Multi-target logistics vehicle dynamic scheduling system based on deep learning

Abstract

The invention relates to the technical field of logistics vehicle scheduling, and discloses a multi-target logistics vehicle dynamic scheduling system based on deep learning. The system comprises a region division module, an order acquisition module, a path calculation module, a dynamic adjustment module, a central control module and the like. The region division module divides scheduling units and generates task allocation rules; the order acquisition module obtains an order parameter set according to a rule; the path calculation module calculates a path cost index; the dynamic adjustment module obtains a real-time traffic state index; and the central control module fuses the scheduling dimension parameters to generate a comprehensive scheduling scheme. The system can also comprise a route optimization module, a storage module and an alarm module which are respectively used for calculating time correlation degree, storing data and generating an alarm instruction. The method can comprehensively consider multiple targets, process dynamic information in real time, effectively mine spatio-temporal data values, improve logistics distribution efficiency and benefits, reduce cost and improve customer satisfaction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of logistics vehicle scheduling, in particular to a multi-target logistics vehicle dynamic scheduling system based on deep learning. BACKGROUND

[0002] In the current era of vigorous development of the logistics industry, logistics distribution efficiency and cost control have become key factors in enterprise competition. Traditional logistics vehicle scheduling methods have gradually exposed many problems in response to increasingly complex logistics demands.

[0003] Early logistics vehicle scheduling mainly relied on human experience. The scheduling personnel arranged vehicle distribution tasks based on their understanding of basic information such as routes and orders. This approach was still maintainable when the logistics business was small, but as the business volume grew rapidly, its drawbacks became increasingly apparent. Manual scheduling was difficult to fully and timely consider various attributes of numerous orders, such as order weight, volume, delivery time requirements, and delivery address, leading to unreasonable vehicle loading, often resulting in vehicle space waste or overloading, increasing transportation costs, and potentially causing safety hazards. At the same time, manual scheduling had a serious lag in understanding traffic conditions, making it impossible to flexibly adjust routes based on real-time traffic conditions, resulting in extended delivery times and reduced customer satisfaction.

[0004] With the development of computer technology, static scheduling systems based on rules and algorithms emerged. Such systems schedule logistics vehicles by setting some fixed rules and algorithms, such as the nearest neighbor algorithm and the savings algorithm. To some extent, it improves the efficiency and accuracy of scheduling and can handle relatively complex order and route planning problems. However, the limitations of static scheduling systems are also very prominent. It assumes that order information and traffic conditions are static and unchanging, and cannot adapt to dynamic changes in logistics business. In the actual logistics distribution process, new orders are constantly generated, and the delivery requirements of orders may change at any time, and traffic conditions are constantly changing, such as traffic accidents, road construction, and peak congestion. Static scheduling systems are difficult to respond to these changes in real time, leading to deviations in the execution of scheduling plans and failing to achieve optimal delivery results, resulting in waste of logistics resources and low distribution efficiency.

[0005] With the rise of big data and machine learning technologies, some logistics companies have begun to explore the use of these technologies for vehicle scheduling. However, existing scheduling solutions based on big data and machine learning still have many defects. On the one hand, these solutions often only focus on the optimization of a single target, such as simply pursuing the lowest transportation cost or the shortest delivery time, while ignoring other important targets such as maximizing vehicle utilization and minimizing carbon emissions. In actual logistics operations, multiple targets are mutually constrained and influenced, and the optimization of a single target cannot achieve the optimal overall benefit. On the other hand, existing technologies have limited capabilities in processing spatio-temporal data. Logistics distribution involves complex data in both time and space dimensions, and existing algorithms are difficult to effectively mine potential associations and patterns in these data, making it impossible to accurately predict changes in traffic conditions and order demand, thereby affecting the accuracy and real-time performance of the scheduling solution. SUMMARY

[0006] The purpose of the present application is to provide a multi-objective logistics vehicle dynamic scheduling system based on deep learning to solve the problems raised in the background art.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical solution: a multi-objective logistics vehicle dynamic scheduling system based on deep learning, comprising:

[0008] a region division module for dividing a target region into multiple scheduling units according to logistics task requirements and generating corresponding task allocation rules for each scheduling unit;

[0009] an order collection module for obtaining order parameter sets of each scheduling unit from a logistics database in real time according to the task allocation rules and taking all parameters in the order parameter sets as scheduling dimension parameters of the scheduling units;

[0010] a path calculation module for calculating path cost indicators of each scheduling unit based on a pre-set path association relationship in the logistics database through a spatio-temporal convolution algorithm and taking the path cost indicators as scheduling dimension parameters of the scheduling units;

[0011] a dynamic adjustment module for obtaining real-time traffic state indicators of each scheduling unit through a pre-set spatio-temporal association analysis method and taking the real-time traffic state indicators as scheduling dimension parameters of the scheduling units;

[0012] a central control module for sending the scheduling unit information to the path calculation module and the dynamic adjustment module, sending the task allocation rules to the order collection module, and further for data fusion of all scheduling dimension parameters of the scheduling units to generate a comprehensive scheduling solution for the scheduling units.

