A big data dynamic path intelligent planning system and method for instant delivery

Abstract

The application discloses a kind of big data dynamic path intelligent planning system and method for instant distribution, its system includes multi-source real-time sensing module, big data fusion processing module, distribution demand prediction module, multi-objective dynamic optimization scheduling module, path reconstruction and adaptive updating module and feedback learning module.The application realizes the quick identification and path dynamic reconstruction to traffic congestion, order change and emergent event by constructing multi-source real-time sensing and feature deviation trigger mechanism, avoid distribution personnel to follow invalid path, simultaneously, to traffic, meteorology, road risk and driver load multi-source heterogeneous data are unified feature modeling and weight adaptive fusion, and combined with local path repair strategy, only abnormal section is re-planned, significantly reduce computational complexity and improve system stability in large-scale concurrent scene.

Description

Technical Field

[0001] This invention relates to the field of intelligent route planning technology, and in particular to a big data dynamic intelligent route planning system and method for on-demand delivery. Background Technology

[0002] With the rapid development of e-commerce and local life services, on-demand delivery services are characterized by high frequency, small batches, multiple time periods, and strong timeliness. The delivery network nodes are dense and orders change frequently. Traditional static route planning methods can no longer meet the needs of high-efficiency and high-reliability delivery. To this end, a big data dynamic route intelligent planning system for on-demand delivery has been proposed in the existing technology. By analyzing order information, vehicle location information, and historical trajectory data, the system realizes automatic planning and scheduling of delivery routes. However, in practical applications, the aforementioned existing technologies still have the following obvious shortcomings: 1. Insufficient real-time perception and updating capability for dynamic environments: Existing systems typically collect vehicle location and order status information in batches at fixed time intervals, resulting in long route planning update cycles. This makes it difficult to reflect dynamic factors such as traffic congestion, sudden accidents, and temporary changes in orders in a timely manner. When the delivery environment changes abruptly, the system cannot complete route reconstruction in a short time, causing delivery personnel to continue using invalid routes, which in turn leads to delays, detours, or even task failures. 2. The ability to integrate big data and optimize intelligently needs to be improved: Existing technologies mostly use single or a small number of data sources for decision-making, failing to effectively integrate multi-source heterogeneous data such as traffic flow, weather information, road risks, and driver load, and lacking a unified feature modeling mechanism, resulting in an insufficient portrayal of the delivery environment. At the same time, in high-concurrency order scenarios, the ability to process large-scale data is limited, and the path optimization model is prone to problems such as slow response and unstable convergence, affecting the overall performance of the system. 3. Lack of multi-objective scheduling and forecasting capabilities: Existing route planning systems usually use the shortest path or the shortest time as a single optimization objective, without comprehensively considering multi-objective constraints such as delivery costs, driver load balancing, energy consumption, and customer satisfaction. This makes it difficult to balance overall efficiency and service quality in the scheduling results. In addition, most systems lack the ability to forecast future order demand and traffic trends, and can only perform passive scheduling, unable to achieve advance planning and proactive optimization. In summary, existing big data dynamic path intelligent planning systems for on-demand delivery still have significant shortcomings in real-time perception, data fusion and processing capabilities, and multi-objective prediction and scheduling. There is an urgent need for a big data dynamic path intelligent planning system and method for on-demand delivery that can integrate multi-source big data, possess real-time perception and prediction-driven capabilities, and support multi-objective dynamic optimization, so as to improve the overall operational efficiency and service quality in on-demand delivery scenarios. Summary of the Invention

[0003] Based on the technical problems existing in the background technology, this invention proposes a big data dynamic path intelligent planning system and method for on-demand delivery, which solves the problems of insufficient real-time perception and updating capabilities of dynamic environments, the need to improve big data fusion and intelligent optimization capabilities, and the lack of multi-objective scheduling and prediction capabilities in the existing technology.

[0004] The present invention proposes a big data dynamic path intelligent planning system for on-demand delivery, which includes a multi-source real-time perception module, a big data fusion processing module, a delivery demand prediction module, a multi-objective dynamic optimization scheduling module, a path reconstruction and adaptive update module, and a feedback learning module. The multi-source real-time sensing module is used to collect delivery vehicle status data, order status data, road traffic status data, and external environment data, and to construct a delivery environment status vector. ; The big data fusion processing module is used to process the delivery environment state vector. Feature mapping and weighted fusion are performed to generate a unified feature vector. ; The delivery demand forecasting module is used to predict the distribution of future delivery demand based on historical order time series data. The multi-objective dynamic optimization scheduling module is used to base on a unified feature vector. Construct a multi-objective joint optimization function with the predicted delivery demand, and output the optimal set of delivery routes; The path reconstruction and adaptive update module is used to trigger path reconstruction when the offset between the current unified feature vector and the unified feature vector at the previous moment is greater than a preset threshold. The feedback learning module is used to update the scheduling model parameters based on historical scheduling execution results.

[0005] Preferably, the delivery environment state vector Represented as: ; in This represents the vehicle's position and velocity vector. Indicates the state of road congestion. Indicates the order queue status. Indicates weather conditions. Indicates the driver's load status. Indicates road risk factors; The unified feature vector The formula for construction is: ; in Let represent the weight of the i-th feature, and n represent the total number of feature dimensions in the delivery environment state vector. Represents the i-th normalized feature. This represents the input data for the i-th original feature.

