Unmanned aerial vehicle multi-order distribution carbon emission reduction benefit prediction method and system

Abstract

The invention discloses an unmanned aerial vehicle multi-order distribution carbon emission reduction benefit prediction method and system, and relates to the technical field of carbon emission reduction benefit prediction, and the method comprises the steps: obtaining unmanned aerial vehicle distribution data, carrying out the processing of the unmanned aerial vehicle distribution data, separating an independent distribution order and an order-combined distribution order, and taking the separated distribution order and order-combined distribution order as order-combined distribution stroke data; variable features are constructed on the basis of the order-combining distribution travel data, feature variables of an unmanned aerial vehicle order-combining distribution travel emission reduction benefit prediction model are generated, and construction of the variable features comprises three levels of time, space and travel; the feature variables of the unmanned aerial vehicle order-combining distribution travel emission reduction benefit prediction model are input into a pre-established optimization prediction model, a carbon emission reduction benefit prediction result is obtained through output, and the pre-established optimization prediction model is obtained after key hyper-parameters are optimized through a Bayesian hyper-parameter optimization method by means of a gradient lifting model.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon emission reduction benefit prediction, and specifically to a method and system for predicting carbon emission reduction benefits of unmanned aerial vehicle multi-order delivery. Background Art

[0002] With the rapid development of society and economy and the accelerated urbanization process, urban transportation systems are under unprecedented pressure. Environmental pollution and energy consumption are becoming increasingly serious, which not only threatens the sustainable development of cities, but also has a profound impact on human health and the ecological environment. In the face of the urgent situation of global climate change, reducing carbon emissions in the transportation sector and promoting green and low-carbon transportation have become the common concerns of governments, scientific research institutions and all sectors of society around the world.

[0003] As an emerging logistics model, drone delivery has shown great potential in alleviating urban traffic pressure and improving delivery efficiency. However, this innovative model is also accompanied by carbon emissions, and its environmental impact needs to be fully evaluated. Although drones have low-carbon advantages in some aspects compared to traditional ground transportation tools, in actual operation, the calculation of their carbon footprint and the exploration of emission reduction potential are still full of challenges. For example, the manufacturing, maintenance, battery production and charging of drones will generate certain carbon emissions. In addition, factors such as route planning, flight frequency and load efficiency of drone delivery also directly affect its overall carbon emission reduction benefits. Summary of the invention

[0004] In order to address the deficiencies mentioned in the above background technology, the purpose of the present invention is to provide a method and system for predicting the carbon emission reduction benefits of drone multi-order delivery, which can optimize the group order delivery service and formulate emission reduction strategies by accurately evaluating and predicting the carbon emission reduction benefits of individual group order delivery trips.

[0005] In the first aspect, the purpose of the present invention can be achieved by the following technical solution: a method for predicting carbon emission reduction benefits of drone multi-order delivery, the method comprising the following steps:

[0006] Obtain drone delivery data, process the drone delivery data, separate individual delivery orders and group delivery orders, and use them as group delivery trip data;

[0007] Based on the group-order delivery trip data, the variable characteristics are constructed to generate the characteristic variables of the drone group-order delivery travel emission reduction benefit prediction model. The construction of variable characteristics includes three levels: time, space and travel.

[0008] The characteristic variables of the drone group delivery emission reduction benefit prediction model are input into a pre-established optimization prediction model, and the carbon emission reduction benefit prediction result is output. The pre-established optimization prediction model is a gradient boosting model obtained by optimizing key hyperparameters through the Bayesian hyperparameter optimization method.

[0009] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: the process of processing the drone delivery data includes: data cleaning, group delivery order identification, indicator calculation and outlier processing.

[0010] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the process of identifying the group order delivery order:

[0011] The group delivery orders and individual delivery orders in the order data set are distinguished by the group delivery trip identification algorithm, that is, two adjacent orders of the same drone sorted in chronological order: If the time periods of the two orders overlap, that is, the end billing time of the first order is later than the start billing time of the second order: ET 1 >ST 2 , then these two orders are group delivery orders, and two group delivery orders constitute one group delivery trip.

[0012] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: the index calculation includes the calculation of trajectory segment indicators, individual delivery trip indicators, group delivery trip indicators, and drone group delivery travel order indicators.

[0013] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the trajectory segment index calculation includes:

[0014] Track segment interval length:

[0015] Assume that the trajectory segment s of drone n in order i is formed by connecting two consecutive trajectory points m and m+1 of drone n in order i, then the interval length of the trajectory segment is It is expressed as:

[0016]

[0017] In the formula, geodesic is a function in Python that calculates the distance between two locations. is the location information of trajectory point m of drone n in order i;

[0018] Track segment interval time:

[0019] The starting time of trajectory segment s The timestamp of trajectory point m End time of trajectory segment s That is the timestamp of trajectory point m+1 Track segment interval time The end time of the track segment and starting time The differences are:

[0020]

[0021] Average speed of trajectory segment:

[0022] Average speed of trajectory segment Equal to the interval length of the track segment With interval time The ratio is expressed as:

[0023]

[0024] Carbon emission factor of trajectory segment It is expressed as:

[0025]

[0026] In the formula, α, β, γ, δ, ε, ζ, and η are the relevant parameters for calculating the carbon emission factor in the COPERT model;

[0027] Carbon emissions per track segment:

[0028] Carbon emissions of trajectory segments Emission factor equal to the trajectory segment Distance between track segments The product of, that is:

[0029]

[0030] Trajectory segment outlier processing:

[0031] Clear the abnormal track segments with an interval time of more than 60s and an interval distance not equal to 0, or an interval distance of more than 500m, or an average speed of more than 120km / h;

[0032] The calculation of individual delivery trip metrics includes:

[0033] Separate delivery travel time:

[0034] For a single delivery trip k of drone n, the travel time of single delivery trip k is Equal to the order duration of the corresponding individual delivery order i, that is, the end billing time ET of the individual delivery order i i n and start billing time ST i nDifference:

[0035]

[0036] Distance of single delivery trip:

[0037] Introducing correction factor It is equal to the actual travel time of the individual delivery trip The cumulative duration of the trajectory segment corresponding to the individual delivery order i The ratio of, that is:

[0038]

[0039] Travel distance for individual delivery trips Equal to the correction factor The cumulative length of the trajectory segments corresponding to the individual delivery orders The product of, that is:

[0040]

[0041] Carbon emissions per delivery trip

[0042] Carbon emissions from a single delivery trip Equal to the correction factor Cumulative carbon emissions of the trajectory segments corresponding to individual delivery orders The product of, that is:

[0043]

[0044] Handling of abnormal values ​​for individual delivery trips

[0045] The abnormal individual delivery trips with serious missing track segments are cleared and only the correction coefficient is retained Individual delivery trips with a value less than or equal to 1.1;

[0046] The process of calculating the group order delivery itinerary index:

[0047] Group delivery travel time:

[0048] For group delivery trip j of drone n, the travel time of group delivery trip j is Equal to the set of group delivery orders it contains Φ j The latest end billing time max(ET i n ) and the earliest billing time min(ST i n ), that is:

[0049]

[0050] Distance of group delivery

[0051] Correction factor for group delivery trips Equal to the travel time of the group delivery trip The cumulative duration of the track segments corresponding to all group delivery orders included in the group delivery trip The product of, that is:

[0052]

