Bus carbon emission factor prediction method and device for unit passengers, and medium

Abstract

The invention discloses a bus carbon emission factor prediction method and device for unit passengers, and a medium. The method comprises the following steps: constructing a bus station level influence factor, a bus link level influence factor and a bus line level influence factor; and taking the bus station level influence factor, the bus link level influence factor and the bus line level influence factor as input factors, processing the input factors through a routing mechanism, calculating a routing score corresponding to each input factor, and distributing each input factor to the corresponding expert sub-model so as to predict the bus carbon emission of the corresponding unit passenger. The key factors of the bus carbon emission are judged, and / or the coupling effect between every two input factors is judged. And introducing a context sensitive frequency and a calling balance degree to evaluate local and global load conditions of the expert sub-model, and quantifying the influence of expert switching on prediction output through expert disturbance sensitivity.

Description

Technical Field

[0001] The present invention belongs to the technical field of urban transportation carbon emission prediction, and in particular relates to a method, device and medium for predicting a public transportation carbon emission factor per passenger. Background Art

[0002] The global climate change problem is becoming more and more serious, and reducing carbon emissions has become a common goal of society. The transportation sector, especially the public transportation system, is one of the important sources of urban carbon emissions. As the core mode of daily travel for urban residents, the carbon emission level of the public transportation system not only determines the quality of the urban environment, but also affects the realization of the city's sustainable development goals.

[0003] Existing research mainly focuses on the prediction of bus system total carbon emissions (BCE). However, since BCE is affected by multiple external factors such as city size, population density and travel structure, its directness and practicality as an absolute indicator in guiding bus system emission reduction are relatively weak. In contrast, the relative indicator Bus Transit System Carbon Emissions Factor per Passenger (PBF) has higher application value. PBF can not only intuitively reflect the carbon emission efficiency and environmental benefits of the bus system, but also provide an accurate basis for evaluating and optimizing the effectiveness of emission reduction strategies.

[0004] However, traditional bus carbon emission prediction methods usually rely on fixed emission factors and static traffic data, which makes it difficult to describe the complex dynamic changes in bus operations. In addition, most existing studies are limited to line-level analysis, ignoring the comprehensive impact of multi-level factors such as stations, links, and lines on PBF, thereby reducing the accuracy of model prediction. More importantly, the current carbon emission prediction model is insufficient in revealing the main effects and coupling effects of influencing factors, resulting in poor interpretability of the carbon emission prediction model. This limitation not only affects the scientificity and effectiveness of emission reduction strategies, but also has a negative impact on fairness and transparency. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a method, device and medium for predicting the carbon emission factor of public transportation per unit passenger.

[0006] In a first aspect, an embodiment of the present invention provides a method for predicting a public transportation carbon emission factor per passenger, the method comprising:

[0007] Construct bus station-level influencing factors, bus link-level influencing factors, and bus line-level influencing factors; wherein the bus station-level influencing factors include: station density, dynamic population density in the station service area, weighted work POI density, and weighted residential POI density; the bus link-level influencing factors include: bus link speed, link real-time passenger load factor, and link road category ratio; the bus line-level influencing factors include: fleet size, vehicle size, line departure interval, bus-bus line connectivity index, bus-subway line connectivity index, and line turnover;

[0008] The bus stop-level influencing factors, bus link-level influencing factors, and bus line-level influencing factors are taken as input factors, processed by the routing mechanism, and the routing score corresponding to each input factor is calculated, so that each input factor is assigned to the corresponding expert sub-model to predict the corresponding unit passenger bus carbon emissions.

[0009] In a second aspect, an embodiment of the present invention provides an electronic device, including:

[0010] at least one processor; and

[0011] a memory communicatively connected to the at least one processor; wherein,

[0012] The memory stores one or more computer programs that can be executed by the at least one processor, and the one or more computer programs are executed by the at least one processor so that the at least one processor can execute the above-mentioned method for predicting the public transportation carbon emission factor per unit passenger.

[0013] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the above-mentioned method for predicting the carbon emission factor of public transportation per unit passenger when executed by a processor.