[0006] Preferably, the delivery demand forecasting module uses a time-series forecasting model to predict the future distribution of delivery demand, and the forecasting formula is as follows: ; in Indicates the future Order demand forecast at any given time. ( () represents the time series prediction model function, and k represents the prediction step size. Both represent order sequences; The multi-objective joint optimization function constructed by the multi-objective dynamic optimization scheduling module is: ; Where J represents the multi-objective joint optimization function, and D represents the total delivery distance. Indicates delivery delay. Indicates scheduling cost, This indicates a loss of customer satisfaction. , , and Each represents the weight coefficient of each objective.

[0007] Preferably, the triggering condition for the path reconstruction and adaptive update module is: ; in Indicates the state offset threshold. This represents the unified feature vector generated in the previous time step. Represents a unified feature vector. The L2 norm operation of vectors is used to quantify the Euclidean distance between two eigenvectors, representing the magnitude of changes in the environmental state. The parameter update method of the feedback learning module is as follows: ; in This represents the updated set of scheduling model parameters at time t+1, including but not limited to the weights of the time series prediction model, the coefficients of the multi-objective optimization function, and the core model parameters. Indicates the current parameters of the model. Indicates the learning rate. This represents the gradient of the loss function.

[0008] This invention also proposes a big data-driven dynamic route intelligent planning method for on-demand delivery, comprising the following steps: S1: Constructing a delivery environment state vector: Through vehicle terminals, driver mobile terminals, and back-end business systems, real-time data on vehicle location and speed, order change information, road congestion index, weather information, and driver load status are collected and uniformly encapsulated into a delivery environment state vector. ; ; in This represents the vehicle's position and velocity vector. Indicates the state of road congestion. Indicates the order queue status. Indicates weather conditions. Indicates the driver's load status. Indicates road risk factors; S2: Generate a unified feature vector: Use the feature mapping function to analyze the delivery environment state vector. Dimension alignment and normalization are performed, and weight coefficients are assigned according to the importance of different data sources to form a unified feature vector. ; S3: Predict future order demand; use a time-series forecasting model to model historical order volume sequences to obtain predicted order demand values ​​for multiple future time periods. This provides a forward-looking basis for path planning; S4: Perform multi-objective path joint optimization: Using a unified feature vector and order demand prediction results as input, construct a multi-objective optimization function that includes delivery distance, time delay, cost, and customer satisfaction, and output the current set of optimal delivery paths. ; S5: Determine whether to trigger path reconstruction: Calculate the offset between the current unified feature vector and the unified feature vector at the previous time step. When it is greater than the preset threshold, the path reconstruction process is automatically triggered to avoid path failure. S6: Update model parameters based on feedback results: Collect feedback data on delays, anomalies, and success rates generated during the delivery process, and adjust model parameters and weight coefficients through gradient updates to achieve adaptive evolution of the system.

[0009] Preferably, the specific logical steps of S2 are as follows: S201: Transfer the delivery environment state vector in S1 Each piece of data in the data is stamped with a uniform timestamp 't' and placed within a time window. The built-in dataset is aligned to ensure that data from different sources can be processed synchronously, using the following formula: ; in The data set representing time t, This indicates vehicle position and speed data. This indicates road congestion status data. This represents order queue data. Represents meteorological status data. This indicates driver load status data. This represents road risk factor data, where Δt represents the length of the data alignment time window, which can be adjusted according to the actual scenario. S202: Perform anomaly detection and removal on the aligned data in S201. Outliers are filtered out by calculating the standard deviation and a threshold to ensure the reliability of the fused data. The formula used for this calculation is: ,when Remove in time; in This represents the i-th data item. Representing data Historical average, Representing data standard deviation Indicates standardization bias. This indicates the threshold for identifying outliers; S203: Use interpolation to complete missing data to avoid incomplete feature vectors due to missing data. The formula used is: ; in This represents the i-th data item at time t. , They represent time respectively and Data; S204: Perform feature mapping and normalization on each type of data to transform data of different dimensions into a unified feature space. The formula used is as follows: ; in Represents the normalized eigenvalues. These represent the maximum and minimum values ​​of the i-th data type, respectively. S205: The weights are dynamically adjusted based on the stability of each data source in its historical operation, so that data with high importance and stability have a greater impact on the fusion result. The formula used is: ; in This represents the weight of the i-th data type at time t. This represents the historical variance of the i-th class of data. This represents the weight decay coefficient. Indicates the total number of data sources. Let represent the historical variance of the j-th data source, where j represents the index variable of the data source, and its value range is the same as that of i. ; S206: Sum the weighted features to generate a unified feature vector. As input for subsequent prediction and optimization, the formula used is: ; in Represents a unified feature vector. This represents the weight of the i-th feature. Let i represent the i-th normalized feature.

[0010] Preferably, the specific logical steps of S3 are as follows: S301: Collect historical order data to construct a time series This provides a basis for predicting future order volumes, and its formula is: ; in This represents the historical order sequence at time t. Indicates time The order volume, where m represents the length of the historical time window; S302: The historical series is stationary by eliminating trends through differencing. The formula used is: ; in This represents the stationary order sequence. This represents the order sequence from the previous time step; S303: Input the stabilized sequence into the time series prediction model ( ), to obtain the order demand forecast for future periods. It provides forward-looking information for path optimization, and the formula used is: ; in Indicates the future Order demand forecast at any given time. ( ) represents the time series prediction model function, and k represents the prediction step size.