[0013] Preferably, the method for obtaining the real-time traffic state index of each scheduling unit by the preset space-time correlation analysis method comprises:

[0014] The current scheduling unit and the adjacent scheduling unit are both taken as dynamic analysis nodes, and the traffic fluctuation mode of each dynamic analysis node in a historical period is determined based on the path historical data of the logistics area to which the scheduling unit belongs;

[0015] Based on the traffic fluctuation mode, the initial traffic index of each dynamic analysis node and the deviation of each path historical data from the initial value are calculated based on the fluctuation amplitude of each path historical data in the corresponding dynamic analysis node;

[0016] Based on the initial traffic index of each dynamic analysis node and the deviation of each path historical data from the initial value, the final traffic index of each dynamic analysis node and the deviation of each path historical data are synchronized and optimized through a dynamic topology network to obtain the real-time traffic state index of each dynamic analysis node.

[0017] The comparison dynamic analysis node is set as any one of all dynamic analysis nodes, the comparison path historical data is set as any one of all path historical data, and based on the traffic fluctuation mode, the traffic change amount of the comparison dynamic analysis node in the comparison path historical data is taken as an independent change amount, the total change amount of the comparison dynamic analysis node in all path historical data is taken as an independent total variable, the change amount of all dynamic analysis nodes in the comparison path historical data is taken as a global change amount, and the total change amount of all dynamic analysis nodes in all path historical data is taken as a global total variable. If the ratio of the independent change amount to the independent total variable exceeds the ratio of the global change amount to the global total variable, it is determined that the comparison dynamic analysis node has a significant congestion feature in the comparison path historical data.

[0018] Preferably, the task allocation rule comprises a fixed parameter type group and a variable parameter type group; the fixed parameter type group comprises a plurality of fixed parameter types and a static weight corresponding to each fixed parameter type; and the variable parameter type group comprises a plurality of variable parameter types and an elastic weight corresponding to each variable parameter type.

[0019] Preferably, the method for obtaining the order parameter set of each scheduling unit from the logistics database in real time according to the task allocation rule comprises:

[0020] The order collection module extracts structured order data of the scheduling unit from a logistics database, extracts corresponding fixed parameters from the structured order data respectively based on each fixed parameter type in the fixed parameter type group, and extracts corresponding variable parameters from the structured order data respectively based on each variable parameter type in the variable parameter type group, and aggregates all the fixed parameters and all the variable parameters into an order parameter set of the scheduling unit.

[0021] Preferably, when the preset path association relationship is a direct path relationship, it is judged whether the scheduling unit is associated to a path node in the logistics database, if yes, the path cost indicator of the scheduling unit is the sum of cost contribution values of all path nodes associated with the scheduling unit, otherwise the path cost indicator of the scheduling unit is the basic path cost of the scheduling unit itself.

[0022] When the preset path association relationship is an indirect path relationship, the path cost indicator of the scheduling unit is a dynamic weighted value of cost contribution values of all path nodes obtained by the scheduling unit through a cross-regional associated path.

[0023] Preferably, the system further comprises a route optimization module connected with the central control module.

[0024] The route optimization module is configured to calculate a time association degree of the scheduling unit and a feasible route by a preset space-time trajectory analysis method according to a real-time road network state, and take the time association degree as a scheduling dimension parameter of the scheduling unit.

[0025] Preferably, the route optimization module calculates the time association degree of the scheduling unit and the feasible route by the preset space-time trajectory analysis method, including:

[0026] obtaining a historical optimal route set of the scheduling unit from a path planning model library, extracting a current feasible route set from a real-time road network database, and taking a space-time coincidence degree of the historical optimal route set and the feasible route set as a first association degree of the scheduling unit and the feasible route;

[0027] dynamically matching a delivery time limit requirement of the scheduling unit within a preset time window with real-time time limit data of the feasible route, and taking a time limit matching degree of the two as a second association degree of the scheduling unit and the feasible route;

[0028] taking the first association degree, the second association degree, or a weighted result of the two as the time association degree.

[0029] Preferably, the central control module linearly superimposes or non-linearly fuses all the scheduling dimension parameters of the scheduling unit to generate the comprehensive scheduling scheme.

[0030] Preferably, the system further comprises a storage module connected with the central control module, and the storage module is used for storing the order parameter set of the scheduling unit, the path cost index, the real-time traffic state index and the comprehensive scheduling scheme.

[0031] Preferably, the system further comprises an alarm module connected with the central control module, and the alarm module is used for generating a multi-level alarm instruction according to the deviation value of the comprehensive scheduling scheme from the preset scheduling benchmark, and sending the alarm instruction to a designated scheduling terminal through a communication gateway.

[0032] Compared with the prior art, the present application has the following beneficial effects:

[0033] In terms of improving scheduling accuracy, the system sets a region division module, which divides the target region into multiple scheduling units according to the logistics task requirements, and formulates exclusive task allocation rules for each scheduling unit. The order acquisition module obtains comprehensive order parameter sets from the logistics database in real time according to these rules, the path calculation module calculates accurate path cost indexes by using the space-time convolution algorithm combined with the preset path correlation relationship, and the dynamic adjustment module obtains real-time traffic state indexes by using the preset space-time correlation analysis method. These rich and accurate scheduling dimension parameters provide a solid data foundation for subsequent scheduling decisions. The central control module fuses these parameters, which can more comprehensively and accurately consider various factors compared with the traditional scheduling method which only relies on part of the information or simple rules, and formulates a scheduling scheme that is more in line with the actual situation, thereby greatly improving the accuracy of scheduling.