[0014] Compared with the prior art, the present invention has the following beneficial effects:

[0015] The present invention provides a method for predicting the carbon emission factor of public transportation per unit passenger. By identifying the walking range of bus stop nodes and determining the service area of ​​the bus stop, etc., a set of influencing factors of the carbon emission factor of public transportation per unit passenger is constructed, which includes influencing factors at the bus stop level, influencing factors at the bus link level, and influencing factors at the bus line level. The present invention constructs a prediction model based on a routing mechanism and a sparse activation strategy to integrate the influencing factors at the bus stop level, influencing factors at the bus link level, and influencing factors at the bus line level. The prediction model introduces an expert equilibrium loss to prevent the expert sub-model from being overloaded or idle, thereby predicting the corresponding carbon emission of public transportation per unit passenger. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0017] Figure 1 A flow chart of a method for predicting a public transportation carbon emission factor per passenger provided in an embodiment of the present invention;

[0018] Figure 2 A structural diagram of a unit passenger public transport carbon emission factor prediction model provided in an embodiment of the present invention;

[0019] Figure 3 A schematic diagram of a factor coupling effect network of a unit passenger public transportation carbon emission factor prediction model provided by an embodiment of the present invention;

[0020] Figure 4 A schematic diagram of the coupling effect and main effect ratio of the unit passenger public transportation carbon emission factor prediction model provided in an embodiment of the present invention;

[0021] Figure 5 A schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] 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.

[0023] It should be noted that, in the absence of conflict, the features in the following embodiments and implementations may be combined with each other.

[0024] like Figure 1 As shown, the present invention provides a method for predicting the carbon emission factor of public transportation per unit passenger, the method comprising the following steps:

[0025] Step S1, obtaining the total area of ​​the bus stop service area.

[0026] Specifically, in step S1, the total service area SA of bus stop s s The calculation process includes:

[0027] Step S101, obtaining a walking network ,in, represents a node in the walking network, Represents an edge in a pedestrian network.

[0028] Step S102: Project the bus stop s onto the pedestrian network to generate a projection point, which is used as the first node. Add to the pedestrian network.

[0029] Step S103: If the link in the walking network If the length is greater than the threshold, a second node is inserted into the link in the pedestrian network. , until the distance between adjacent nodes is less than the length threshold, thus obtaining the optimized pedestrian network In this example, the length threshold is set to 10 meters.

[0030] Step S104: After the pedestrian network is optimized In the first node As the starting point, calculate the first node The walking range corresponding to the walking threshold; generating a convex hull area based on the walking range, and taking the convex hull area as the total service area SA of the bus stop s s In this example, the walking threshold is set to 400 meters.

[0031] It should be noted that the bus stop service area refers to the area that pedestrians can reach within a certain time or walking distance from the bus stop. Traditional service area identification methods usually use a fixed radius buffer method, such as setting a 400-meter buffer radius around the bus stop. However, this method ignores the complexity of the actual road network and the impact of geographical obstacles, resulting in a deviation between the recognition results and the actual situation. Therefore, this example combines the optimized pedestrian network and generates a convex hull area based on the walking range, which can more accurately reflect the actual walking conditions, thereby obtaining a more accurate total area of ​​the bus stop service area.

[0032] Step S2, constructing bus stop-level influencing factors, wherein the bus stop-level influencing factors include: station density, dynamic population density in the station service area, weighted work POI density, and weighted residential POI density.

[0033] Among them, bus routes Site density The expression is as follows:

[0034]

[0035] In the formula, It is a bus route The number of sites on It is a bus route Total operating length.

[0036] It should be noted that station density has a dual effect on the bus carbon emission factor per passenger PBF: high station density may lead to frequent stops and starts of vehicles, thereby increasing the bus carbon emission factor per passenger PBF; but at the same time, it can also improve the accessibility of the bus system and attract more passengers, which may reduce the bus carbon emission factor per passenger PBF.