[0053] Travel distance for group delivery trips Equal to the correction factor The cumulative length of the track segments corresponding to all group delivery orders included in the group delivery trip The product of, that is:

[0054]

[0055] Carbon emissions from group delivery trips:

[0056] Carbon emissions from group delivery trips Equal to the correction factor Cumulative carbon emissions of the trajectory segments corresponding to all group delivery orders included in the group delivery trip The product of, that is:

[0057]

[0058] Carbon emission reduction of group-order delivery trips: First, calculate the median of their carbon emissions as the basis for calculating the carbon emission reduction of group-order delivery trips, expressed as follows:

[0059]

[0060] In the formula, is the median of carbon emissions of all individual delivery trips replaced by group delivery order i. median{} represents the median operation, which represents the baseline carbon emissions of individual delivery trips replaced by group delivery order i. k and D k represent the grids where the starting point and the end point of a single delivery trip k are located, O i and D i They represent the grids where the starting point and the end point of group delivery order i are located respectively;

[0061] Carbon emission reduction of group delivery trip j Equal to the sum of the baseline carbon emissions of the individual delivery trips replaced by all group delivery orders i included in the group delivery trip Actual carbon emissions from group-order delivery trips The difference, that is:

[0062]

[0063] Carbon emission reduction per unit distance of group delivery trip:

[0064] Carbon emission reduction per unit distance of group delivery trip j Carbon emissions reduction for group delivery trips Distance from group delivery trips The ratio is calculated as follows:

[0065]

[0066] Delay time and detour distance for group delivery:

[0067] The median travel time and distance of alternative individual delivery orders are calculated as a benchmark for calculating detours and delays in drone group delivery, which are:

[0068]

[0069] In the formula, is the median travel time of all individual delivery trips replaced by group delivery order i, representing the benchmark travel time of the individual delivery trips replaced by group delivery order i; is the median of the travel distances of all individual delivery trips replaced by group delivery order i, representing the benchmark travel distance of the individual delivery trips replaced by group delivery order i;

[0070] Delay time of group delivery trip j Equal to the actual travel time of the group delivery trip The sum of the benchmark travel times of the individual delivery trips replaced by all group delivery orders i included in the group delivery trip The difference, that is:

[0071]

[0072] Detour distance of group delivery trip j Equal to the actual travel distance of the group delivery trip The sum of the benchmark travel distances of the individual delivery trips replaced by all group delivery orders i included in the group delivery trip The difference, that is:

[0073]

[0074] Shared delivery trips share time and distance:

[0075] Shared time of group delivery trip j The length of the overlapping time period of all group delivery orders included in the group delivery trip:

[0076]

[0077] Shared distance of group delivery trip j It is the product of the cumulative length of all trajectory segments of the group delivery trip during the shared time and the group delivery trip correction coefficient:

[0078]

[0079] Handling abnormal values ​​in group delivery trips

[0080] Abnormal group-order delivery trips with serious missing track segments are cleared, and only shared trips with a correction coefficient less than or equal to 1.1 are retained. Abnormal group-order delivery trips with a shared time of less than 1 minute and a shared distance of less than 500m are cleared;

[0081] Delivery time for group orders

[0082] Order duration T for group delivery orders i n Equal to the end billing time ET of the order i n and start billing time ST i n The difference, that is:

[0083] T i n =ET i n -ST i n

[0084] Travel distance for group delivery orders

[0085] Travel distance of drone n for group delivery order i in group delivery trip j Equal to the group order delivery trip correction factor The sum of the lengths of all trajectory segments of the drone during the order period The product of, that is:

[0086]

[0087] Carbon emissions from group delivery orders

[0088] Carbon emissions from group delivery orders Equal to the group order delivery trip correction factor The cumulative emissions of all trajectory segments of the drone during the order period The product of, that is:

[0089]

[0090] In combination with the first aspect, in some implementations of the first aspect, the method also includes: the construction of variable characteristics based on group delivery trip data: by analyzing drone delivery orders and trajectory data, and combining POI point of interest data, road network data and land use data, the characteristic variables of the drone group delivery travel emission reduction benefit prediction model are constructed from three aspects: time characteristics, spatial characteristics, and travel characteristics.

[0091] In combination with the first aspect, in certain implementations of the first aspect, the method also includes: the characteristic variables of the drone group delivery travel emission reduction benefit prediction model include: emission reduction per unit distance of group delivery trip, peak hours, traffic conditions, land use mix, distance to the CBD, POI density, bus and subway station density, road network density, number of group deliveries, trip distance, average trip speed, shared distance ratio, detour rate and maximum and minimum order distance ratio.

[0092] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the process of constructing variable features based on the group-order delivery trip data includes:

[0093] The constructed features are screened using a multi-criteria integrated feature selection method to determine the characteristic variables that affect the prediction of carbon emission reduction benefits of drone group delivery travel; among them, the method selects the recursive feature elimination method in the parcel feature selection method and the elastic network method and random forest method in the embedded feature selection method;

[0094] Feature selection is performed on the characteristic variables at the three levels of time, space, and travel, and the features with average standardized scores less than the preset threshold, including peak hours, distance from the city center, density of bus and subway stations, and maximum and minimum order distance ratio, are eliminated. The remaining features with average standardized scores greater than the preset threshold are retained, and finally the characteristic variables of the drone group delivery travel emission reduction benefit prediction model are obtained.

[0095] In combination with the first aspect, in some implementations of the first aspect, the method further includes: interpreting and analyzing the carbon emission reduction benefit prediction results using a SHAP model interpretation method, and proposing a drone group-order delivery travel emission reduction strategy based on the analysis results;

[0096] The process includes: global model interpretation analysis, single-factor model interpretation analysis, two-factor interaction model interpretation analysis, individual model interpretation analysis, and based on the model interpretation results, proposing corresponding drone group delivery travel emission reduction strategies.

[0097] In the second aspect, in order to achieve the above-mentioned purpose, the present invention discloses a carbon emission reduction benefit prediction system for drone multi-order delivery, comprising:

[0098] The data processing module is used to obtain drone delivery data, process the drone delivery data, separate individual delivery orders and group delivery orders, and use them as group delivery trip data;

[0099] The feature construction module is used to construct variable features based on the group-order delivery trip data, and generate characteristic variables of the drone group-order delivery travel emission reduction benefit prediction model, where the construction of variable features includes three levels: time, space, and travel;

[0100] The benefit prediction module is used to input the characteristic variables of the drone group delivery emission reduction benefit prediction model into a pre-established optimization prediction model, and output the carbon emission reduction benefit prediction results, wherein the pre-established optimization prediction model is a gradient boosting model obtained by optimizing key hyperparameters through the Bayesian hyperparameter optimization method.