[0034] In terms of responding to dynamic changes, the system modules work cooperatively and have strong real-time response capability. The order acquisition module obtains order information in real time, and can timely incorporate newly generated orders; the dynamic adjustment module continuously monitors the traffic state, and can quickly feed back the real-time traffic state index to the central control module once the traffic condition changes, such as congestion, accidents, etc. The central control module re-optimizes the scheduling scheme based on these dynamic information, ensuring that the vehicle can timely avoid congested road sections and adjust the delivery route during driving. Compared with the traditional static scheduling system, the present system can closely follow the dynamic changes in logistics business, and always maintain efficient delivery efficiency, avoiding delivery delay caused by information lag.

[0035] In terms of comprehensive optimization of multiple objectives, the system fully considers multiple key objectives in logistics distribution. The calculation of path cost indicators helps control transportation costs, reduces vehicle mileage and fuel consumption by reasonable route planning; the use of real-time traffic state indicators can improve distribution efficiency and ensure timely delivery of goods; vehicle utilization is also improved, and through comprehensive analysis of order parameters and vehicle information, reasonable vehicle loading is achieved to avoid empty or overloaded vehicles. In addition, the system can also consider environmental factors to some extent, such as reducing carbon emissions by optimizing routes to achieve a balance between economic and environmental benefits. This comprehensive optimization of multiple objectives breaks the limitations of traditional single-objective optimization, enabling logistics enterprises to achieve better results in cost control, service quality improvement, and sustainable development.

[0036] In terms of data mining and analysis, the system uses deep learning-related technologies to deeply mine a large amount of logistics data. Spatio-temporal convolution algorithms and preset spatio-temporal correlation analysis methods can effectively mine potential correlations and patterns of logistics data in time and space dimensions. For example, by analyzing historical traffic data and order data, the system can predict traffic flow changes and order demand trends in different time periods and regions, providing strong support for formulating reasonable scheduling strategies in advance. This deep data mining capability helps logistics enterprises better grasp market dynamics, optimize resource allocation, and improve overall operational management.

[0037] In terms of system scalability and compatibility, the invention is designed reasonably, with good scalability and compatibility. Each module in the system works independently and cooperatively, and when the business scale of logistics enterprises expands or business needs change, individual modules can be easily upgraded or expanded. For example, the route optimization module can introduce more advanced spatio-temporal trajectory analysis algorithms according to actual needs, and the storage module can expand storage capacity or optimize storage structure. At the same time, the system can be well compatible with existing logistics databases, communication gateways, and various scheduling terminals, facilitating logistics enterprises to upgrade and integrate the system on the existing basis, reducing the cost and difficulty of system upgrading. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 The working principle diagram of the deep learning-based multi-objective logistics vehicle dynamic scheduling system described in the invention;

[0039] Figure 2 The flowchart for obtaining the order parameter set according to the task allocation rule;

[0040] Figure 3 The flowchart for calculating path cost indicators under different path association relationships;

[0041] Figure 4A flow chart for calculating time correlation for the route optimization module;

[0042] Figure 5 A flow chart for generating a comprehensive scheduling scheme for the central control module. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0044] Please refer to Figures 1-5 The present application provides a multi-target logistics vehicle dynamic scheduling system based on deep learning, and the specific implementation steps are as follows:

[0045] The system includes a region division module, an order collection module, a path calculation module, a dynamic adjustment module, and a central control module. The region division module divides the target region according to the logistics task demand, and divides it into multiple scheduling units. In the division process, the distribution characteristics of the logistics task, the geographical region characteristics, and the traffic conditions and other factors are fully considered to ensure that the division of each scheduling unit is reasonable and efficient. After the division is completed, the corresponding task allocation rules are generated for each scheduling unit, which are important basis for subsequent order collection and scheduling.

[0046] The order collection module acquires the order parameter set of each scheduling unit from the logistics database in real time according to the task allocation rules generated by the region division module. The module establishes a stable data connection with the logistics database, accurately filters out the order data that meets the requirements of each scheduling unit according to the task allocation rules, and takes all parameters in the order parameter set as the scheduling dimension parameters of the scheduling unit.

[0047] The path calculation module calculates the path cost index of each scheduling unit based on the pre-set path correlation relationship in the logistics database using the space-time convolution algorithm. In the calculation process, the space-time characteristics of the path are fully considered, the path information in the logistics database is combined, and the accurate path cost index is obtained through the algorithm, which is used as the scheduling dimension parameter of the scheduling unit.

[0048] The dynamic adjustment module obtains the real-time traffic state index of each scheduling unit through a pre-set space-time correlation analysis method. The module analyzes the traffic data of the scheduling unit and its adjacent scheduling units, considers factors such as traffic fluctuation mode and historical data, calculates the real-time traffic state index, and uses it as the scheduling dimension parameter of the scheduling unit.

[0049] The central control module plays a core role in coordinating and controlling the whole system. It sends scheduling unit information to the path calculation module and the dynamic adjustment module to provide necessary data support for the calculation and analysis of the two modules; and sends task allocation rules to the order collection module to ensure the accurate performance of order collection. In addition, the central control module performs data fusion on all scheduling dimension parameters of the scheduling unit, and generates a comprehensive scheduling scheme of the scheduling unit through a specific fusion algorithm.

