehicle usage data based on the spatial dependency network, and to establish a spatial autocorrelation correction usage model; The deployment strategy module 204 is configured to predict vehicle demand at rental points based on the spatial autocorrelation correction usage model, and generate a vehicle deployment strategy that takes into account the mutual influence between rental points based on the vehicle demand.

[0051] In an embodiment of the present invention, the rental vehicle usage data analysis system operates the rental vehicle usage data analysis method described above. The system performs spatiotemporal labeling on historical vehicle usage data distributed across different rental locations, constructing a spatiotemporal data matrix and a spatial weight matrix to generate a historical vehicle usage dataset. Based on the historical vehicle usage dataset, the system identifies spatial clustering patterns and spatial heterogeneity characteristics, constructing a spatial dependency network of rental vehicle usage behavior. Based on the spatial dependency network, the system parameterizes the spatial autocorrelation in the historical vehicle usage data to establish a spatial autocorrelation correction usage model. Based on the spatial autocorrelation correction usage model, the system predicts vehicle demand at rental locations and generates a vehicle allocation strategy that takes into account the mutual influence between rental locations. By introducing spatial metrology methods, the present invention effectively identifies and corrects spatial autocorrelation in rental vehicle usage data, thereby improving the accuracy of vehicle demand forecasts and the rationality of vehicle allocation decisions.

[0052] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0053] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0054] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for analyzing usage data of a rental vehicle, characterized in that: The method for analyzing usage data of a rental vehicle comprises: Performing spatiotemporal labeling on historical vehicle usage data distributed across different rental locations to construct a spatiotemporal data matrix for vehicle usage, and calculating a spatial weight matrix between rental locations. Based on the spatiotemporal data matrix and the spatial weight matrix, a historical vehicle usage dataset is generated. Identifying spatial clustering patterns and spatial heterogeneity characteristics in the historical vehicle usage data based on the historical vehicle usage dataset, and constructing a spatial dependency network of rental vehicle usage behavior using the spatial clustering patterns and the spatial heterogeneity characteristics; Based on the spatial dependency network, parameterizing the spatial autocorrelation in the historical vehicle usage data to establish a spatial autocorrelation correction usage model; Based on the spatial autocorrelation correction usage model, vehicle demand for rental points is predicted, and a vehicle deployment strategy that takes into account the mutual influence between rental points is generated based on the vehicle demand.

2. The method for analyzing usage data of a rental vehicle according to claim 1, characterized in that: The process of performing spatiotemporal tagging on the historical vehicle usage data distributed at different rental points, constructing a spatiotemporal data matrix for vehicle usage, and calculating a spatial weight matrix between rental points, and generating a historical vehicle usage dataset based on the spatiotemporal data matrix and the spatial weight matrix includes: The GPS trajectory data of rental vehicles distributed at different rental points are sampled in segments according to time granularity, and each sampling point is associated with the nearest rental point to form a vehicle-rental point spatiotemporal association record; Extracting features from the spatiotemporal association records of the vehicle and rental point, calculating the vehicle usage frequency, average usage duration, peak usage period, and idle rate of each rental point in different time windows, and constructing a spatiotemporal data matrix containing multiple rental points and multiple time windows; Based on the geographic coordinate information of the rental points, the Euclidean distance between each rental point is calculated, and the Euclidean distance is converted into a spatial weight reflecting spatial proximity using the inverse distance weighting method to construct a spatial weight matrix between the rental points; Performing algebraic operations on the spatiotemporal data matrix and the spatial weight matrix, calculating a weighted vehicle usage index of each rental point affected by adjacent rental points, and combining the original usage index with the weighted vehicle usage index to generate a historical vehicle usage dataset.

3. The method for analyzing usage data of a rental vehicle according to claim 1, characterized in that: The identifying of spatial clustering patterns and spatial heterogeneity characteristics in the historical vehicle usage data based on the historical vehicle usage dataset, and constructing a spatial dependency network of rental vehicle usage behavior using the spatial clustering patterns and the spatial heterogeneity characteristics includes: Calculating a global Moran's index value for the historical vehicle usage data set, and determining whether there is statistically significant spatial autocorrelation in the overall rental network based on the global Moran's index value using a Monte Carlo simulation method; If yes, then calculate the local spatial autocorrelation index of each rental point based on the historical vehicle usage dataset, perform a significance test on the local spatial autocorrelation index, and identify statistically significant high-value cluster areas, low-value cluster areas, and spatial outliers; Based on the distribution characteristics of the high-value clustering areas, low-value clustering areas, and spatial outliers, the spatial influence coefficients between rental points are calculated, and a spatial flow matrix representing the mutual influence of vehicle usage behaviors between rental points is constructed; The spatial flow matrix is converted into a directed weighted network structure to form a spatial dependency relationship network of rental vehicle usage behavior.