[0011] Preferably, the specific logical steps of S4 are as follows: S401: A multi-objective optimization function is constructed based on a unified feature vector and predicted order demand. This function comprehensively considers delivery distance, delivery delay, scheduling costs, and customer satisfaction. The multi-objective optimization function is as follows: ; Where J represents the multi-objective joint optimization function, and D represents the total delivery distance. Indicates delivery delay. Indicates scheduling cost, This indicates a loss of customer satisfaction. , , and Each represents a weighting coefficient for each objective. + + , This represents the weighting coefficient of the total delivery distance D. This represents the weighting coefficient for the delivery delay T. This represents the weighting coefficient of the scheduling cost C. The weighting coefficient representing the customer satisfaction loss E; S402: Generate a set of candidate delivery routes p in the road network to ensure that all orders are deliverable within the time window and vehicle constraints; ; in Each represents a specific candidate path in set P. It is the first candidate path. This is the second one, and so on, where j represents the total number of candidate paths; S403: Evaluate and rank candidate paths by calculating a multi-objective cost function, and select the optimal set of paths. For actual scheduling, the formula used is: ; in Representing a path The corresponding multi-objective cost function value, This represents the variable corresponding to the minimum value. This represents a specific path in the set of candidate delivery paths P.

[0012] Preferably, the specific logical steps of S5 are as follows: S501: Real-time calculation of the offset between the current unified feature vector and the feature vector at the previous time step. The formula used to quantify the degree of environmental change is: ; in Indicates the uniform feature offset. These represent the unified feature vectors of the current time step and the previous time step, respectively. Represents the L2 norm operation of vectors; S502: When offset or Path reconstruction is triggered in real time to ensure the effectiveness of path planning. Indicates the state offset threshold. Indicates the actual delay time. Indicates the maximum allowed delay threshold; S503: Under triggered conditions, perform local route repair on abnormal or congested road sections and regenerate the optimal route. This avoids redundant calculations of the global path and improves system efficiency.

[0013] Preferably, the specific logical steps of S6 are as follows: S601: Collect actual completion time during delivery execution. Actual delivery cost and customer satisfaction Feedback data, used to evaluate scheduling effectiveness, is expressed by the following formula: ; Where R represents the set of feedback data for a single delivery task; S602: Construct a loss function to quantify execution deviations and reflect the gap between the planned path and the actual execution results. The formula used is: ; Where L represents the loss function. Indicates the planned time. Indicates planned cost, , Both represent loss weights; S603: Update model parameters based on the gradient of the loss function By incorporating multi-objective weights, the system becomes more accurate and intelligent in the next round of scheduling, achieving closed-loop adaptive optimization. The formula used is: ; in This represents the updated set of scheduling model parameters at time t+1. Indicates the current parameters of the model. Indicates the learning rate. This represents the gradient of the loss function.

[0014] Compared with existing technologies, the beneficial effects of this invention are: 1. By constructing a multi-source real-time perception mechanism, continuous data collection is performed on vehicle location, order status, traffic conditions, and emergency events. Environmental changes are quantified based on feature offsets. Exceeding the threshold The system automatically triggers dynamic route reconstruction, enabling it to quickly update delivery routes after traffic congestion, order changes, or unexpected accidents, preventing delivery personnel from using invalid routes, thereby significantly reducing delay and detour rates and improving the reliability and real-time response capability of delivery tasks. 2. By performing unified feature modeling on multi-source heterogeneous data and constructing environmental feature vectors using a weighted adaptive fusion method, the system achieves deep integration of information such as traffic, weather, road risk, and driver load, making the system's characterization of the delivery environment more comprehensive and accurate. At the same time, through a local path repair mechanism, only abnormal road segments are replanned, which significantly reduces the computational complexity in large-scale concurrent scenarios and significantly improves the stability and real-time performance of the path planning model under high load conditions. 3. By constructing a multi-objective joint optimization cost function, factors such as delivery distance, timeliness, cost, and energy consumption are incorporated into a unified optimization framework to achieve multi-objective collaborative scheduling and avoid the overall efficiency imbalance caused by focusing solely on the shortest path or shortest time. At the same time, an order time series prediction model is introduced to predict future order demand in advance, enabling the system to adjust vehicle layout and route strategies in advance, transforming from passive scheduling to proactive planning, thereby effectively improving order completion rate and customer satisfaction during peak hours. This invention constructs a multi-source real-time perception and feature offset triggering mechanism to achieve rapid identification and dynamic path reconstruction of traffic congestion, order changes, and emergencies, preventing delivery personnel from following failed paths. Simultaneously, it performs unified feature modeling and adaptive weight fusion of multi-source heterogeneous data on traffic, weather, road risks, and driver load, and combines this with a local path repair strategy to replan only abnormal road segments, significantly reducing computational complexity and improving system stability in large-scale concurrent scenarios. Furthermore, by establishing a multi-objective joint optimization cost function and introducing an order time series prediction model, it achieves multi-objective collaborative scheduling of delivery distance, timeliness, cost, and energy consumption, transforming the system from passive response to proactive planning, thereby comprehensively improving the real-time performance, overall efficiency, and customer satisfaction of the instant delivery system. Attached Figure Description

[0015] Figure 1 This is a block diagram of a big data dynamic route intelligent planning system for on-demand delivery proposed in this invention. Figure 2 This is a flowchart of a big data dynamic path intelligent planning method for on-demand delivery proposed in this invention. Figure 3This is a flowchart of multi-objective dynamic optimization scheduling and optimal path selection in a big data dynamic path intelligent planning method for on-demand delivery proposed in this invention. Figure 4 This is a curve comparing the accuracy of delivery demand prediction in a big data dynamic route intelligent planning method for on-demand delivery proposed in this invention. Figure 5 This is a bar chart comparing the response time of path reconstruction in a big data dynamic path intelligent planning method for on-demand delivery proposed in this invention. Detailed Implementation

[0016] The present invention will be further explained below with reference to specific embodiments.