[0037] Furthermore, the population density in the station service area affects the efficiency and passenger volume of the bus system, thereby indirectly affecting the bus carbon emission factor per passenger PBF. In this example, in order to capture the dynamic impact of population changes on the bus carbon emission factor per passenger PBF, the time variable is introduced .exist During the period, bus stops Dynamic population density in the service area The expression is as follows:

[0038]

[0039] In the formula, express Sites within the time period The total population in the service area, Indicates bus stops The total service area.

[0040] Furthermore, the weighted work POI density and weighted residential POI density are used to evaluate the impact of work-residence separation and commuting activities on PBF. In order to reflect the dynamic nature of POI access, the calculation of POI density is combined with access weighting. The calculation process of weighted work POI density and weighted residential POI density includes:

[0041]

[0042] In the formula, represents the weighted density of POI visits of category c in the service area of ​​bus stop s during period t, Indicates bus routes The number of POIs of category c in the service area of ​​bus stop s, where category c is work or residence. It is a bus route The service area of ​​the bus stop s, is the bus route during time period t The total number of visits to bus stop s, Indicates bus routes On site The number of

[0043] Step S3, constructing bus link-level influencing factors, wherein the bus link-level influencing factors include: bus link speed, link real-time passenger load factor, and link road category ratio.

[0044] Furthermore, the impact of bus link speed on the bus carbon emission factor per passenger PBF is multifaceted. In traffic congestion, frequent stops and starts will increase PBF, while appropriate bus link speed can improve fuel efficiency and reduce the bus carbon emission factor per passenger PBF. In addition, a higher bus link speed can shorten the travel time cost of passengers, thereby attracting more passengers to use public transportation and further reducing the bus carbon emission factor per passenger PBF.

[0045] Among them, During this period, bus routes Middle Link Speed The expression is as follows:

[0046]

[0047] In the formula, Indicates link Length, Indicates user exist Links during the period The time consumed. It is during the period Inside vehicle Through the link Passenger capacity at that time.

[0048] Furthermore, bus load factor has a significant impact on PBF. A higher passenger load factor reduces the total vehicle emissions per passenger, thereby reducing the carbon emission factor per passenger. During this period, bus routes Middle Link Passenger load factor The expression is as follows:

[0049]

[0050] In the formula, It is a public transport vehicle The size of a vehicle is measured by its maximum passenger capacity.

[0051] Furthermore, the road category ratio of a link directly and indirectly affects PBF by affecting bus operating efficiency, traffic flow, and fuel consumption. For example, expressways and trunk roads are usually designed with high capacity, providing higher average speeds and better fuel efficiency, but congestion during peak hours may increase PBF. The road category ratio is calculated as follows:

[0052]

[0053] in, It is a bus route link The road category is The proportion of highways, trunk roads, primary roads, secondary roads and tertiary road types are taken into account. It is a bus route link The road category is The total length of Indicates bus routes The total length of

[0054] Step S4, constructing bus line-level influencing factors, wherein the bus line-level influencing factors include: fleet size, vehicle size, line departure interval, bus-bus line connectivity index, bus-subway line connectivity index and line turnover.

[0055] Furthermore, fleet size and vehicle dimensions indirectly affect PBF by changing the composition of the fleet. For example, increasing the number of articulated buses may lead to increased carbon emissions in the production, maintenance, and scrapping stages. However, when articulated buses carry a large number of passengers, resources are used efficiently, which effectively reduces PBF.

[0056] Line departure intervals have a dual effect on PBF: frequent departures not only reduce passengers’ waiting time and attract more passengers to use public transportation, but may also lead to an increase in direct carbon emissions.

[0057] In addition, the connectivity of bus lines with other bus and subway lines determines the convenience of transfer, which in turn affects PBF; the bus-bus line connectivity index is the bus line The number of bus routes that can be transferred within all bus stops The bus-subway line connectivity index is the bus line The number of subway lines that can be transferred within all bus stops .

[0058] Furthermore, routes with high turnover usually carry a large number of commuting demands and connect important transportation hubs. The expression of the route turnover is as follows:

[0059]

[0060] In the formula, Indicates the bus routes in time period t The line turnover, is the number of vehicles in period t Through the link The passenger capacity at that time, Indicates link Length.