[0101] Beneficial effects of the present invention:

[0102] The present invention can deeply evaluate the specific impact of individual group-order delivery trips on the urban traffic environment, thereby providing more accurate decision-making support for urban traffic management and planning. The present invention comprehensively considers the influencing factors of multiple dimensions such as time, space and travel mode, and comprehensively analyzes the carbon emission reduction benefits of drone group-order delivery travel by constructing characteristic variables and applying multi-criteria feature selection methods, thereby improving the comprehensiveness and meticulousness of the evaluation. By using Bayesian hyperparameter optimization technology to tune the gradient boosting model, the present invention significantly improves the accuracy of the prediction of the carbon emission reduction benefits of drone group-order delivery travel and the generalization ability of the model. Using the SHAP model interpretation method, the present invention can clearly show the contribution of each influencing factor to the prediction result, enhance the interpretability of the model, and make the prediction result easier to understand and accept. The present invention conducts innovative research on the carbon emission reduction benefits of drone group-order delivery travel, especially in dealing with the complexity brought about by the inconsistency between the starting and ending points of drones and ordering users, showing high research value and application potential. The present invention establishes an environmental cost function that conforms to the characteristics of urban shared transportation, quantifies the impact of different travel modes on the urban traffic environment, and provides a new quantitative tool for evaluating and improving the urban traffic environment. By clarifying the environmental impact of ride-hailing drone group delivery, the present invention helps relevant departments take effective measures to promote the development of urban transportation in a greener and more sustainable direction. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0104] Figure 1 It is a schematic flow chart of the method of the present invention;

[0105] Figure 2 It is a schematic diagram of a group delivery itinerary including two group delivery orders of the present invention;

[0106] Figure 3 is a schematic diagram of a group order delivery itinerary including three group order delivery orders of the present invention;

[0107] Figure 4 It is a schematic diagram of a group delivery itinerary of the present invention and a separate delivery itinerary that it replaces;

[0108] Figure 5 It is a schematic diagram of the research scope of the present invention;

[0109] Figure 6 It is a schematic diagram of the multi-criteria integrated feature selection process of the present invention;

[0110] Figure 7 It is a schematic diagram of the framework of the carbon emission reduction benefit prediction model for the group-order delivery trip of the present invention;

[0111] Figure 8 It is a visualization diagram of evaluation indicators of different models of the present invention;

[0112] Fig. 9 It is a schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0113] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0114] Embodiment 1:

[0115] like Figure 1 As shown, a method for predicting carbon emission reduction benefits of drone multi-order delivery includes the following steps:

[0116] S101: Acquire drone delivery data, process the drone delivery data, separate individual delivery orders and group delivery orders, and use them as group delivery trip data;

[0117] The process of processing drone delivery data includes: data cleaning, group delivery order identification, indicator calculation and outlier processing;

[0118] Step 1: Data cleaning

[0119] Before identifying orders for group delivery and calculating indicators, the following types of data need to be cleaned:

[0120] (1) Longitude and latitude out of range data

[0121] Delete the order data and its corresponding trajectory data whose latitude and longitude of the order pickup location or the goods delivery location are out of range, including some erroneous data with negative longitude and latitude.

[0122] (2) Repeated and redundant data

[0123] The duplicate and redundant data in both the order dataset and the trajectory dataset are removed, and only the first and last records of research significance are retained in chronological order.

[0124] (3) Unreasonable order duration data

[0125] The order duration of each order can be obtained through the start billing time and the end billing time of the order data, and the order data with unreasonable order duration and its corresponding trajectory data are deleted. In the present invention, orders with an order duration of less than one minute or more than two hours are eliminated.

[0126] Step 2: Identify group delivery orders

[0127] The group delivery orders and individual delivery orders in the order data set are distinguished by the group delivery trip identification algorithm, that is, two adjacent orders of the same drone (i.e. the same drone) are sorted in chronological order. If the time periods of the two orders overlap, that is, the end billing time of the first order is later than the start billing time (ET) of the second order, the group delivery orders and individual delivery orders in the order data set are distinguished. 1 >ST 2 ), then these two orders are group delivery orders, and these two group delivery orders constitute a group delivery trip. Figure 2 As shown, there are two possible scenarios for a group delivery trip with a group delivery number of 2.

[0128] For a group delivery trip that includes three group delivery orders, the situation is more diverse, but can still be converted into the above situation. If the latest end billing time of the two group delivery orders that have been identified as the same group delivery trip is later than the start billing time of the third order (max(ET 1 ,ET 2 )>ST 3), the third order will be added to this group delivery itinerary, and the group delivery quantity of the group delivery itinerary will also change from 2 to 3. Figure 3 As shown, the figure illustrates one of the scenarios (Scenario 3) of a group delivery trip with 3 group deliveries.

[0129] Step 3: Indicator calculation and outlier processing

[0130] If the dataset does not directly provide relevant information such as emissions, speed, and travel distance, it is necessary to calculate indicators in order to extract appropriate features for subsequent feature analysis and prediction model construction. In addition, since trajectory data may be missing or abnormal, indicator calculation can also be used to find outliers in the data and process them to improve the accuracy and reliability of the data and ensure the quality of the data used. Indicator calculations are performed in the order of trajectory segments, individual delivery trips, group delivery trips, and group delivery orders, starting from the trajectory points.

[0131] ① Calculation of trajectory segment index

[0132] Connect two consecutive GPS track points and convert them into track segments. According to the acquisition timestamps and longitude and latitude of the two track points, the interval time and distance of each track segment can be calculated, so as to obtain the average speed of each track segment, and the carbon emission factor and carbon emissions can be calculated based on the average speed. Finally, the abnormal track segments are cleared according to the interval time, distance and average speed of the track segments.

[0133] (1) Track segment interval length

[0134] Assume that the trajectory segment s of drone n in order i is formed by connecting two consecutive trajectory points m, m+1 of drone n in order i, then the interval length of the trajectory segment is It can be expressed as:

[0135]

[0136] In the formula, geodesic is a function in Python that calculates the distance between two locations. is the location information of the trajectory point m of drone n in order i, including the longitude and latitude of the trajectory point.

[0137] (2) Track segment interval time

[0138] At this time, the starting time of trajectory segment s The timestamp of trajectory point m End time of trajectory segment s That is the timestamp of trajectory point m+1 Track segment interval time The end time of the track segment and starting time The differences are:

[0139]

[0140] (3) Average speed of trajectory segment

[0141] Average speed of trajectory segment Equal to the interval length of the track segment With interval time The ratio is expressed as:

[0142]

[0143] (4) Carbon emission factor of trajectory segment

[0144] The research object selected by this method is a small passenger car using gasoline, and the emission factor of the trajectory segment is Expressed as:

[0145]

[0146] Where α, β, γ, δ, ε, ζ, and η are the relevant parameters for calculating the carbon emission factor in the COPERT model. When the emission standard is Euro IV, the fuel type is gasoline, and the vehicle type is a small passenger car, their corresponding values ​​are shown in Table 1.

[0147] Table 1 COPERT model related parameter values

[0148]

[0149] In practical applications, if the vehicle type is an electric vehicle or other new energy vehicle, the carbon emission factor calculation function can be adjusted according to the different types of vehicles. Then, based on this method, the carbon emissions corresponding to different vehicle types can be obtained, and then the emission reduction benefits of drone group delivery under different vehicle types can be analyzed.

[0150] (5) Carbon emissions of trajectory segment

[0151] Carbon emissions of trajectory segments Emission factor equal to the trajectory segment Distance between track segments The product of, that is:

[0152]

[0153] (6) Trajectory segment outlier processing

[0154] To avoid errors, abnormal track segments with an interval time of more than 60s and an interval distance not equal to 0, or an interval distance of more than 500m, or an average speed of more than 120km / h are cleared. The abnormal track segments account for about 0.18% of the total track segments.