[0017] Example 1 Reference Figure 1 This embodiment proposes a big data dynamic path intelligent planning system for on-demand delivery, including a multi-source real-time perception module, a big data fusion processing module, a delivery demand prediction module, a multi-objective dynamic optimization scheduling module, a path reconstruction and adaptive update module, and a feedback learning module. The multi-source real-time sensing module is used to collect delivery vehicle status data, order status data, road traffic status data, and external environment data, and to construct a delivery environment status vector. ; The delivery environment state vector Represented as: ; in This represents the vehicle's position and velocity vector. Indicates the state of road congestion. Indicates the order queue status. Indicates weather conditions. Indicates the driver's load status. Indicates road risk factors; The big data fusion processing module is used to analyze the state vector of the delivery environment. Feature mapping and weighted fusion are performed to generate a unified feature vector. ; Among them, the unified feature vector The formula for construction is: ; in Let represent the weight of the i-th feature, and n represent the total number of feature dimensions in the delivery environment state vector. Represents the i-th normalized feature. This represents the input data for the i-th original feature; The delivery demand forecasting module is used to predict the distribution of future delivery demand based on historical order time-series data; The delivery demand forecasting module uses a time-series forecasting model to predict the distribution of future delivery demand. The forecasting formula is as follows: ; in Indicates the future Order demand forecast at any given time. ( () represents the time series prediction model function, and k represents the prediction step size. Both represent order sequences; The multi-objective dynamic optimization scheduling module is used for scheduling based on a unified feature vector. Construct a multi-objective joint optimization function with the predicted delivery demand, and output the optimal set of delivery routes; The multi-objective joint optimization function constructed by the multi-objective dynamic optimization scheduling module is: ; Where J represents the multi-objective joint optimization function, and D represents the total delivery distance. Indicates delivery delay. Indicates scheduling cost, This indicates a loss of customer satisfaction. , , and Each represents a weight coefficient for each objective. The path reconstruction and adaptive update module is used to trigger path reconstruction when the offset between the current unified feature vector and the unified feature vector at the previous time step is greater than a preset threshold. The triggering conditions for the path reconstruction and adaptive update module are: ; in Indicates the state offset threshold. This represents the unified feature vector generated in the previous time step. Represents a unified feature vector. The L2 norm operation of vectors is used to quantify the Euclidean distance between two eigenvectors, representing the magnitude of changes in the environmental state. The feedback learning module is used to update the scheduling model parameters based on historical scheduling execution results; The parameter update method for the feedback learning module is as follows: ; in This represents the updated set of scheduling model parameters at time t+1, including but not limited to the weights of the time series prediction model, the coefficients of the multi-objective optimization function, and the core model parameters. Indicates the current parameters of the model. Indicates the learning rate. This represents the gradient of the loss function.