[0050] The embodiments of the present application are further illustrated below through specific examples:

[0051] Example 1:

[0052] In obtaining the real-time traffic state index, the system takes both the current scheduling unit and the adjacent scheduling unit as dynamic analysis nodes. Based on the path historical data of the scheduling unit belonging to the logistics area, the system deeply analyzes the traffic fluctuation patterns of each dynamic analysis node in the historical period. Through the mining and sorting of a large amount of historical data, the traffic fluctuation rules in different time periods and under different road conditions are extracted. On the basis of these traffic fluctuation patterns, the system calculates the initial traffic index of each dynamic analysis node and the deviation of each path historical data from the initial value according to the fluctuation amplitude of each path historical data in the corresponding dynamic analysis node.

[0053] Based on the initial traffic index of each dynamic analysis node and the deviation of each path historical data from the initial value, the system uses the dynamic topology network to realize the synchronous optimization of the final traffic index of each dynamic analysis node and the deviation value of each path historical data, thereby obtaining the real-time traffic state index of each dynamic analysis node. In the analysis process, the system also sets a comparison dynamic analysis node as any one of all dynamic analysis nodes, and a comparison path historical data as any one of all path historical data. At the same time, based on the traffic fluctuation pattern, the system takes the traffic change amount of the comparison dynamic analysis node in the comparison path historical data as an independent change amount, takes the total change amount of the comparison dynamic analysis node in all path historical data as an independent total variable, takes the change amount of all dynamic analysis nodes in the comparison path historical data as a global change amount, and takes the total change amount of all dynamic analysis nodes in all path historical data as a global total variable. If the ratio of the independent change amount to the independent total variable exceeds the ratio of the global change amount to the global total variable, it is determined that the comparison dynamic analysis node has a significant congestion feature in the comparison path historical data. In this way, the system can timely discover traffic congestion and provide accurate basis for subsequent scheduling decisions.

[0054] Suppose there is a large logistics distribution area divided into multiple scheduling units, among which the A scheduling unit and its adjacent B and C scheduling units are dynamic analysis nodes.

[0055] The logistics system has long accumulated rich path history data in the region, such as the traffic volume and speed of each road segment at different times of the day over the past year. Through in-depth analysis of this data, the system found that during the morning rush hour on weekdays, the traffic volume on the main road connecting the A and B dispatch units increased significantly, and the speed decreased significantly, showing a regular traffic fluctuation pattern. At night, the overall traffic volume in the region is small, and the traffic conditions are relatively smooth, and the fluctuation pattern is also different.

[0056] Based on the fluctuation amplitude of each path history data in the corresponding dynamic analysis node, the system begins to calculate the initial traffic index and the deviation from the initial value. Taking a main distribution route in the A dispatch unit as an example, during the morning rush hour on a certain weekday, its traffic volume increased by 50% compared to usual. According to the set calculation rules, the system calculates the initial traffic index of the A dispatch unit during this period. At the same time, by comparing the traffic volume of this period with the average traffic volume of this route, the deviation of this path history data from the initial value is obtained. Similarly, the path history data in the B and C dispatch units is also calculated.

[0057] Next, the system uses the dynamic topology network to synchronize the optimization of the final traffic index of each dynamic analysis node and the deviation value of each path history data. This dynamic topology network continuously collects real-time traffic data and adjusts the calculation method based on the relationship between nodes. For example, when a traffic accident occurs on a road segment in the A dispatch unit, causing a sharp decrease in traffic volume, the dynamic topology network quickly captures this change and recalculates the final traffic index of the A dispatch unit. At the same time, it adjusts the traffic index of the B and C dispatch units related to it, ensuring that accurate real-time traffic status indicators are obtained for each dynamic analysis node.

[0058] In the analysis process, assume that the A dispatch unit is the comparative dynamic analysis node, and the path history data of the trunk road connecting the A and B dispatch units in a certain time period is the comparative path history data. During this time period, the traffic volume of the A dispatch unit on this trunk road increases by 300 vehicles, which is the independent change amount; the total change amount of the traffic volume of the A dispatch unit in all path history data is 1000 vehicles, which is the independent total variable. The total change amount of the traffic volume of all dynamic analysis nodes (A, B, and C dispatch units) in this comparative path history data is 800 vehicles, which is the global change amount; the total change amount of the traffic volume of all dynamic analysis nodes in all path history data is 2500 vehicles, which is the global total variable. After calculation, the ratio of the independent change amount to the independent total variable is 0.3, and the ratio of the global change amount to the global total variable is 0.32. Since 0.3 does not exceed 0.32, it is determined that the A dispatch unit does not have a significant congestion feature in this comparative path history data at this time. If the ratio exceeds, it is determined that there is a significant congestion feature, and the system will timely issue a warning to provide an important reference for subsequent logistics vehicle scheduling decisions, such as guiding vehicles to avoid congested road sections and re-planning more efficient distribution routes.