[0155] ② Calculation of separate delivery trip indicators

[0156] The main indicators of a single delivery trip include the single delivery trip time, the single delivery trip distance, and the single delivery trip carbon emissions, which can be calculated through the corresponding single delivery orders and their trajectory segments.

[0157] (1) Single delivery journey time

[0158] For a single delivery trip k of drone n, the travel time of single delivery trip k is Equal to the order duration of the corresponding individual delivery order i, that is, the end billing time ET of the individual delivery order i i n and start billing time ST i n The difference is:

[0159]

[0160] (2) Single delivery trip distance

[0161] Introducing correction factor It is equal to the actual travel time of the individual delivery trip The cumulative duration of the trajectory segment corresponding to the individual delivery order i The ratio of, that is:

[0162]

[0163] Travel distance for individual delivery trips Equal to the correction factor The cumulative length of the trajectory segments corresponding to the individual delivery orders The product of, that is:

[0164]

[0165] (3) Carbon emissions from a single delivery trip

[0166] Carbon emissions from a single delivery trip Equal to the correction factor Cumulative carbon emissions of the trajectory segments corresponding to individual delivery orders The product of, that is:

[0167]

[0168] (4) Handling of abnormal values ​​for individual delivery trips

[0169] The abnormal individual delivery trips with serious missing track segments are cleared and only the correction coefficient is retained For separate delivery trips with a rate less than or equal to 1.1, abnormal separate delivery trips account for approximately 3.5% of the total number of separate delivery trips.

[0170] ③Calculation of group delivery itinerary indicators

[0171] The main indicators of group delivery trips include group delivery trip time, group delivery trip distance, group delivery trip carbon emissions, group delivery trip carbon emission reduction, delay time and detour distance, sharing time and sharing distance, which can be calculated through the corresponding group delivery orders, trajectory segments and the individual delivery trips replaced by the group delivery trip.

[0172] (1) Group delivery travel time

[0173] For group delivery trip j of drone n, the travel time of group delivery trip j is Equal to the set of group delivery orders it contains Φ j The latest end billing time max(ET i n ) and the earliest billing time min(ST i n ), that is:

[0174]

[0175] (2) Distance of group delivery

[0176] Similar to individual delivery trips, the correction factor for group delivery trips Equal to the travel time of the group delivery trip The cumulative duration of the track segments corresponding to all group delivery orders included in the group delivery trip The product of, that is:

[0177]

[0178] Travel distance for group delivery trips Equal to the correction factor The cumulative length of the track segments corresponding to all group delivery orders included in the group delivery trip The product of, that is:

[0179]

[0180] (3) Carbon emissions from group-order delivery trips

[0181] Carbon emissions from group delivery trips Equal to the correction factor Cumulative carbon emissions of the trajectory segments corresponding to all group delivery orders included in the group delivery trip The product of, that is:

[0182]

[0183] (4) Carbon emissions reduction from group-order delivery trips

[0184] like Figure 4 As shown in the figure, a spatial aggregation method is used to find possible replacement single delivery trips from the existing order data set. The study area is gridded with 500m×500m as a grid unit, so that the study area can be divided into 289 grids. The start and end OD pairs of the single delivery trip and the group delivery order are aggregated into the grid. When the grid where the start and end of the single delivery trip are located is the same as the grid where the start and end of the group delivery order are located, the single delivery trip is regarded as a single delivery trip replaced by the group delivery order. Assuming that a group delivery trip consists of N group delivery orders, it can replace N single delivery trips with the same OD.

[0185] In order to represent the general characteristics of these alternative individual delivery trips, their median carbon emissions are calculated as a benchmark for calculating the carbon emission reduction of group delivery trips, expressed as:

[0186]

[0187] In the formula, is the median of carbon emissions of all individual delivery trips replaced by group delivery order i. median{} represents the median operation, which represents the baseline carbon emissions of individual delivery trips replaced by group delivery order i. k and D k represent the grids where the starting point and the end point of a single delivery trip k are located, O i and D i They respectively represent the starting and ending grids of group delivery order i.

[0188] Carbon emission reduction of group delivery trip j Equal to the sum of the baseline carbon emissions of the individual delivery trips replaced by all group delivery orders i included in the group delivery trip Actual carbon emissions from group-order delivery trips The difference, that is:

[0189]

[0190] (5) Carbon emissions reduction per unit distance of group-order delivery trips

[0191] Carbon emission reduction per unit distance of group delivery trip j Carbon emissions reduction for group delivery trips Distance from group delivery trips The ratio measures the amount of carbon dioxide emissions saved per kilometer through group delivery. In this method, it is used to evaluate the carbon emission reduction benefits of group delivery trips. The calculation formula is as follows:

[0192]

[0193] (6) Delay time and detour distance for group delivery

[0194] Similarly, the median travel time and distance of these alternative individual delivery orders are calculated as a benchmark for calculating detours and delays in drone group delivery, which are:

[0195]

[0196] In the formula, is the median travel time of all individual delivery trips replaced by group delivery order i, representing the benchmark travel time of the individual delivery trips replaced by group delivery order i; It is the median of the travel distances of all individual delivery trips replaced by group delivery order i, representing the benchmark travel distance of the individual delivery trips replaced by group delivery order i.

[0197] Delay time of group delivery trip j Equal to the actual travel time of the group delivery trip The sum of the benchmark travel times of the individual delivery trips replaced by all group delivery orders i included in the group delivery trip The difference, that is:

[0198]

[0199] Detour distance of group delivery trip j Equal to the actual travel distance of the group delivery trip The sum of the benchmark travel distances of the individual delivery trips replaced by all group delivery orders i included in the group delivery trip The difference, that is:

[0200]

[0201] (7) Shared delivery trips share time and distance

[0202] Shared time of group delivery trip j It is the length of the overlapping time period of all group delivery orders included in the group delivery trip. Taking a group delivery trip containing two group delivery orders as an example, the shared time of the group delivery trip is the order end time min (ET) of the first group delivery order. i n ) and the order start time of the second group delivery order max(ST i n ) is the difference between .

[0203]

[0204] Shared distance of group delivery trip j It is the product of the cumulative length of all trajectory segments of the group delivery trip in the shared time and the group delivery trip correction coefficient. Taking the group delivery trip containing two group delivery orders as an example, the shared distance of the group delivery trip is the order start time max(ST i n ) and the order end time min(ET i n ) is the cumulative length of the trajectory segments between .

[0205]

[0206] (8) Handling abnormal values ​​in group delivery trips

[0207] Abnormal group delivery trips with serious missing trajectory segments were cleared, and only the shared trips with a correction coefficient less than or equal to 1.1 were retained. Abnormal group delivery trips with a sharing time of less than 1 minute and a sharing distance of less than 500m were also cleared. Abnormal group delivery trips accounted for about 3.9% of the total group delivery trips.

[0208] ④ Calculation of drone group-order delivery travel order indicators

[0209] The relevant indicators of group delivery orders are calculated through the group delivery itinerary. The main indicators of group delivery orders include the duration of group delivery orders, the travel distance of group delivery orders, and the carbon emissions of group delivery orders.