[0061] In summary, the statistical results of bus stop-level influencing factors, bus link-level influencing factors, and bus route-level influencing factors are shown in Table 1.

[0062] Table 1: Summary of statistical results of factors affecting carbon emission factors of multi-level unit passenger buses

[0063]

[0064] In step S5, the bus stop-level influencing factors, bus link-level influencing factors, and bus line-level influencing factors are used as input factors, processed by the routing mechanism, and the routing score corresponding to each input factor is calculated, so that each input factor is assigned to the corresponding expert sub-model, thereby predicting the corresponding unit passenger bus carbon emissions.

[0065] Specifically, in this example, Figure 2 As shown in the figure, a unit passenger bus carbon emissions prediction model (MST-Boost) is constructed. The unit passenger bus carbon emissions prediction model (MST-Boost) includes dynamically calculating the routing score of each sample on different expert sub-models through a routing mechanism to determine which expert sub-model will process the sample. At the same time, in order to improve the collaboration efficiency between expert sub-models and reduce the waste of computing resources, the unit passenger bus carbon emissions prediction model (MST-Boost) also introduces a sparse activation mechanism to effectively integrate multi-level factors (including bus station-level influencing factors, bus link-level influencing factors, and bus line-level influencing factors).

[0066] Furthermore, the routing score corresponding to each input factor is calculated, and the expression is as follows:

[0067]

[0068] In the formula, Represents the input sample In the expert submodel The routing score on is the input sample The factor matrix (consisting of site, link and line level factors), is a learnable routing matrix.

[0069] Furthermore, the expression of the sparse activation mechanism is as follows:

[0070]

[0071] In the formula, For input sample The predicted value of carbon emissions per passenger of public transportation, is the number of expert sub-models, e represents the expert sub-model, Indicates the highest K routing scores in the expert sub-model. In this example, the unit passenger bus carbon emissions prediction model (MST-Boost) takes K=1, that is, only one expert sub-model is activated for each input sample.

[0072] It should be noted that, unlike the traditional mixture of experts (MoE) model in which multiple experts are activated simultaneously, the sparse activation mechanism provided by the unit passenger bus carbon emissions prediction model (MST-Boost) in this example has the following advantages: Improve the degree of differentiation: Each expert sub-model focuses on processing a certain type of specific level factors to avoid interference between tasks; Improve model scalability: The number of expert sub-models can be increased without significantly increasing the computational complexity; Avoid training instability: By reducing parameter competition and gradient conflicts, the stability of model training is significantly improved.

[0073] Furthermore, the training process of the expert sub-model includes:

[0074] Set the loss function, the expression is as follows:

[0075]

[0076]

[0077] In the formula, is the total loss of the expert sub-model, is the prediction loss, is the expert equilibrium loss, is the weight of the expert equilibrium loss term, Expert sub-model The probability of being selected (this probability is calculated through the statistical distribution of routing scores. By encouraging experts to choose a more even distribution, we can avoid some experts from participating in the calculation too frequently, while ensuring that all experts can be effectively utilized, thereby improving the robustness and generalization ability of the model).

[0078] Further, in this example, the expert sub-model can be selected from a neural network (NN) and an extreme gradient boosting model (XGBoost).

[0079] In addition, this example considers other comparison models, including support vector machine (SVM), random forest (RF), lightweight gradient boosting machine (LightGBM), neural network (NN), XGBoost, ensemble learning (EL), deep gradient boosting machine (DeepGBM), deep extreme gradient boosting (DeepBoost) and hybrid boosting (MoE-Boost). As shown in Table 2, the model performance comparison results show that MST-Boost performs well in various indicators, with a root mean square error (RMSE) of 0.261, a mean absolute error (MAE) of 0.208, and a determination coefficient (R²) of 0.932. This outstanding performance is mainly due to the efficient routing mechanism and sparse activation strategy adopted by MST-Boost, which significantly improve the efficiency of collaborative integration among multiple expert systems.