[0059] Embodiment 2:

[0060] The task allocation rule includes a fixed parameter type group and a variable parameter type group. The fixed parameter type group contains multiple fixed parameter types, and each fixed parameter type corresponds to a static weight. These fixed parameter types are determined according to the basic characteristics and long-term experience of logistics business, such as the weight and volume of orders, and their static weights remain stable within a certain period of time. The variable parameter type group contains multiple variable parameter types, and each variable parameter type corresponds to an elastic weight. Variable parameter types may include the urgency of orders, special requirements of customers, etc., and their elastic weights will be dynamically adjusted according to actual business conditions. When obtaining the order parameter set, the order collection module extracts the structured order data of the dispatch unit from the logistics database. For each fixed parameter type in the fixed parameter type group, the order collection module accurately extracts the corresponding fixed parameter from the structured order data; for each variable parameter type in the variable parameter type group, the corresponding variable parameter is also extracted from the structured order data. Finally, all extracted fixed parameters and variable parameters are combined into the order parameter set of the dispatch unit. This task allocation rule and order parameter set acquisition method can flexibly adapt to different logistics task requirements, improving the accuracy and efficiency of scheduling.

[0061] Suppose in a logistics distribution scenario in a city, there is a logistics distribution center responsible for the distribution of goods in multiple regions, and the city is divided into multiple dispatch units, with the central area as a dispatch unit for analysis.

[0062] The task allocation rule set by the logistics system includes a fixed parameter type group and a variable parameter type group. In the fixed parameter type group, there are two fixed parameter types, order weight and order volume, and static weights are set for them respectively. For example, the static weight of order weight is set to 0.4, and the static weight of order volume is set to 0.3. This is because in this logistics distribution scenario, according to long-term operation experience, the weight and volume of goods have a greater and relatively stable impact on vehicle loading planning and transportation cost, so they are given fixed weights.

[0063] In the variable parameter type group, there are two variable parameter types, order urgency and customer special requirements. Order urgency is divided into three levels, ordinary, urgent and special urgent, corresponding to different elastic weights, ordinary is 0.1, urgent is 0.2, and special urgent is 0.4; for customer special requirements, if the customer requires delivery within a specific time period, an elastic weight of 0.2 is given, and if there is no special requirement, the weight is 0. The elastic weight will be dynamically adjusted according to the actual business situation, for example, during the shopping festival and other order peak periods, the weight of urgent orders may be appropriately increased to prioritize the delivery of urgent orders.

[0064] When the order collection module starts working, it extracts structured order data of the city center dispatch unit from the logistics database. For one of the orders, the order weight is 50 kg and the order volume is 2 cubic meters, these two data are fixed parameters extracted from the order data according to the requirements in the fixed parameter type group. Then, check the urgency of the order as urgent, and the customer requires delivery between 10am-11am, these two data are variable parameters extracted according to the variable parameter type group.

[0065] After that, the order collection module summarizes all the extracted fixed parameters and variable parameters into the order parameter set of the dispatch unit. In this example, the order parameter set contains order weight 50 kg, order volume 2 cubic meters, order urgency (urgent), customer special requirements (delivery between 10am-11am) and other parameters. These parameters are used as dispatch dimension parameters of the dispatch unit, which will be used for more accurate logistics vehicle dispatch planning in the future. For example, when the system arranges vehicles, it will consider these parameters comprehensively, and preferentially arrange vehicles with appropriate load weight and meet the delivery time requirements of urgent orders, and reasonably plan the loading scheme of the vehicle according to the volume of goods, to improve the efficiency and service quality of logistics distribution.

[0066] Embodiment 3:

[0067] When the preset path association relationship is a direct path relationship, the system first determines whether the dispatch unit is associated to a path node in the logistics database. If the dispatch unit is associated to a path node, the path cost index of the dispatch unit is the sum of the cost contribution values of all path nodes associated thereto. The cost contribution value is determined according to the nature, location and association degree of the dispatch unit and other factors. If the dispatch unit is not associated to a path node, the path cost index of the dispatch unit is the basic path cost of the dispatch unit itself, which is preset based on the geographical location, surrounding traffic conditions and other factors of the dispatch unit. When the preset path association relationship is an indirect path relationship, the path cost index of the dispatch unit is the dynamic weighted value of the cost contribution values of all path nodes obtained by the dispatch unit through the cross-regional associated path. In calculating the dynamic weighted value, the system considers multiple factors such as the length, traffic conditions and transportation time of the cross-regional path, and calculates the cost contribution value by a specific algorithm to obtain an accurate path cost index. This way of calculating the path cost index according to different path association relationships can more accurately reflect the actual transportation cost and provide a reliable basis for path planning.

[0068] Suppose in a logistics distribution network covering multiple cities, there is a dispatch unit responsible for delivering goods from a logistics center in city A to several small towns around.

[0069] When the preset path association relationship is a direct path relationship, the system first determines whether the dispatch unit is associated to a path node in the logistics database. If the dispatch unit is associated to a path node, the path cost index of the dispatch unit is the sum of the cost contribution values of all path nodes associated thereto. The cost contribution value is determined according to the nature, location and association degree of the dispatch unit and other factors. If the dispatch unit is not associated to a path node, the path cost index of the dispatch unit is the basic path cost of the dispatch unit itself, which is preset based on the geographical location, surrounding traffic conditions and other factors of the dispatch unit. When the preset path association relationship is an indirect path relationship, the path cost index of the dispatch unit is the dynamic weighted value of the cost contribution values of all path nodes obtained by the dispatch unit through the cross-regional associated path. In calculating the dynamic weighted value, the system considers multiple factors such as the length, traffic conditions and transportation time of the cross-regional path, and calculates the cost contribution value by a specific algorithm to obtain an accurate path cost index. This way of calculating the path cost index according to different path association relationships can more accurately reflect the actual transportation cost and provide a reliable basis for path planning. n1 n2 n3 The cost contribution value may be determined according to the operating cost, maintenance cost and relevance to the distribution task of the node, etc. Then, the calculation formula of the path cost index C 直 of the dispatch unit to town A is: C 直 =C n1 +C n2 +C n3 .