[0210] (1) Delivery time for group orders

[0211] For group delivery orders, the order duration T is the same as for individual delivery orders. i n Equal to the end billing time ET of the order i n and start billing time ST i n The difference, that is:

[0212] T i n =ET i n -ST i n

[0213] (2) Travel distance for group delivery orders

[0214] Travel distance of drone n for group delivery order i in group delivery trip j Equal to the group order delivery trip correction factor The sum of the lengths of all trajectory segments of the drone during the order period The product of, that is:

[0215]

[0216] (3) Carbon emissions from group delivery orders

[0217] Carbon emissions from group delivery orders Equal to the group order delivery trip correction factor The cumulative emissions of all trajectory segments of the drone during the order period The product of, that is:

[0218]

[0219] S102: constructing variable features based on the group-order delivery trip data to generate characteristic variables of the drone group-order delivery travel emission reduction benefit prediction model, wherein the construction of variable features includes three levels: time, space, and travel;

[0220] Construction of variable characteristics based on group delivery trip data: By analyzing drone delivery orders and trajectory data, and combining POI point of interest data, road network data and land use data, the characteristic variables of the drone group delivery travel emission reduction benefit prediction model are constructed from three aspects: time characteristics, spatial characteristics, and travel characteristics.

[0221] Time characteristics refer to characteristics related to the travel period of drone group delivery. According to the results of time characteristic analysis, in this method, the constructed time feature is whether the start time of the group delivery trip is during the lunch rush hour (13-14 o'clock) or the evening rush hour (17-18 o'clock).

[0222] Spatial characteristics refer to the characteristics related to the geographical space through which the drone group delivery travel passes. According to the results of spatial characteristic analysis, in this method, the constructed spatial characteristics are the weighted average of the traffic operation conditions and built environment factors of all grids through which the group delivery travel trajectory segment passes, including the current hourly average vehicle speed, land use mixture, distance to the CBD, POI density, bus and subway station density, and the total density of the road network such as expressways and trunk roads of all grids through which the group delivery travel trajectory segment passes, representing the overall situation of the space passed by the group delivery travel route.

[0223] Travel characteristics refer to characteristics related to the drone group delivery trip itself. According to the results of travel characteristics analysis, there are serious multicollinearity problems between the detour rate and the delay rate, between the travel distance and the travel time, between the shared distance ratio and the shared time ratio, and between the maximum and minimum order distance ratio and the maximum and minimum order time ratio, which may reduce the stability, reliability and interpretability of the model.

[0224] The characteristic variables of the drone group delivery travel emission reduction benefit prediction model include: emission reduction per unit distance of group delivery trip (g / km), peak hours, traffic operation conditions (km / h), land use mix, distance to CBD (m), POI density (units / km2), bus and subway station density (units / km2), road network density (km / km2), number of group deliveries, trip distance (km), average trip speed (km / h), sharing distance ratio, detour rate and maximum and minimum order distance ratio.

[0225] The process of constructing variable features based on group delivery trip data includes:

[0226] The constructed features are screened using a multi-criteria integrated feature selection method to determine the characteristic variables that have an important impact on the prediction of carbon emission reduction benefits of drone group delivery travel;

[0227] Specifically, the method body is to select the recursive feature elimination method in the wrapper feature selection method and the elastic network method and random forest method in the embedded feature selection method;

[0228] Feature selection is performed on the characteristic variables at the three levels of time, space, and travel, and the features with average standardized scores less than the preset threshold of 0.1, including peak hours, distance from the city center, density of bus and subway stations, and maximum and minimum order distance ratio, are eliminated. The remaining features with average standardized scores greater than the preset threshold of 0.1 are retained. Finally, the characteristic variables of the drone group delivery travel emission reduction benefit prediction model are obtained and substituted into the final prediction model for training.

[0229] S103: Input the characteristic variables of the drone group delivery emission reduction benefit prediction model into a pre-established optimization prediction model, and output the carbon emission reduction benefit prediction result, wherein the pre-established optimization prediction model is a gradient boosting model obtained by optimizing key hyperparameters through the Bayesian hyperparameter optimization method.

[0230] Specifically, one or more of GBDT, XGBoost, LightGBM, and CatBoost are used as prediction models;

[0231] The sample data is divided into training set and test set in the ratio of 8:2, and the 5-fold cross-validation method is used to construct four gradient boosting models: GBDT, XGBoost, LightGBM, and CatBoost.

[0232] The Bayesian hyperparameter optimization method is used to optimize the key hyperparameters of the selected model, such as the number of trees, learning rate, and maximum depth of the tree.

[0233] The data set was cross-validated 5-fold, and based on the Bayesian hyperparameter optimization method, the three key hyperparameters of the number of trees, learning rate, and maximum tree depth of the four gradient boosting models were optimized with the goal of maximizing R2. The number of trees ranged from [50, 1000], the learning rate ranged from [0.005, 0.5], and the maximum tree depth ranged from [3, 8]. The performance of the previous model was compared with the parameter optimization. Finally, the performance parameters of all models were compared, and the model with the best performance was selected as the emission reduction prediction model for group delivery trips of this method.

[0234] The carbon emission reduction benefit prediction results are interpreted and analyzed using the SHAP model interpretation method, and a drone group delivery travel emission reduction strategy is proposed based on the analysis results;

[0235] The process includes: global model interpretation analysis, single-factor model interpretation analysis, two-factor interaction model interpretation analysis, individual model interpretation analysis, and based on the model interpretation results, proposing corresponding drone group delivery travel emission reduction strategies.

[0236] Step 1: Global model interpretation and analysis

[0237] First, a global model interpretation analysis is performed to evaluate the overall impact of each feature on the carbon emission reduction benefits of drone group delivery by calculating the average SHAP value of all features. The absolute value of the SHAP average of each feature is output as a histogram and sorted by size to indicate the importance of the feature. Based on the SHAP values ​​of all test samples, a density scatter plot is drawn to further explore the direction of each feature's effect on the emission reduction per unit distance of group delivery trips.

[0238] Step 2: Model interpretation and analysis based on single factor

[0239] The single-factor model interpretation analysis shows the marginal effect of a single feature on the model prediction results by fixing the values ​​of other features. After obtaining the influence of each feature variable on the emission reduction per unit distance of group delivery trips, further research is conducted on the specific impact of a single feature on the emission reduction per unit distance of group delivery trips in order of feature importance. A single-factor dependency graph is constructed, and all test samples are visualized.

[0240] Step 3: Model interpretation and analysis based on two-factor interaction

[0241] Based on the consideration of feature interactions, this method uses a two-factor interaction model to interpret and analyze, identifying possible interactions between features and their joint impact on the carbon emission reduction benefits of drone group delivery. The two travel characteristics of detour rate and trip distance are coupled with the number of group deliveries, and a two-factor dependency graph is constructed to analyze the impact of two-factor interactions on the SHAP value by example.

[0242] Step 4: Individual-based model interpretation analysis

[0243] The individual-based model interpretation analysis provides a detailed explanation of each group-delivery trip sample. Through the individual contribution graph of the SHAP value, it is possible to identify the features that have a significant impact on the emission reduction benefits of a specific group-delivery trip and customize emission reduction strategies for each trip.

[0244] Two samples are randomly selected from each of the emission-reduced group-batch delivery trips and the non-emission-reduced group-batch delivery trips, and the prediction process of these samples is visualized through the SHAP waterfall chart, so as to more deeply analyze the prediction process of each sample and the key factors affecting the model decision.