[0080] Table 2: Model performance comparison results

[0081]

[0082] Furthermore, the method further comprises:

[0083] Step S6, by removing each input factor in sequence, calculating the SHAP value corresponding to each input factor, and sorting them, and taking the first K input factors as the key factors of public transportation carbon emissions;

[0084] and / or,

[0085] By removing the input factors pairwise in sequence and calculating the coupling effects between the input factors pairwise, the coupling effects between the input factors pairwise can be determined.

[0086] Among them, the SHAP value corresponding to each input factor is calculated, that is, the first input factor is calculated Main Effect , the expression is as follows:

[0087]

[0088] In the formula, F represents the set of all input factors, Indicates that the first input factor is excluded from the set F A subset of Consider a subset , the first input factor The predicted value of carbon emissions per passenger per bus output by the model is: Consider a subset The predicted value of carbon emissions per passenger per bus output by the model.

[0089] Among them, the coupling effect between the two input factors is calculated, that is, the first input factor is calculated With the second input factor The coupling effect , the expression is as follows:

[0090]

[0091] In the formula, Represents the first input factor With the second input factor The coupling effect between them, F represents the set of all input factors, Indicates that the first input factor is excluded from the set F With the second input factor A subset of Consider a subset , the first input factor With the second input factor The predicted value of carbon emissions per passenger per bus output by the model is: Consider a subset , the first input factor The predicted value of carbon emissions per passenger per bus output by the model is: Consider a subset , the second input factor The predicted value of carbon emissions per passenger per bus output by the model is: Consider a subset The predicted value of carbon emissions per passenger per bus output by the model.

[0092] Among them, the factor coupling effect network of the unit passenger bus carbon emission prediction model (MST-Boost) is as follows: Figure 3 As shown, the significant coupling effect between line turnover and other factors is revealed. Figure 4 The ratio of coupling effect to main effect is shown, and the results show that the relative main effect of line turnover is 97.4%, the highest among all factors; while the relative coupling effect of the proportion of primary roads is the most prominent, reaching 17.0%.

[0093] It should be noted that this example improves the interpretability of the expert sub-model from the following three dimensions:

[0094] Model load balancing analysis: In order to comprehensively evaluate the usage of expert sub-models in different contexts and detect whether there is overload (a certain expert is over-called) or idle (a certain expert is hardly called), this example proposes two load evaluation indicators: context-sensitive frequency and call balance , which are used to measure the load distribution of the expert model in the local context and the global scope. In the local scope, the context label set is introduced , such as "peak hour", "off-peak hour", "main road section", etc., which are used to divide the sample set to analyze the expert's call in a specific context. Expert models in context The selection frequency in is:

[0095]

[0096] in, It is context The total number of samples in ; is the indicator function, if the sample Assigned to an expert ,but , otherwise it is 0. This indicator can be used to analyze the actual participation of each expert model in a specific context, thereby revealing whether there is a "biased load" problem in the expert model under specific conditions. In the global scope, in order to measure the overall call balance of all expert models, a call balance based on information entropy is proposed , defined as follows:

[0097]

[0098] In the formula, the call balance H ranges from When the call balance H is close to 100%, it means that the call distribution of the expert model is more uniform; when the call balance H is close to 0%, it means that there is a serious call imbalance in the model, which may cause some experts to be overloaded or idle, thus affecting the robustness and generalization ability of the model.

[0099] Routing decision graph analysis: To analyze the correlation between expert choice and sample distribution, t-SNE is used to analyze the input features. Dimensionality reduction:

[0100]

[0101] In the formula, Represents the features after dimensionality reduction.

[0102] Each sample Represented in two-dimensional space, label the expert (or expert set) to which it belongs:

[0103]

[0104] By generating a two-dimensional routing map , to analyze the coupling relationship between expert selection and input feature structure.