4. The method for analyzing usage data of a rental vehicle according to claim 3, characterized in that: Calculating a global Moran's index value for the historical vehicle usage dataset and determining whether there is statistically significant spatial autocorrelation in the overall rental network based on the global Moran's index value using a Monte Carlo simulation method include: Standardizing the vehicle usage index in the historical vehicle usage dataset to obtain a standardized usage index; Calculating a global Moran's index value based on the standardized usage index and the spatial weight matrix using a spatial statistical algorithm; Randomly permuting the standardized usage index using a Monte Carlo simulation method, recalculating the global Moran's index value after each permutation, and repeating the random permutation a preset number of times to form a zero distribution of the global Moran's index; Based on the null distribution of the global Moran's index, the p-value of the actual observation value is calculated. When the p-value is less than the preset significance level, it is determined that there is statistically significant spatial autocorrelation in the overall rental network. When the p-value is greater than or equal to the preset significance level, it is determined that there is no spatial autocorrelation and subsequent analysis is terminated.

5. The method for analyzing usage data of a rental vehicle according to claim 1, characterized in that: The parameterized modeling of the spatial autocorrelation in the historical vehicle usage data based on the spatial dependency network and the establishment of a spatial autocorrelation correction usage model include: Based on the structural characteristics of the spatial dependency network, a corresponding spatial econometric model type is selected; According to the selected spatial econometric model type, the usage indicators in the historical vehicle usage dataset are used as dependent variables, the external factors affecting vehicle usage are used as independent variables, and the spatial weights of the spatial dependency network are combined to construct the corresponding model equations; Applying the maximum likelihood estimation method to solve the model equations to obtain the spatial autocorrelation coefficient, regression coefficient and standard error, and evaluating the statistical significance of the spatial autocorrelation coefficient through a likelihood ratio test; The spatial autocorrelation coefficient, the regression coefficient and the standard error are substituted into the model equation group to construct a spatial autocorrelation correction model.

6. The method for analyzing usage data of a rental vehicle according to claim 5, characterized in that: The applying the maximum likelihood estimation method to solve the model equations to obtain the spatial autocorrelation coefficient, regression coefficient and standard error, and evaluating the statistical significance of the spatial autocorrelation coefficient by a likelihood ratio test includes: Constructing a log-likelihood function corresponding to the model equations, and applying a numerical optimization algorithm to the log-likelihood function to iteratively solve for the optimal estimate of the spatial autocorrelation coefficient, the regression coefficient vector, and the error variance, while obtaining the corresponding maximum log-likelihood value; Obtaining a spatial autocorrelation coefficient, a regression coefficient vector, and a standard error of the error variance based on a second-order derivative matrix of the log-likelihood function at the optimal estimate; Constructing a log-likelihood function of a constraint model that does not include spatial autocorrelation terms, solving parameter estimates of the log-likelihood function of the constraint model using a numerical optimization algorithm, and obtaining a maximum log-likelihood value of the constraint model; A likelihood ratio statistic is calculated according to the maximum log-likelihood value of the model equation group and the maximum log-likelihood value of the constraint model, and whether the spatial autocorrelation coefficient is statistically significant is determined based on the likelihood ratio statistic.

7. The method for analyzing usage data of a rental vehicle according to claim 1, characterized in that: The method of predicting vehicle demand at rental points based on the spatial autocorrelation correction usage model and generating a vehicle allocation strategy that considers the mutual influence between rental points based on the vehicle demand includes: Inputting the usage status data and external environmental factor data of the current rental point into the spatial autocorrelation correction usage model to generate a vehicle demand forecast value considering spatial correlation; Identifying rental points with imbalanced vehicle supply and demand based on the vehicle demand forecast values, and calculating the vehicle allocation demand between the rental points in combination with the spatial dependency network; Based on the vehicle allocation demand, a multi-objective optimization problem is constructed with the goals of minimizing the total allocation cost and maximizing the demand satisfaction rate, with the distance cost between rental points and the influence intensity in the spatial dependency network as constraints; A heuristic algorithm is used to solve the multi-objective optimization problem and generate a vehicle deployment strategy including deployment path, deployment quantity and priority.

8. A rental vehicle usage data analysis system, characterized in that: The rental vehicle usage data analysis system includes: A data processing module is used to perform spatiotemporal labeling on historical vehicle usage data distributed across different rental locations, construct a spatiotemporal data matrix for vehicle usage, and calculate a spatial weight matrix between rental locations. Based on the spatiotemporal data matrix and the spatial weight matrix, a historical vehicle usage dataset is generated. a pattern recognition module for identifying spatial clustering patterns and spatial heterogeneity characteristics in the historical vehicle usage data based on the historical vehicle usage dataset, and constructing a spatial dependency network of rental vehicle usage behavior using the spatial clustering patterns and the spatial heterogeneity characteristics; a model building module for parameterizing spatial autocorrelation in historical vehicle usage data based on the spatial dependency network and establishing a spatial autocorrelation correction usage model; A deployment strategy module is used to predict the vehicle demand of the rental points based on the spatial autocorrelation correction usage model, and generate a vehicle deployment strategy that takes into account the mutual influence between the rental points based on the vehicle demand.

Images