[0245] Step 5: Based on the model interpretation results, a corresponding drone group delivery travel emission reduction strategy is proposed

[0246] Specifically, the present invention is further described below by way of embodiments:

[0247] The drone delivery dataset used in this method comes from the Gaia data open plan of a certain software. Due to data privacy protection, only the trajectory data of part of the second ring road in a certain city center is open. The coordinate system used by the Gaia data open plan is the GCJ-02 Mars coordinate system. In order to correctly project the data to the corresponding geographic space and ensure the accuracy of subsequent spatial analysis, it is converted to the WGS84 geographic coordinate system. The longitude and latitude ranges of the full-month trajectory dataset are counted, and their intersection is calculated. The results show that the trajectory data area is between 104.04° and 104.12° east longitude and 30.66° and 30.73° north latitude in the WGS84 coordinate system. Therefore, this area is used as the research scope of this method. Figure 5 The area is approximately 6.5×6.5km2.

[0248] The drone delivery dataset used in this example consists of an online car-hailing order dataset and an online car-hailing trajectory dataset. The two datasets can be linked through the order ID. The dataset covers more than six million Didi drone deliveries in a certain city from November 1 to November 30, 2016.

[0249] The order dataset records the start and end time of each travel order and the location where the goods were delivered. The fields of the order dataset include order ID, start billing time, end billing time, longitude of the order pickup location, latitude of the order pickup location, longitude of the goods delivery location, and latitude of the goods delivery location. Specific examples of the order dataset are shown in Tables 2 and 3.

[0250] The average sampling interval of the trajectory dataset is about 3 seconds, which records the GPS location points of each order at different times during the journey. The fields of the trajectory dataset include drone ID, order ID, GPS time, longitude of trajectory point, and latitude of trajectory point. Specific examples of the trajectory dataset are shown in Tables 4 and 5.

[0251] Table 2 Drone order data structure

[0252]

[0253]

[0254] Table 3. Sample drone order data

[0255]

[0256] Table 4 UAV trajectory data structure

[0257]

[0258] Table 5. Examples of drone trajectory data

[0259]

[0260] After the above preprocessing steps, the number of valid orders screened out is 1,178,811, of which 1,148,274 are single delivery orders, accounting for 97.4% of the total orders, and 30,537 are group delivery orders, accounting for only 2.6% of the total orders, which indicates that the utilization rate of group delivery service is low and the attractiveness of group delivery service needs to be further improved. There are a total of 14,929 group delivery trips, of which 14,250 include two group delivery orders, accounting for 95.5% of the total number of group delivery trips, and 679 include three group delivery orders, accounting for 4.5% of the total number of group delivery trips, which indicates that the number of passengers carried by group delivery service is generally low and the passenger carrying efficiency of group delivery service needs to be improved.

[0261] The original data set does not directly provide order type (i.e., whether the order is a group delivery order) and related information such as emissions, speed, and travel distance. Therefore, steps such as group delivery order identification and indicator calculation are required to extract appropriate features for subsequent modeling. In addition, due to factors such as data privacy protection, interference from buildings and weather, and equipment and network failures, online car-hailing trajectory data may be missing, redundant, and abnormal. In order to ensure the accuracy and reliability of the data, steps such as data screening, deduplication, and outlier processing are also required during the preprocessing process.

[0262] After the above preprocessing steps, the number of valid orders screened out is 1,178,811, of which 1,148,274 are single delivery orders, accounting for 97.4% of the total orders, and 30,537 are group delivery orders, accounting for only 2.6% of the total orders, which indicates that the utilization rate of group delivery service is low and the attractiveness of group delivery service needs to be further improved. There are a total of 14,929 group delivery trips, of which 14,250 include two group delivery orders, accounting for 95.5% of the total number of group delivery trips, and 679 include three group delivery orders, accounting for 4.5% of the total number of group delivery trips, which indicates that the number of passengers carried by group delivery service is generally low and the passenger carrying efficiency of group delivery service needs to be improved.

[0263] Table 6 Proportion of order types

[0264]

[0265] Table 7 Proportion of group-order delivery trip types

[0266]

[0267] The constructed features are screened using a multi-criteria integrated feature selection method to determine the characteristic variables that have an important impact on the prediction of the carbon emission reduction benefits of drone group delivery travel.

[0268] Specifically, the recursive feature elimination method in the package feature selection method and the elastic network method and random forest method in the embedded feature selection method are selected. The multi-criteria integrated feature selection process is as follows Figure 6 shown.

[0269] Based on the multi-criteria integrated feature selection method proposed above, including recursive feature elimination, elastic network method, and random forest algorithm, feature selection is performed on the feature variables at the three levels of time, space, and travel selected by this method. The results are shown in Table 8 Multi-criteria Integrated Feature Selection Results. Too many features may not only increase the training time and prediction time of the model, but also increase the complexity and overfitting risk of the model. In order to select the most relevant features to improve model performance and generalization ability, the peak hours, distance from the city center, bus and subway station density, and maximum and minimum order distance ratio, which have an average standardized score of less than 0.1, are eliminated. The remaining features with an average standardized score greater than 0.1 are retained and substituted into the final prediction model for training.

[0270] Table 8 Multi-criteria integrated feature selection results

[0271]

[0272]

[0273] The sample data was divided into training set and test set in a ratio of 8:2, and four gradient boosting models, GBDT, XGBoost, LightGBM, and CatBoost, were constructed using a 5-fold cross-validation method. Then, the data set was cross-validated 5-fold, and based on the Bayesian hyperparameter optimization method, the three key hyperparameters of the number of trees, learning rate, and maximum tree depth of the four gradient boosting models were optimized with the goal of maximizing R2. The number of trees ranged from [50, 1000], the learning rate ranged from [0.005, 0.5], and the maximum tree depth ranged from [3, 8]. The performance of the previous model was compared after parameter optimization. Finally, the performance parameters of all models were compared, and the model with the best performance was selected as the emission reduction prediction model for group delivery trips of this method.

[0274] 4) Carbon emission reduction benefit prediction: Use the optimized model to predict the carbon emission reduction benefits of group-order delivery trips;

[0275] The evaluation indicators of all models are shown in Table 9 and Figure 8 , from Table 9 and Figure 8It can be seen that among all the models, BO-CatBoost has the highest R2 of 0.9531, and the MSE, RMSE, and MAE are the lowest among all models, which are 88.6276, 9.4121, and 7.1680, respectively. That is, compared with other models, BO-CatBoost has the best prediction performance. Therefore, BO-CatBoost is selected as the emission reduction benefit prediction model for group delivery trips in this method.

[0276] Table 9 Comparison of evaluation indicators of different models

[0277]

[0278] Model interpretation and strategy formulation: The SHAP model interpretation method is used to interpret and analyze the prediction results, and a drone group delivery travel emission reduction strategy is proposed.

[0279] Through the SHAP-based model interpretation method, the impact of the relevant characteristics of group-order delivery trips on the emission reduction benefits of group-order delivery trips was explored from the global to the individual, from single-factor analysis to two-factor interaction analysis. Based on the analysis results of the previous steps, this section summarizes the emission reduction strategies for drone group-order delivery trips.