[0105] Expert redundancy analysis: given a sample feature set , which is distributed by the router to A subset of activated expert models This example introduces an adversarial perturbation function , whose goal is to maximize the switching of the expert router output without significantly changing the input characteristics, that is:

[0106]

[0107] in, Expert selection probability distribution for router output; represents the KL divergence, which is used to measure the difference between the expert distribution before and after the perturbation; Control the disturbance amplitude. Then define the expert disturbance sensitivity index , which is used to measure the impact of expert switching on the prediction output:

[0108]

[0109] in, is the expert selection set after disturbance. , indicating that expert switching has a weak effect on prediction, there is redundancy or the router is insensitive to the feature; otherwise, it means that expert selection has a significant effect on prediction.

[0110] In this example, in order to quantify the degree of global redundancy, a redundancy index is proposed :

[0111]

[0112] in, is the tolerance threshold for redundant judgment (set to 5% relative error); is the indicator function, when It is recorded as 1 when it is, otherwise it is 0; is the total number of samples. The higher the value, the more redundant experts there are in the system.

[0113] In summary, the present invention provides a method for predicting the bus carbon emission factor per unit passenger, by identifying the walking range of bus stop nodes, determining the service area of ​​the bus stop, etc., so as to construct a set of influencing factors of bus carbon emission factors per unit passenger covering bus stop-level influencing factors, bus link-level influencing factors, and bus line-level influencing factors; the present invention is based on a routing mechanism and a sparse activation strategy to construct a prediction model for integrating bus stop-level influencing factors, bus link-level influencing factors, and bus line-level influencing factors. The prediction model introduces an expert equilibrium loss to prevent the expert sub-model from being overloaded or idle, thereby predicting the corresponding bus carbon emissions per unit passenger.

[0114] Accordingly, the present application also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method for predicting the carbon emission factor of public transportation per unit passenger. Figure 5 As shown, a hardware structure diagram of any device with data processing capability for the method for predicting the carbon emission factor of public transportation per passenger provided by an embodiment of the present invention is shown, except Figure 5 In addition to the processor, memory and network interface shown, any device with data processing capability in which the apparatus in the embodiment is located may also include other hardware, generally based on the actual functions of the device with data processing capability, which will not be described in detail.

[0115] Accordingly, the present application also provides a computer-readable storage medium on which computer instructions are stored, and when the instructions are executed by a processor, the method for predicting the carbon emission factor of public transportation per unit passenger as described above is implemented. The computer-readable storage medium may be an internal storage unit of any device with data processing capabilities described in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be an external storage device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), an SD card, a flash card (Flash Card), etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capabilities. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and may also be used to temporarily store data that has been output or is to be output.

[0116] The above embodiments are only used to illustrate the design ideas and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, any equivalent changes or modifications made based on the principles and design ideas disclosed by the present invention are within the protection scope of the present invention.

Claims

1. A method for predicting the carbon emission factor of public transportation per passenger, characterized in that: The method comprises: Construct bus station-level influencing factors, bus link-level influencing factors, and bus line-level influencing factors; wherein the bus station-level influencing factors include: station density, dynamic population density in the station service area, weighted work POI density, and weighted residential POI density; the bus link-level influencing factors include: bus link speed, link real-time passenger load factor, and link road category ratio; the bus line-level influencing factors include: fleet size, vehicle size, line departure interval, bus-bus line connectivity index, bus-subway line connectivity index, and line turnover; The bus stop-level influencing factors, bus link-level influencing factors, and bus line-level influencing factors are taken as input factors, processed by the routing mechanism, and the routing score corresponding to each input factor is calculated, so that each input factor is assigned to the corresponding expert sub-model to predict the corresponding unit passenger bus carbon emissions.

2. The method for predicting the carbon emission factor of public transportation per passenger according to claim 1, characterized in that: The calculation process of the dynamic population density within the station service area includes: During time period t, the population density PD in the service area of ​​bus stop s is t,s is the total population in the service area of ​​bus stop s during period t and the total area of ​​the service area of ​​bus stop s SA s ratio; Among them, the total service area of ​​bus stop s is SA s The calculation process includes: Get walking network; Project the bus stop s onto the walking network, thereby adding the first node to the walking network; If the link in the walking network is greater than the length threshold, a second node is inserted into the link in the walking network until the distance between adjacent nodes is less than the length threshold, thereby obtaining an optimized walking network; In the optimized pedestrian network, the first node is taken as the starting point, and the walking range corresponding to the first node is calculated; a convex hull area is generated based on the walking range, and the convex hull area is used as the total service area SA of the bus stop s s .