[0018] Example 2 Based on the same inventive concept as the big data dynamic route intelligent planning system for instant delivery in the foregoing embodiments, such as Figures 2-5 This embodiment proposes a big data-driven dynamic route intelligent planning method for on-demand delivery, including the following steps: S1: Constructing a delivery environment state vector: Through vehicle terminals, driver mobile terminals, and back-end business systems, real-time data on vehicle location and speed, order change information, road congestion index, weather information, and driver load status are collected and uniformly encapsulated into a delivery environment state vector. ; ; in This represents the vehicle's position and velocity vector. This indicates the road congestion status, with a quantification value ranging from [0,1], where 0 represents no congestion and 1 represents severe congestion. This indicates the order queue status, which includes the number of orders, delivery address distribution, and delivery time requirements. This indicates the meteorological condition, which includes quantitative indicators such as rainfall, wind speed, and visibility. Indicates the driver's load status. This represents road risk factors, which include risk levels such as construction sections, accident-prone areas, and temporary traffic control. The value range is [0,1]. Construction sections / temporary traffic control are marked with 0.8-1.0, accident-prone areas with 0.5-0.7, and normal road sections with 0-0.2. The data source is real-time announcements from the transportation department. S2: Generate a unified feature vector: Use the feature mapping function to analyze the delivery environment state vector. Dimension alignment and normalization are performed, and weight coefficients are assigned according to the importance of different data sources to form a unified feature vector. ; The specific logical steps are as follows: S201: Transfer the delivery environment state vector in S1 Each piece of data in the data is stamped with a uniform timestamp 't' and placed within a time window. The built-in dataset is aligned to ensure that data from different sources can be processed synchronously, using the following formula: ; in The data set representing time t, This indicates vehicle position and speed data. This indicates road congestion status data. This represents order queue data. Represents meteorological status data. This indicates driver load status data. This represents road risk factor data, where Δt represents the length of the data alignment time window, which can be adjusted according to the actual scenario. S202: Perform anomaly detection and removal on the aligned data in S201. Outliers are filtered out by calculating the standard deviation and a threshold to ensure the reliability of the fused data. The formula used for this calculation is: ,when Remove in time; in This represents the i-th data item. Representing data Historical average, Representing data standard deviation Indicates standardization bias. This indicates the threshold for identifying outliers; S203: Use interpolation to complete missing data to avoid incomplete feature vectors due to missing data. The formula used is: ; in This represents the i-th data item at time t. , They represent time respectively and Data; S204: Perform feature mapping and normalization on each type of data to transform data of different dimensions into a unified feature space. The formula used is as follows: ; in Represents the normalized eigenvalues. Let represent the maximum and minimum values ​​of the i-th type of data, respectively; this formula is a min-max normalization, the core function of which is to eliminate the influence of data with different dimensions (such as vehicle speed km / h and rainfall mm), so that all feature values ​​are on the same order of magnitude, facilitating subsequent weighted calculations; if the congestion index of a certain road segment... historical maximum value =1, minimum value =0, current =0.6, then = =0.6; S205: The weights are dynamically adjusted based on the stability of each data source in its historical operation, so that data with high importance and stability have a greater impact on the fusion result. The formula used is: ; in This represents the weight of the i-th data type at time t. This represents the historical variance of the i-th data category. The smaller the variance, the more stable the data, and the larger the weight. This represents the weight decay coefficient. Indicates the total number of data sources. Let represent the historical variance of the j-th data source, where j represents the index variable of the data source, and its value range is the same as that of i. In addition, when a certain type of data is core data (such as order data), a base weight of 0.1 will be added to the variance calculation to ensure the influence of the core data; S206: Sum the weighted features to generate a unified feature vector. As input for subsequent prediction and optimization, the formula used is: ; in Represents a unified feature vector. This represents the weight of the i-th feature. Represents the i-th normalized feature; In addition, step S2 uses a series of key processing steps, such as time alignment, anomaly removal, missing data completion, normalization, and dynamic weighting, to transform multi-source heterogeneous delivery environment data into a feature vector Ft with a unified dimension and scale by using feature mapping functions and dynamic weight allocation algorithms. This not only eliminates the scale differences and noise interference from different data sources, but also highlights the influence of core data through dynamic weight adjustment. This provides high-quality and highly available input features for subsequent order demand prediction and multi-objective path optimization, effectively improving the computational efficiency and accuracy of subsequent algorithms. S3: Predict future order demand; use a time-series forecasting model to model historical order volume sequences to obtain predicted order demand values ​​for multiple future time periods. This provides a forward-looking basis for path planning; The specific logical steps are as follows: S301: Collect historical order data to construct a time series This provides a basis for predicting future order volumes, and its formula is: ; in This represents the historical order sequence at time t. Indicates time The order volume, where m represents the length of the historical time window; S302: The historical series is stationary by eliminating trends through differencing. The formula used is: ; in This represents the stationary order sequence. This represents the order sequence from the previous time step; If the stationarized sequence still does not meet the stationarity test, repeat the difference operation until it does. S303: Input the stabilized sequence into the time series prediction model ( ), to obtain the order demand forecast for future periods. It provides forward-looking information for path optimization, and the formula used is: ; in Indicates the future Order demand forecast at any given time. ( ) represents the time series prediction model function, and k represents the prediction step size; In addition, step S3 is based on historical order time series. It eliminates data trend and periodic interference through stabilization processing, and then uses a time series prediction model to model the processed sequence, accurately outputting the order demand prediction values ​​for multiple future time periods. This solves the "foresight" problem of route planning, allowing the scheduling system to predict high-incidence areas and time periods of orders in advance, providing data support for advance vehicle deployment and route pre-planning, and avoiding problems such as vehicle scheduling delays and delivery timeouts caused by sudden orders. S4: Perform multi-objective path joint optimization: Using a unified feature vector and order demand prediction results as input, construct a multi-objective optimization function that includes delivery distance, time delay, cost, and customer satisfaction, and output the current set of optimal delivery paths. ; The specific logical steps are as follows: S401: A multi-objective optimization function is constructed based on a unified feature vector and predicted order demand. This function comprehensively considers delivery distance, delivery delay, scheduling costs, and customer satisfaction. The multi-objective optimization function is as follows: ; Where J represents the multi-objective joint optimization function, and D represents the total delivery distance. Indicates delivery delay. Indicates scheduling cost, E represents customer satisfaction loss, with a value ranging from [0,1]. It is calculated based on factors such as timeout duration and damage rate. The longer the timeout and the higher the damage rate, the closer E is to 1. , , and Each represents a weighting coefficient for each objective. + + , This represents the weighting coefficient of the total delivery distance D. This represents the weighting coefficient for the delivery delay T. This represents the weighting coefficient of the scheduling cost C. This