[0280] (1) Optimize the route planning algorithm for group delivery to reduce the detour rate and increase the shared distance ratio

[0281] The detour rate is the most important factor affecting the emission reduction benefits of group-order delivery trips. An increase in the detour rate will reduce the carbon emission reduction benefits of group-order delivery trips. When there are no detours in group-order delivery trips, it may still have a negative effect on the emission reduction benefits of group-order delivery trips. In addition, when the sharing distance is relatively small, it may also have a negative effect on the carbon emission reduction benefits of group-order delivery trips. Therefore, the drone delivery platform should optimize the group-order delivery route planning algorithm, minimize the detour distance of group-order delivery trips and increase the route overlap rate of group-order delivery users, so that the detour rate is less than -0.2 and the sharing distance ratio is greater than 0.7, so as to improve the emission reduction benefits of group-order delivery trips.

[0282] (2) Encourage short-distance drone group delivery and non-peak drone group delivery

[0283] When the travel distance is less than 6km, the corresponding SHAP value is positive, which has a positive effect on the emission reduction per unit distance of the group delivery trip. The proportion of emission-reducing group delivery trips and the emission reduction per unit distance of group delivery trips during non-peak hours are relatively high. Therefore, short-distance drone group delivery trips and non-peak drone group delivery trips should be encouraged to improve the emission reduction benefits of group delivery trips.

[0284] (3) Choose a reasonable itinerary

[0285] The POI density, traffic conditions, road network density, and land use mix of the area where the group-order delivery trip passes through have a strong linear relationship with the carbon emission reduction benefits of the group-order delivery trip. When the POI density of the area where the group-order delivery trip passes through is greater than 1250 / km2, the average vehicle speed is greater than 37.5km / h, the road network density is less than 6km / km2, and the land use mix is ​​greater than 0.6, the corresponding SHAP value is positive, which has a positive effect on the emission reduction per unit distance of the group-order delivery trip. Therefore, when choosing a route for a group-order delivery trip, you should try to pass through areas with good traffic conditions, high POI density and land use mix, and low road network density.

[0286] (4) Optimize the matching algorithm for group order delivery

[0287] The data preprocessing results show that 16.08% of group delivery trips still fail to achieve emission reduction effects. In addition, the number of group delivery orders only accounts for 2.6% of the total number of travel orders, and the number of group delivery trips containing three group delivery orders only accounts for 4.5% of the total number of group delivery trips. Therefore, drone group delivery travel still has great potential for emission reduction. Using the prediction model proposed by this method, when receiving a group delivery matching request, the drone delivery platform can immediately predict the potential carbon emission reduction of the group delivery trip, thereby avoiding matching group delivery trips that do not reduce emissions.

[0288] (5) Implement incentive policies for drone group delivery

[0289] The model of this method can also help calculate the carbon emission reduction of each drone group delivery trip. The government and drone delivery platforms can formulate relevant incentive policies based on the results of carbon emission reduction, such as imposing carbon taxes, conducting carbon emissions trading, providing financial subsidies and rewards, etc., so that more people can understand their own contributions to energy conservation and emission reduction, encourage more people to choose drone group delivery, and increase the utilization rate and emission reduction efficiency of drone group delivery travel services.

[0290] Embodiment 2: Fig. 9 As shown, a carbon emission reduction benefit prediction system for drone multi-order delivery is characterized by comprising:

[0291] The data processing module 11 is used to obtain drone delivery data, process the drone delivery data, separate individual delivery orders and group delivery orders, and use them as group delivery itinerary data;

[0292] The feature construction module 12 is used to construct variable features based on the group-order delivery trip data to generate feature variables of the drone group-order delivery travel emission reduction benefit prediction model, wherein the construction of variable features includes three levels: time, space and travel;

[0293] The benefit prediction module 13 is used to input the characteristic variables of the drone group delivery emission reduction benefit prediction model into a pre-established optimization prediction model, and output the carbon emission reduction benefit prediction result, wherein the pre-established optimization prediction model is a gradient boosting model obtained by optimizing key hyperparameters through the Bayesian hyperparameter optimization method.

[0294] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors, and a memory for storing one or more computer programs; the program includes program instructions, and the processor is used to execute the program instructions stored in the memory. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is used to implement one or more instructions, specifically for loading and executing one or more instructions in a computer storage medium to implement the above method.

[0295] It needs to be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium, on which a computer program is stored, and the computer program is executed by a processor to execute the above method. The storage medium can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0296] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0297] The above shows and describes the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited by the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present disclosure. Without departing from the spirit and scope of the present disclosure, the present disclosure may have various changes and improvements, and these changes and improvements fall within the scope of the present disclosure to be protected.

Claims

1. A method for predicting carbon emission reduction benefits of drone multi-order delivery, characterized in that: The method comprises the following steps: Obtain drone delivery data, process the drone delivery data, separate individual delivery orders and group delivery orders, and use them as group delivery trip data; Based on the group-order delivery trip data, the variable characteristics are constructed to generate the characteristic variables of the drone group-order delivery travel emission reduction benefit prediction model. The construction of variable characteristics includes three levels: time, space and travel. The characteristic variables of the drone group delivery emission reduction benefit prediction model are input into a pre-established optimization prediction model, and the carbon emission reduction benefit prediction result is output. The pre-established optimization prediction model is a gradient boosting model obtained by optimizing key hyperparameters through the Bayesian hyperparameter optimization method.

2. The method for predicting carbon emission reduction benefits of drone multi-order delivery according to claim 1 is characterized in that: The process of processing drone delivery data includes: data cleaning, group delivery order identification, indicator calculation and outlier processing.

3. The method for predicting carbon emission reduction benefits of drone multi-order delivery according to claim 2 is characterized in that: The process of identifying the group order delivery order: The group delivery orders and individual delivery orders in the order data set are distinguished through the group delivery itinerary identification algorithm, that is, two adjacent orders of the same drone sorted in chronological order: if the time periods of the two orders overlap, that is, the end billing time of the first order is later than the start billing time of the second order: ET1>ST2, then the two orders are group delivery orders, and the two group delivery orders constitute a group delivery itinerary.

4. The method for predicting carbon emission reduction benefits of drone multi-order delivery according to claim 2 is characterized in that: The index calculation includes the calculation of trajectory segment index, individual delivery trip index, group delivery trip index and drone group delivery travel order index.