3. The method for predicting the carbon emission factor of public transportation per passenger according to claim 1, characterized in that: The calculation process of weighted work POI density and weighted residential POI density includes: Calculate the weighted POI density of category c in the service area of ​​bus stop s during time period t, where the weighted POI density is equal to the number of POIs of category c in the service area of ​​bus stop s multiplied by the total number of visits to bus stop s during time period t, divided by the product of the service area area of ​​bus stop s and the total number of visits to all bus stops on the bus line during time period t; Among them, category c is work or residence.

4. The method for predicting the carbon emission factor of public transportation per passenger according to claim 1, characterized in that: The bus-bus line connectivity index is the number of bus lines that can be transferred within all bus stops on the bus line; The bus-subway line connectivity index is the number of subway lines that can be transferred to at all bus stops on the bus line; The calculation process of the line turnover includes: a bus line includes several links, and the line turnover of the bus line in a period of time t is equal to the passenger volume when the vehicle passes through the link in the period of time t multiplied by the link length, and then the sum of all links is calculated.

5. The method for predicting the carbon emission factor of public transportation per passenger according to claim 1, characterized in that: The training process of the expert sub-model includes: Setting a loss function, wherein the loss function is a weighted sum of prediction loss and expert equilibrium loss; The expert balance loss is used to make the use probability distribution of the expert sub-model more uniform, so as to avoid overloading or idling of some expert sub-models.

6. The method for predicting the carbon emission factor of public transportation per passenger according to claim 1, characterized in that: The method further comprises: By removing each input factor in order, calculating the SHAP value corresponding to each input factor and sorting them, the first K input factors are taken as the key factors of public transportation carbon emissions; and / or, By removing the input factors pairwise in sequence and calculating the coupling effects between the input factors pairwise, the coupling effects between the input factors pairwise can be determined.

7. The method for predicting the carbon emission factor of public transportation per passenger according to claim 1, characterized in that: The process of calculating the coupling effect between two input factors includes: Based on the set F of all input factors, the subset S' excluding the first input factor and the second input factor in the set F, the predicted value of carbon emissions per passenger bus output by the model considering the subset S', the first input factor and the second input factor, the predicted value of carbon emissions per passenger bus output by the model considering the subset S' and the first input factor, the predicted value of carbon emissions per passenger bus output by the model considering the subset S' and the second input factor, and the predicted value of carbon emissions per passenger bus output by the model considering the subset S', calculate the coupling effect between any two input factors.

8. The method for predicting the carbon emission factor of public transportation per passenger according to claim 1, characterized in that: The method further comprises: Analyze the routing mechanism and the interpretability of the expert sub-model; including: Constructing context-sensitive frequency and call balance, wherein the context-sensitive frequency is used to measure the call frequency of the expert sub-model in the context, and the call balance is used to measure the load balancing state in the global scope to identify overload or idle problems of the expert sub-model; generating a two-dimensional expert routing distribution map, wherein the two-dimensional expert routing distribution map is used to analyze the coupling correlation between the selection of the expert sub-model and the input features; Evaluate the expert disturbance sensitivity index, which is used to evaluate the impact of expert routing changes on prediction results under small input disturbances, thereby reflecting the switching stability and prediction robustness between expert sub-models; A redundancy index is calculated, where the redundancy index is used to quantify the redundancy degree of the expert sub-model to assist in optimizing the configuration of the expert sub-model and the design of the routing mechanism.

9. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores one or more computer programs that can be executed by the at least one processor, and the one or more computer programs are executed by the at least one processor to enable the at least one processor to execute the method for predicting the public transportation carbon emission factor per unit passenger as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the computer program implements the method for predicting the public transportation carbon emission factor per unit passenger according to any one of claims 1 to 7.

Images