represents the weighting coefficient for customer satisfaction loss E, set during peak hours (e.g., 11:00-13:00, 17:00-19:00). =0.4、 =0.3、 =0.15、 =0.15; Off-peak hours are set =0.4、 =0.3、 =0.15、 =0.15; S402: Generate a set of candidate delivery routes p in the road network to ensure that all orders are deliverable within the time window and vehicle constraints; ; in Each represents a specific candidate path in set P. It is the first candidate path. This is the second one, and so on, where j represents the total number of candidate paths; The constraints include: maximum vehicle load capacity, maximum driving range, and order delivery time window (such as delivery between 10:00 and 10:30 as specified by the customer). S403: Evaluate and rank candidate paths by calculating a multi-objective cost function, and select the optimal set of paths. For actual scheduling, the formula used is: ; in Representing a path The corresponding multi-objective cost function value, This represents the variable corresponding to the minimum value. This represents a specific path in the candidate delivery path set P; In addition, step S4 takes the unified feature vector and order demand prediction results as input to construct a multi-objective optimization function covering delivery distance, time delay, cost and customer satisfaction. Combined with the weight differentiation strategy for peak and off-peak periods, the function outputs the set of paths with the best overall performance through candidate path generation and cost evaluation. The core is to achieve a dynamic balance between delivery efficiency, operating cost and service quality under multiple constraints such as vehicle load, range, and order time window, so as to meet the cost control needs of enterprises and ensure the delivery experience of customers. S5: Determine whether to trigger path reconstruction: Calculate the offset between the current unified feature vector and the unified feature vector at the previous time step. When it is greater than the preset threshold, the path reconstruction process is automatically triggered to avoid path failure. The specific logical steps are as follows: S501: Real-time calculation of the offset between the current unified feature vector and the feature vector at the previous time step. The formula used to quantify the degree of environmental change is: ; in Indicates the uniform feature offset. These represent the unified feature vectors of the current time step and the previous time step, respectively. Represents the L2 norm operation of vectors; S502: When offset or Path reconstruction is triggered in real time to ensure the effectiveness of path planning. Indicates the state offset threshold. Indicates the actual delay time. Indicates the maximum allowed delay threshold; S503: Under triggered conditions, perform local route repair on abnormal or congested road sections and regenerate the optimal route. This avoids redundant calculations of global paths and improves system efficiency. The specific steps for local route repair of abnormal or congested road sections are as follows: S5031: Represent the current execution path as a sequence of nodes: And calculate the real-time travel cost offset for each edge segment: ,when When, mark the edge. This is an abnormal road section; in This represents the real-time toll cost offset rate for the i-th road segment, used to quantify the degree of anomaly in the toll consumption of that road segment. This represents the real-time actual travel time for the i-th road segment. This represents the historical average travel time for the i-th road segment. This indicates the threshold for identifying abnormal road sections. This represents the i-th node in the path node sequence. Let i represent the i-th edge. Indicates road segment The starting node, Indicates road segment The endpoint This represents the complete delivery route currently being executed, consisting of k nodes in sequence; S5032: Using the abnormal road segment as the center, expand r nodes forward and backward to construct the sub-path to be repaired: ; in This indicates the subpath to be repaired. Indicates the starting node of the sub-path. Indicates the start and end points of the abnormal road segment. This indicates the end node of the subpath, represented by 'r'. S5033: From the original global road network map Extracting subgraphs: And update the subgraph edge weights in real time: ; in Let represent the set of nodes in the subgraph, and V represent the set of all nodes in the global road network. Represents the set of all road segments in the global road network. Represents the set of edges of a subgraph. This represents the total travel cost of the j-th edge in the subgraph. , , and These correspond to the weights of distance, time, cost, and customer satisfaction, respectively. This represents the physical distance of the j-th edge. This represents the real-time travel time of the j-th edge. Let $j$ represent the travel cost of the $j$-th edge. This represents the customer satisfaction loss along the j-th edge; S5034: In the subgraph Starting from above end Running the constrained Dijkstra algorithm: ; in Indicates the starting offset, usually , Indicates in the subgraph Find the comprehensive toll cost The shortest path This represents the optimized new sub-path; S5035: Remove the original path Replace with the new subpath: ; in Indicates the first to the second position in the original path. A sequence of nodes, Indicates the first in the original path A sequence up to k nodes, This indicates a set merging operation, which concatenates three path segments into a complete path. S5036: If Then update the current delivery route to And synchronized to the delivery terminal; in Indicates the optimal path after repair. The overall cost, This represents the total cost of the original path P before repair; In addition, step S5 constructs a dual triggering mechanism by calculating the feature vector offset and order delay time to accurately identify scenarios such as sudden changes in road conditions and order anomalies that cause path failure. The core adopts the "local path repair" strategy, which quickly generates the optimal path without recalculating the global path by marking abnormal road segments, extracting sub-paths, optimizing subgraphs, and stitching paths. This ensures the real-time effectiveness of the path, significantly reduces computational complexity, and improves the system's adaptability and operational stability in dynamic environments. S6: Update model parameters based on feedback results: Collect feedback data on delays, anomalies, and success rates generated during delivery execution, and adjust model parameters and weight coefficients through gradient updates to achieve adaptive evolution of the system. The specific logical steps are as follows: S601: Collect actual completion time during delivery execution. Actual delivery cost and customer satisfaction Feedback data, used to evaluate scheduling effectiveness, is expressed by the following formula: ; Where R represents the set of feedback data for a single delivery task; S602: Construct a loss function to quantify execution deviations and reflect the gap between the planned path and the actual execution results. The formula used is: ; Where L represents the loss function. Indicates the planned time. Indicates planned cost, , Both represent loss weights; S603: Update model parameters based on the gradient of the loss function By incorporating multi-objective weights, the system becomes more accurate and intelligent in the next round of scheduling, achieving closed-loop adaptive optimization. The formula used is: ; in This represents the updated set of scheduling model parameters at time t+1. Indicates the current parameters of the model. Indicates the learning rate. Represents the gradient of the loss function; In addition, step S6 collects feedback data such as actual delays, costs, and customer satisfaction during the delivery process, constructs a loss function to quantify the deviation between the plan and the actual situation, and then uses the gradient descent algorithm to update the prediction model parameters and multi-objective optimization weights in reverse, forming a closed-loop adaptive system of "data collection - path planning - execution feedback - parameter optimization". The core is to solve the problem of "accuracy decay" in the long-term operation of the system, so that the system can continuously learn the changing patterns of the actual delivery scenario and continuously improve the accuracy and intelligence of scheduling decisions. This embodiment achieves rapid dynamic reconstruction of delivery routes through multi-source real-time perception and feature offset triggering mechanisms. Combined with multi-source data unified modeling and local path repair strategies, it reduces computational complexity and improves system stability in high-concurrency scenarios. At the same time, it introduces multi-objective joint optimization and order demand prediction models to realize the transformation from passive scheduling to proactive planning, thereby comprehensively improving the real-time performance, overall efficiency and service quality of instant delivery.