5. The method for predicting carbon emission reduction benefits of drone multi-order delivery according to claim 4 is characterized in that: The trajectory segment index calculation includes: Track segment interval length: Assume that the trajectory segment s of drone n in order i is formed by connecting two consecutive trajectory points m and m+1 of drone n in order i, then the interval length of the trajectory segment is It is expressed as: In the formula, geodesic is a function in Python that calculates the distance between two locations. is the location information of trajectory point m of drone n in order i; Track segment interval time: The starting time of trajectory segment s The timestamp of trajectory point m End time of trajectory segment s That is the timestamp of trajectory point m+1 Track segment interval time The end time of the track segment and starting time The differences are: Average speed of trajectory segment: Average speed of trajectory segment Equal to the interval length of the track segment With interval time The ratio is expressed as: Carbon emission factor of trajectory segment It is expressed as: In the formula, α, β, γ, δ, ε, ζ, and η are the relevant parameters for calculating the carbon emission factor in the COPERT model; Carbon emissions per track segment: Carbon emissions of trajectory segments Emission factor equal to the trajectory segment Distance between track segments The product of, that is: Trajectory segment outlier processing: Through the trajectory preset conditions, the abnormal trajectory segments that do not meet the trajectory preset conditions are cleared; The calculation of individual delivery trip metrics includes: Separate delivery travel time: For a single delivery trip k of drone n, the travel time of single delivery trip k is Equal to the order duration of the corresponding individual delivery order i, that is, the end billing time ET of the individual delivery order i i n and start billing time ST i n Difference: Distance of single delivery trip: Introducing correction factor It is equal to the actual travel time of the individual delivery trip The cumulative duration of the trajectory segment corresponding to the individual delivery order i The ratio of, that is: Travel distance for individual delivery trips Equal to the correction factor The cumulative length of the trajectory segments corresponding to the individual delivery orders The product of, that is: Carbon emissions per delivery trip Carbon emissions from a single delivery trip Equal to the correction factor Cumulative carbon emissions of the trajectory segments corresponding to individual delivery orders The product of, that is: Handling of abnormal values ​​for individual delivery trips Clear abnormal individual delivery trips with serious missing track segments; The process of calculating the group order delivery itinerary index: Group delivery travel time: For group delivery trip j of drone n, the travel time of group delivery trip j is Equal to the set of group delivery orders it contains Φ j The latest end billing time max(ET i n ) and the earliest billing time min(ST i n ), that is: Distance of group delivery Correction factor for group delivery trips Equal to the travel time of the group delivery trip The cumulative duration of the track segments corresponding to all group delivery orders included in the group delivery trip The product of, that is: Travel distance for group delivery trips Equal to the correction factor The cumulative length of the track segments corresponding to all group delivery orders included in the group delivery trip The product of, that is: Carbon emissions from group delivery trips: Carbon emissions from group delivery trips Equal to the correction factor Cumulative carbon emissions of the trajectory segments corresponding to all group delivery orders included in the group delivery trip The product of, that is: Carbon emission reduction of group-order delivery trips: First, calculate the median of their carbon emissions as the basis for calculating the carbon emission reduction of group-order delivery trips, expressed as follows: In the formula, is the median of carbon emissions of all individual delivery trips replaced by group delivery order i. median{} represents the median operation, which represents the baseline carbon emissions of individual delivery trips replaced by group delivery order i. k and D k represent the grids where the starting point and the end point of a single delivery trip k are located, O i and D i They represent the grids where the starting point and the end point of group delivery order i are located respectively; Carbon emission reduction of group delivery trip j Equal to the sum of the baseline carbon emissions of the individual delivery trips replaced by all group delivery orders i included in the group delivery trip Actual carbon emissions from group-order delivery trips The difference, that is: Carbon emission reduction per unit distance of group delivery trip: Carbon emission reduction per unit distance of group delivery trip j Carbon emissions reduction for group delivery trips Distance from group delivery trips The ratio is calculated as follows: Delay time and detour distance for group delivery: The median travel time and distance of alternative individual delivery orders are calculated as a benchmark for calculating detours and delays in drone group delivery, which are: In the formula, is the median travel time of all individual delivery trips replaced by group delivery order i, representing the benchmark travel time of the individual delivery trips replaced by group delivery order i; is the median of the travel distances of all individual delivery trips replaced by group delivery order i, representing the benchmark travel distance of the individual delivery trips replaced by group delivery order i; Delay time of group delivery trip j Equal to the actual travel time of the group delivery trip The sum of the benchmark travel times of the individual delivery trips replaced by all group delivery orders i included in the group delivery trip The difference, that is: Detour distance of group delivery trip j Equal to the actual travel distance of the group delivery trip The sum of the benchmark travel distances of the individual delivery trips replaced by all group delivery orders i included in the group delivery trip The difference, that is: Shared delivery trips share time and distance: Shared time of group delivery trip j The length of the overlapping time period of all group delivery orders included in the group delivery trip: Shared distance of group delivery trip j It is the product of the cumulative length of all trajectory segments of the group delivery trip during the shared time and the group delivery trip correction coefficient: Handling abnormal values ​​in group delivery trips Set delivery preset conditions and clear group delivery trips that do not meet the delivery preset conditions; Delivery time for group orders Order duration T for group delivery orders i n Equal to the end billing time ET of the order i n and start billing time ST i n The difference, that is: T i n =ET i n -ST i n Travel distance for group delivery orders Travel distance of drone n for group delivery order i in group delivery trip j Equal to the group order delivery trip correction factor The sum of the lengths of all trajectory segments of the drone during the order period The product of, that is: Carbon emissions from group delivery orders Carbon emissions from group delivery orders Equal to the group order delivery trip correction factor Cumulative emissions from all trajectory segments of the drone during the order period The product of, that is:

6. The method for predicting carbon emission reduction benefits of drone multi-order delivery according to claim 1 is characterized in that: The variable characteristics are constructed based on the group delivery trip data: by analyzing the drone delivery orders and trajectory data, and combining the POI point of interest data, road network data and land use data, the characteristic variables of the drone group delivery travel emission reduction benefit prediction model are constructed from three aspects: time characteristics, spatial characteristics and travel characteristics.

7. The method for predicting carbon emission reduction benefits of drone multi-order delivery according to claim 6 is characterized in that: The characteristic variables of the drone group delivery travel emission reduction benefit prediction model include: emission reduction per unit distance of group delivery trip, peak hours, traffic operation conditions, land use mix, distance to CBD, POI density, bus and subway station density, road network density, number of group deliveries, trip distance, average trip speed, shared distance ratio, detour rate and maximum and minimum order distance ratio.

8. The method for predicting carbon emission reduction benefits of drone multi-order delivery according to claim 7 is characterized in that: The process of constructing variable features based on group-order delivery trip data includes: The constructed features are screened using a multi-criteria integrated feature selection method to determine the characteristic variables that affect the prediction of carbon emission reduction benefits of drone group delivery travel; among them, the method selects the recursive feature elimination method in the parcel feature selection method and the elastic network method and random forest method in the embedded feature selection method; Feature selection is performed on the characteristic variables at the three levels of time, space, and travel, and the features with average standardized scores less than the preset threshold, including peak hours, distance from the city center, density of bus and subway stations, and maximum and minimum order distance ratio, are eliminated. The remaining features with average standardized scores greater than the preset threshold are retained, and finally the characteristic variables of the drone group delivery travel emission reduction benefit prediction model are obtained.

9. The method for predicting carbon emission reduction benefits of drone multi-order delivery according to claim 1 is characterized in that: The carbon emission reduction benefit prediction results are explained and analyzed using the SHAP model interpretation method, and a drone group delivery travel emission reduction strategy is proposed based on the analysis results; The process includes: global model interpretation analysis, single-factor model interpretation analysis, two-factor interaction model interpretation analysis, individual model interpretation analysis, and based on the model interpretation results, proposing corresponding drone group delivery travel emission reduction strategies.

10. A carbon emission reduction benefit prediction system for drone multi-order delivery, characterized in that: include: The data processing module is used to obtain drone delivery data, process the drone delivery data, separate individual delivery orders and group delivery orders, and use them as group delivery trip data; The feature construction module is used to construct variable features based on the group-order delivery trip data, and generate characteristic variables of the drone group-order delivery travel emission reduction benefit prediction model, where the construction of variable features includes three levels: time, space, and travel; The benefit prediction module is used to input the characteristic variables of the drone group delivery emission reduction benefit prediction model into a pre-established optimization prediction model, and output the carbon emission reduction benefit prediction results, wherein the pre-established optimization prediction model is a gradient boosting model obtained by optimizing key hyperparameters through the Bayesian hyperparameter optimization method.

Images