[0070] If it is to town B, the logistics database shows that there is no directly associated path node, and at this time the path cost index of the dispatch unit to town B is the basic path cost of the dispatch unit itself. The basic path cost is estimated according to the approximate distance from the logistics center in city A to town B, the estimated transportation time, the vehicle empty load loss and other factors, and is assumed to be C 基 .

[0071] ​​When the preset path association relationship is an indirect path relationship, the scheduling unit needs to send the goods to town C through the cross-regional associated path. Town C is not directly connected to the logistics center of city A, but is transferred through the logistics node of city B. From the logistics center of city A to the logistics node of city B to town C, multiple path nodes are involved. Suppose the path nodes obtained through the cross-regional associated path are m1, m2, m3…m n , and their respective cost contribution values are D m1 , D m2 , D m3 …D mn , the dynamic weight of each path node is w m1 , w m2 , w m3 …w mn . The dynamic weight is adjusted in real time according to real-time traffic conditions of the path, busy degree of transportation period, priority of different paths, etc. Then, the calculation formula of the path cost index C 间 of the scheduling unit to town C is: For example, after real-time calculation, the cost contribution value D m1 of m1 node is 10, the dynamic weight w m1 is 0.8; the cost contribution value D m2 of m2 node is 15, the dynamic weight w m2 is 0.9, etc. Adding the product of the cost contribution value and the dynamic weight of all related nodes can obtain the path cost index of the scheduling unit to town C.

[0072] Through this way of calculating the path cost index according to different path association relationships, the logistics system can more accurately evaluate the cost of different distribution routes, provide scientific basis for subsequent vehicle scheduling and path planning, and then optimize the logistics distribution process and reduce the operating cost.

[0073] Embodiment 4:

[0074] The system also includes a route optimization module connected to the central control module. The route optimization module calculates the time correlation degree between the scheduling unit and the feasible route according to the real-time road network state through the preset space-time trajectory analysis method, and takes the time correlation degree as the scheduling dimension parameter of the scheduling unit. When calculating the time correlation degree, the route optimization module obtains the historical optimal route set of the scheduling unit from the path planning model library, and extracts the current feasible route set from the real-time road network database. Then, the time-space coincidence degree between the historical optimal route set and the feasible route set is taken as the first correlation degree between the scheduling unit and the feasible route. Next, the delivery time limit requirement of the scheduling unit within the preset time window is dynamically matched with the real-time time limit data of the feasible route to obtain the time limit matching degree between the two as the second correlation degree between the scheduling unit and the feasible route. Finally, the route optimization module can select the first correlation degree, the second correlation degree, or the weighted result of the two as the time correlation degree. In this way, the route optimization module can consider historical experience and real-time road conditions to select a more suitable route for the scheduling unit and improve delivery efficiency.

[0075] Suppose in a medium-sized city, a logistics enterprise is responsible for the delivery of goods within the city and its surrounding areas, and a certain scheduling unit is responsible for the delivery task in the northeast region of the city.

[0076] The route optimization module starts to work, first obtaining the historical optimal route set of the scheduling unit from the path planning model library. This historical optimal route set is based on the long-term accumulated delivery data of the logistics enterprise and is obtained by algorithm analysis. For example, under normal traffic conditions during the day on weekdays, a historical optimal route from the logistics center to multiple delivery points in the northeast scheduling unit of the city, which takes into account factors such as traffic signal duration, road speed limit, average traffic volume on each section, etc., can make the delivery time relatively shortest.

[0077] At the same time, the route optimization module extracts the current feasible route set from the real-time road network database. Suppose at 10 o'clock in the morning on a certain weekday, due to temporary construction in part of the northeast region of the city, some regular routes cannot be used. The real-time road network database updates this information in time, and the route optimization module extracts the current feasible route set accordingly, which includes some routes that are less used normally but can be used as alternatives under the current circumstances.

[0078] The spatio-temporal coincidence degree of the historical optimal route set and the feasible route set is calculated, and is used as the first correlation degree between the scheduling unit and the feasible route. Assuming that there is a 5-kilometer route in the historical optimal route, and in the current feasible route set, there are 3 kilometers that coincide with it, and in terms of time, the estimated travel time of the historical optimal route on the route and the estimated travel time of the feasible route on the coinciding route also have a certain degree of coincidence. Through a specific calculation method (not involving formulas), it is obtained that the spatio-temporal coincidence degree of the two route sets is 60%, that is, the first correlation degree is 60%.

[0079] The delivery time efficiency requirement of the scheduling unit within the preset time window is dynamically matched with the real-time time efficiency data of the feasible route, and the time efficiency matching degree of the two is obtained as the second correlation degree between the scheduling unit and the feasible route. Assuming that a batch of goods received by the scheduling unit needs to be delivered in full before 11:30 am, which is the delivery time efficiency requirement within the preset time window. For a certain feasible route, according to the real-time traffic information and the vehicle travel speed prediction, the vehicle travels along this route and completes all delivery tasks and is estimated to take 1.5 hours. If it starts from the current time (10:00 am), 1.5 hours later is 11:30 am, which just meets the delivery time efficiency requirement. Through calculation (not involving formulas), it is obtained that the time efficiency matching degree of the feasible route and the delivery time efficiency requirement is 100%, that is, the second correlation degree is 100%.