[0019] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A big data-driven dynamic route intelligent planning system for on-demand delivery, characterized in that: It includes a multi-source real-time perception module, a big data fusion and processing module, a delivery demand prediction module, a multi-objective dynamic optimization and scheduling module, a path reconstruction and adaptive update module, and a feedback learning module; The multi-source real-time sensing module is used to collect delivery vehicle status data, order status data, road traffic status data, and external environment data, and to construct a delivery environment status vector. ; The big data fusion processing module is used to process the delivery environment state vector. Feature mapping and weighted fusion are performed to generate a unified feature vector. ; The delivery demand forecasting module is used to predict the distribution of future delivery demand based on historical order time series data. The multi-objective dynamic optimization scheduling module is used to base on a unified feature vector. Construct a multi-objective joint optimization function with the predicted delivery demand, and output the optimal set of delivery routes; The path reconstruction and adaptive update module is used to trigger path reconstruction when the offset between the current unified feature vector and the unified feature vector at the previous moment is greater than a preset threshold. The feedback learning module is used to update the scheduling model parameters based on historical scheduling execution results.

2. The big data dynamic route intelligent planning system for on-demand delivery according to claim 1, characterized in that, The delivery environment state vector Represented as: ； in Represents the vehicle's position and velocity vector. Indicates the state of road congestion. Indicates the order queue status. Indicates weather conditions. Indicates the driver's load status. Indicates road risk factors; The unified feature vector The formula for construction is: ； in Let represent the weight of the i-th feature, and n represent the total number of feature dimensions in the delivery environment state vector. Represents the i-th normalized feature. This represents the input data for the i-th original feature.

3. The big data dynamic route intelligent planning system for on-demand delivery according to claim 1, characterized in that, The delivery demand forecasting module uses a time-series forecasting model to predict the future distribution of delivery demand. The forecasting formula is as follows: ； in Indicates the future Order demand forecast at any given time. ( () represents the time series prediction model function, and k represents the prediction step size. Both represent order sequences; The multi-objective joint optimization function constructed by the multi-objective dynamic optimization scheduling module is: ； Where J represents the multi-objective joint optimization function, and D represents the total delivery distance. Indicates delivery delay. Indicates scheduling cost, This indicates a loss of customer satisfaction. , , and Each represents the weight coefficient of each objective.

4. The big data dynamic route intelligent planning system for on-demand delivery according to claim 1, characterized in that, The triggering condition for the path reconstruction and adaptive update module is as follows: ； in Indicates the state offset threshold. This represents the unified feature vector generated in the previous time step. Represents a unified feature vector. The L2 norm operation of vectors is used to quantify the Euclidean distance between two eigenvectors, representing the magnitude of changes in the environmental state. The parameter update method of the feedback learning module is as follows: ; in This represents the updated set of scheduling model parameters at time t+1, including but not limited to the weights of the time series prediction model, the coefficients of the multi-objective optimization function, and the core model parameters. Indicates the current parameters of the model. Indicates the learning rate. This represents the gradient of the loss function.

5. A big data-driven dynamic route intelligent planning method for on-demand delivery, applied to the system described in any one of claims 1-4, characterized in that, Includes the following steps: S1: Constructing a delivery environment state vector: Through vehicle terminals, driver mobile terminals, and back-end business systems, real-time data on vehicle location and speed, order change information, road congestion index, weather information, and driver load status are collected and uniformly encapsulated into a delivery environment state vector. ; ； in Represents the vehicle's position and velocity vector. Indicates the state of road congestion. Indicates the order queue status. Indicates weather conditions. Indicates the driver's load status. Indicates road risk factors; S2: Generate a unified feature vector: Use the feature mapping function to analyze the delivery environment state vector. Dimension alignment and normalization are performed, and weight coefficients are assigned according to the importance of different data sources to form a unified feature vector. ; S3: Predict future order demand; use a time-series forecasting model to model historical order volume sequences to obtain predicted order demand values ​​for multiple future time periods. This provides a forward-looking basis for path planning; S4: Perform multi-objective path joint optimization: Using a unified feature vector and order demand prediction results as input, construct a multi-objective optimization function that includes delivery distance, time delay, cost, and customer satisfaction, and output the current set of optimal delivery paths. ; S5: Determine whether path reconstruction is triggered: Calculate the offset between the current unified feature vector and the unified feature vector at the previous time step. When it is greater than the preset threshold, the path reconstruction process is automatically triggered to avoid path failure. S6: Update model parameters based on feedback results: Collect feedback data on delays, anomalies, and success rates generated during the delivery process, and adjust model parameters and weight coefficients through gradient updates to achieve adaptive evolution of the system.