[0080] The route optimization module can select the first correlation degree, the second correlation degree, or the weighted result of the two as the time correlation degree. Assuming that according to the business focus of the logistics enterprise, the weight of the first correlation degree is set to 0.4, and the weight of the second correlation degree is set to 0.6. Then for this feasible route, the time correlation degree is 60%*0.4+100%*0.6=84%. Through such calculation on all feasible routes, the route optimization module can select the route with the highest time correlation degree and recommend it to the scheduling system, and the scheduling system can plan a more reasonable delivery route for the logistics vehicle according to these information, thereby improving the delivery efficiency and ensuring the timely delivery of goods.

[0081] Embodiment 5:

[0082] The system also includes a storage module and an alarm module connected with the central control module. The storage module is used to store the order parameter set of the scheduling unit, the path cost index, the real-time traffic state index and the comprehensive scheduling scheme. The storage module uses an efficient database management system to store and manage these data in a classified manner, ensuring the security and queryability of the data. The alarm module generates multi-level alarm instructions according to the deviation value of the comprehensive scheduling scheme from the preset scheduling benchmark. When generating the alarm instructions, the alarm module analyzes and judges the deviation value to determine the alarm level according to the size and nature of the deviation. Then, the alarm module sends the alarm instructions to the designated scheduling terminal through the communication gateway. The scheduling personnel can adjust the scheduling scheme in a timely manner according to the alarm instructions to ensure that the scheduling work of the logistics vehicles can be carried out according to the preset standard, improving the stability and reliability of the logistics operation.

[0083] Suppose a large logistics enterprise is responsible for the logistics distribution of a region, which is divided into multiple scheduling units. Take one of the scheduling units responsible for the distribution of the city business district as an example.

[0084] The storage module is connected with the central control module and undertakes the key task of data storage. In the daily distribution process, various data generated by the scheduling unit will be stored by the storage module in a classified manner. For example, the order parameter set of a batch of orders is recorded, including the delivery address, weight, volume, customer required delivery time and other detailed information of the order. The path cost index will also be stored, assuming that from the logistics warehouse to several delivery points in the business district, the cost index of different routes will be different due to distance, road conditions, transportation time and other factors. These indexes will be accurately recorded. Real-time traffic state indexes will also be stored, such as congestion conditions and vehicle speed of major roads in the business district during a certain period of time. In addition, the comprehensive scheduling scheme generated by the central control module will also be stored, such as the detailed plan of arranging a vehicle to start at a specific time and follow a specific route to deliver to multiple delivery points in sequence. The storage module uses an advanced database management system to store these data in an orderly manner according to their respective categories and time sequence, facilitating subsequent query and analysis.

[0085] The alarm module is also connected to the central control module and plays an important role in monitoring and early warning. The logistics enterprise sets a dispatch benchmark in advance, for example, it requires that the goods must be delivered within plus or minus 30 minutes of the specified delivery time of the customer. When the deviation occurs during the execution of the comprehensive scheduling scheme, the alarm module will start. Suppose that in a certain delivery task, the originally planned vehicle should deliver the goods to the customer at 10 o'clock in the morning, but according to the real-time monitoring data, due to a sudden traffic accident on the road, the vehicle is expected to arrive at 11 o'clock in the morning, with a deviation of 60 minutes, far exceeding the preset 30 minutes. After the alarm module detects this deviation, it will generate multi-level alarm instructions according to the preset alarm rules. Because this deviation is large, a high-level alarm instruction may be generated. The alarm module sends the alarm instruction to the designated dispatch terminal through the communication gateway, such as the dispatchers' mobile phones or computer terminals. After receiving the alarm instruction, the dispatcher can take timely measures, such as contacting the customer to explain the situation, and at the same time, re-planning the vehicle route to guide the vehicle to avoid congested road sections, or coordinating other vehicles to perform relay delivery, so as to minimize the impact of the delay on the customer and ensure the stability and reliability of the logistics delivery service.

[0086] It should be noted that the relational terms herein such as first and second and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0087] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A multi-objective logistics vehicle dynamic scheduling system based on deep learning, characterized by: include: The regional division module is used to divide the target area into multiple scheduling units according to the logistics task requirements and generate corresponding task allocation rules for each scheduling unit; An order collection module is used to obtain the order parameter set of each scheduling unit from the logistics database in real time according to the task allocation rule, and use all parameters in the order parameter set as scheduling dimension parameters of the scheduling unit; A path calculation module, configured to calculate the path cost index of each scheduling unit based on the path association relationship preset in the logistics database by using a spatiotemporal convolution algorithm, and use the path cost index as a scheduling dimension parameter of the scheduling unit; A dynamic adjustment module is used to obtain the real-time traffic status index of each scheduling unit through a preset spatiotemporal correlation analysis method, and use the real-time traffic status index as a scheduling dimension parameter of the scheduling unit; The central control module is used to send the scheduling unit information to the path calculation module and the dynamic adjustment module, send the task allocation rules to the order collection module, and perform data fusion on all the scheduling dimension parameters of the scheduling unit to generate a comprehensive scheduling plan for the scheduling unit.