6. The big data dynamic path intelligent planning method for on-demand delivery according to claim 5, characterized in that, The specific logical steps of S2 are as follows: S201: Transfer the delivery environment state vector in S1 Each piece of data in the data is stamped with a uniform timestamp 't' and placed within a time window. The built-in dataset is aligned to ensure that data from different sources can be processed synchronously, using the following formula: ； in The data set representing time t, This indicates vehicle position and speed data. This indicates road congestion status data. This represents order queue data. Represents meteorological status data. This indicates driver load status data. This represents road risk factor data, where Δt represents the length of the data alignment time window, which can be adjusted according to the actual scenario. S202: Perform anomaly detection and removal on the aligned data in S201. Outliers are filtered out by calculating the standard deviation and a threshold to ensure the reliability of the fused data. The formula used for this calculation is: ,when Remove in time; in This represents the i-th data item. Representing data Historical average, Representing data standard deviation Indicates standardization bias. This indicates the threshold for identifying outliers; S203: Use interpolation to complete missing data to avoid incomplete feature vectors due to missing data. The formula used is: ； in This represents the i-th data item at time t. , They represent time respectively and Data; S204: Perform feature mapping and normalization on each type of data to transform data of different dimensions into a unified feature space. The formula used is as follows: ； in Represents the normalized eigenvalues. These represent the maximum and minimum values ​​of the i-th data type, respectively. S205: The weights are dynamically adjusted based on the stability of each data source in its historical operation, so that data with high importance and stability have a greater impact on the fusion result. The formula used is: ； in This represents the weight of the i-th data type at time t. This represents the historical variance of the i-th class of data. This represents the weight decay coefficient. Indicates the total number of data sources. Let represent the historical variance of the j-th data source, where j represents the index variable of the data source, and its value range is the same as that of i. ; S206: Sum the weighted features to generate a unified feature vector. As input for subsequent prediction and optimization, the formula used is: ； in Represents a unified feature vector. This represents the weight of the i-th feature. Let i represent the i-th normalized feature.

7. The big data dynamic path intelligent planning method for on-demand delivery according to claim 6, characterized in that, The specific logical steps of S3 are as follows: S301: Collect historical order data to construct a time series This provides a basis for predicting future order volumes, and its formula is: ； in This represents the historical order sequence at time t. Indicates time The order volume, where m represents the length of the historical time window; S302: The historical series is stationary by eliminating trends through differencing. The formula used is: ； in This represents the stationary order sequence. This represents the order sequence from the previous time step; S303: Input the stabilized sequence into the time series prediction model ( ), to obtain the order demand forecast for future periods. It provides forward-looking information for path optimization, and the formula used is: ； in Indicates the future Order demand forecast at any given time. ( ) represents the time series prediction model function, and k represents the prediction step size.

8. The big data dynamic route intelligent planning method for on-demand delivery according to claim 7, characterized in that, The specific logical steps of S4 are as follows: S401: A multi-objective optimization function is constructed based on a unified feature vector and predicted order demand. This function comprehensively considers delivery distance, delivery delay, scheduling costs, and customer satisfaction. The multi-objective optimization function is as follows: ； Where J represents the multi-objective joint optimization function, and D represents the total delivery distance. Indicates delivery delay. Indicates scheduling cost, This indicates a loss of customer satisfaction. , , and Each represents a weighting coefficient for each objective. + + + , This represents the weighting coefficient of the total delivery distance D. This represents the weighting coefficient for the delivery delay T. This represents the weighting coefficient of the scheduling cost C. The weighting coefficient representing the customer satisfaction loss E; S402: Generate a set of candidate delivery routes p in the road network to ensure that all orders are deliverable within the time window and vehicle constraints; ； in Each represents a specific candidate path in set P. It is the first candidate path. This is the second one, and so on, where j represents the total number of candidate paths; S403: Evaluate and rank candidate paths by calculating a multi-objective cost function, and select the optimal set of paths. For actual scheduling, the formula used is: ； in Representing a path The corresponding multi-objective cost function value, This represents the variable corresponding to the minimum value. This represents a specific path in the set of candidate delivery paths P.

9. The big data dynamic path intelligent planning method for on-demand delivery according to claim 8, characterized in that, The specific logical steps of S5 are as follows: S501: Real-time calculation of the offset between the current unified feature vector and the feature vector at the previous time step. The formula used to quantify the degree of environmental change is: ； in Indicates the uniform feature offset. These represent the unified feature vectors of the current time step and the previous time step, respectively. Represents the L2 norm operation of vectors; S502: When offset or Path reconstruction is triggered in real time to ensure the effectiveness of path planning. Indicates the state offset threshold. Indicates the actual delay time. Indicates the maximum allowed delay threshold; S503: Under triggered conditions, perform local route repair on abnormal or congested road sections and regenerate the optimal route. This avoids redundant calculations of the global path and improves system efficiency.

10. A big data dynamic path intelligent planning method for on-demand delivery according to claim 9, characterized in that, The specific logical steps of S6 are as follows: S601: Collect actual completion time during delivery execution. Actual delivery cost and customer satisfaction Feedback data, used to evaluate scheduling effectiveness, is expressed by the following formula: ； Where R represents the set of feedback data for a single delivery task; S602: Construct a loss function to quantify execution deviations and reflect the gap between the planned path and the actual execution results. The formula used is: ； Where L represents the loss function. Indicates the planned time. Indicates planned cost, , Both represent loss weights; S603: Update model parameters based on the gradient of the loss function By incorporating multi-objective weights, the system becomes more accurate and intelligent in the next round of scheduling, achieving closed-loop adaptive optimization. The formula used is: ; in This represents the updated set of scheduling model parameters at time t+1. Indicates the current parameters of the model. Indicates the learning rate. This represents the gradient of the loss function.

Images