2. The multi-objective logistics vehicle dynamic scheduling system based on deep learning according to claim 1 is characterized in that: The method of obtaining the real-time traffic status indicators of each dispatching unit by using a preset spatiotemporal correlation analysis method includes: The current dispatch unit and adjacent dispatch units are used as dynamic analysis nodes, and based on the historical path data of the logistics area to which the dispatch unit belongs, the traffic fluctuation pattern of each dynamic analysis node in the historical period is determined; Based on the traffic fluctuation pattern, and based on the fluctuation amplitude of each path history data in the corresponding dynamic analysis node, calculating the initial traffic index of each dynamic analysis node and the deviation initial value of each path history data; Based on the initial traffic index of each dynamic analysis node and the initial deviation value of each path historical data, the final traffic index of each dynamic analysis node and the deviation value of each path historical data are synchronously optimized through the dynamic topology network to obtain the real-time traffic status index of each dynamic analysis node; Set the comparison dynamic analysis node to any one of all dynamic analysis nodes, and the comparison path historical data to any one of all path historical data. At the same time, based on the traffic fluctuation pattern, take the flow change of the comparison dynamic analysis node in the comparison path historical data as the independent change, take the total change of the comparison dynamic analysis node in all path historical data as the independent total variable, take the change of all dynamic analysis nodes in the comparison path historical data as the global change, and take the total change of all dynamic analysis nodes in all path historical data as the global total variable. If the ratio of the independent change to the independent total variable exceeds the ratio of the global change to the global total variable, then it is determined that the comparison dynamic analysis node has significant congestion characteristics in the comparison path historical data.

3. The multi-objective logistics vehicle dynamic scheduling system based on deep learning according to claim 1 is characterized in that: The task allocation rules include a fixed parameter type group and a variable parameter type group; the fixed parameter type group includes multiple fixed parameter types and a static weight corresponding to each fixed parameter type; The variable parameter type group includes multiple variable parameter types and an elastic weight corresponding to each variable parameter type.

4. The multi-objective logistics vehicle dynamic scheduling system based on deep learning according to claim 2 is characterized in that: The step of acquiring the order parameter set of each scheduling unit from the logistics database in real time according to the task allocation rule includes: The order collection module extracts the structured order data of the scheduling unit from the logistics database, extracts corresponding fixed parameters from the structured order data based on each fixed parameter type in the fixed parameter type group, and extracts corresponding variable parameters from the structured order data based on each variable parameter type in the variable parameter type group, and summarizes all the fixed parameters and all the variable parameters into an order parameter set of the scheduling unit.

5. The multi-objective logistics vehicle dynamic scheduling system based on deep learning according to claim 1 is characterized in that: When the preset path association relationship is a direct path relationship, determine whether the scheduling unit is associated with a path node in the logistics database. If so, the path cost index of the scheduling unit is the sum of the cost contribution values ​​of all path nodes associated with the scheduling unit. Otherwise, the path cost index of the scheduling unit is the basic path cost of the scheduling unit itself. When the preset path association relationship is an indirect path relationship, the path cost index of the scheduling unit is a dynamic weighted value of the cost contribution values ​​of all path nodes obtained by the scheduling unit through the cross-region association path.

6. The multi-objective logistics vehicle dynamic scheduling system based on deep learning according to claim 1 is characterized in that: Also included is a route optimization module connected to the central control module; The route optimization module is used to calculate the time correlation between the scheduling unit and the feasible route according to the real-time road network status through a preset spatiotemporal trajectory analysis method, and use the time correlation as the scheduling dimension parameter of the scheduling unit.

7. The multi-objective logistics vehicle dynamic scheduling system based on deep learning according to claim 6 is characterized in that: The route optimization module calculates the time correlation between the scheduling unit and the feasible route by a preset spatiotemporal trajectory analysis method, including: Obtaining a historical optimal route set of the dispatching unit from a path planning model library, extracting a current feasible route set from a real-time road network database, and using a spatiotemporal overlap between the historical optimal route set and the feasible route set as a first correlation between the dispatching unit and the feasible route; Dynamically matching the delivery time efficiency requirement of the dispatching unit within a preset time window with the real-time time efficiency data of the feasible route, and obtaining a time efficiency matching degree between the two as a second correlation degree between the dispatching unit and the feasible route; The first correlation degree, the second correlation degree, or a weighted result of the two is used as the time correlation degree.

8. The multi-objective logistics vehicle dynamic scheduling system based on deep learning according to claim 1 is characterized in that: The central control module performs linear superposition or nonlinear fusion on all the scheduling dimension parameters of the scheduling unit to generate the comprehensive scheduling solution.

9. The multi-objective logistics vehicle dynamic scheduling system based on deep learning according to claim 1 is characterized in that: It also includes a storage module connected to the central control module, and the storage module is used to store the order parameter set, path cost index, real-time traffic status index and comprehensive scheduling plan of the scheduling unit.

10. The multi-objective logistics vehicle dynamic scheduling system based on deep learning according to claim 1 is characterized in that: It also includes an alarm module connected to the central control module, which is used to generate multi-level alarm instructions according to the deviation value between the comprehensive scheduling plan and the preset scheduling benchmark, and send the alarm instructions to the designated scheduling terminal through the communication gateway.

Images