Method for determining a quantity of at least one pollutant emitted by a set of vehicles within a road network

Abstract

The present invention relates to a method for determining the quantity of pollutants (Q) emitted by at least one vehicle on a section of a road network, in which an average speed (VIT) of the vehicles on the sections of the road network is acquired, and a flow rate model (MDV) is used to determine the flow rate of vehicles on the sections of the road network, and a unitary pollutant emission model (MUE) is used to determine a quantity of pollutant emissions from a single vehicle. Finally, for all vehicles in the road network, an aggregated quantity (Q) of the pollutants emitted is determined using the determined flow rate (Db), the determined unitary quantity (Qu), as well as the distribution of driving styles (DSC) and the thermal state of the vehicles (DTH). Figure 1 to be published

Description

Title of the invention: Method for determining a quantity of at least one pollutant emitted by a set of vehicles within a road network Technical field

[0001] The present invention relates to the field of determining pollutant emissions emitted by a set of vehicles on a road network.

[0002] Today, the World Health Organization (WHO) describes air pollution as one of the greatest environmental risks to health, having a direct impact on strokes, heart disease, lung cancer, and respiratory conditions, both chronic and acute, including asthma. In 2019, 99% of the world's population lived in places where pollutant concentrations did not meet the thresholds recommended by the WHO. The responsibility of road traffic is evident in terms of air pollution. In France, the Ministry of Ecological Transition states that road transport is responsible for 65% to 100% of all transport emissions depending on the pollutants. In Paris, 65% of nitrogen oxide emissions and 50% of total PM 10 particle emissions are due to road traffic.

[0003] Transport, all modes combined, is responsible for approximately 50% of global nitrogen oxide (NOX) emissions and approximately 10% of PM2.5 particulate emissions (i.e. particles with a diameter of less than 2.5 microns). Road transport alone represents a considerable part of this transport contribution, with 58% of NOX emissions and 73% of PM2.5 particulate emissions. These emissions are mainly due to three factors: exhaust emissions, abrasion emissions, and evaporative emissions. While heavy goods vehicles are the main emitters of pollutants, it is private vehicles, more represented in densely populated urban areas, which have the highest impact on citizens' exposure to poor air quality.

[0004] Measures implemented at local level to manage transport use (such as better transport planning and measures to encourage modal shift, i.e. changing transport modes), as well as the gradual renewal of the vehicle fleet, have helped to limit exhaust gas emissions from road transport in cities and urban areas. Indeed, worldwide, road transport activity has increased by a quarter over the last decade, while NOX emissions have increased by 5% and particulate emissions have decreased by 6%. Despite these improvements, levels pollution levels still exceed the thresholds set by the WHO in many cities.

[0005] To significantly improve air quality in their territory, local authorities in urban areas must take action to reduce emissions linked to road transport. For example, at present, this sector represents two-thirds of total nitrogen oxide (NOX) emissions and one-third of total PM10 particle emissions in the Lyon Metropolitan Area. However, to date, the teams of the Urban Mobility Roads department of this authority do not have tools enabling them to determine the impact of road developments on air quality. Decision-making relating to these subjects therefore does not take into account the impact on emissions, due to a lack of tools.

[0006] More specifically, today, the operational tools that allow cities and road managers to estimate pollutant and greenhouse gas emissions on the different sections of a road network are macroscopic models based on emission factors (a multiplicative coefficient depending on the average speed on the section). In general, it is known that at parity of average speed, two different road sections (for example an urban section with free-flowing traffic and a congested motorway section) can present very contrasting emission levels, which would be very poorly represented by a single emission factor.

[0007] Consequently, it is difficult for cities to make the right decisions regarding road infrastructure planning and legislation without having precise tools for assessing and projecting the impact of the measures envisaged on polluting emissions from road transport and air quality. These new tools should ideally make it possible to assess the impact of measures on very fine temporal and spatial scales (of the order of one minute, and of the order of ten meters). Prior art

[0008] The first approach to estimating vehicle emissions from its dynamics is simply to perform emissions tests on a dynamometric bench. Measuring pollutants makes it possible, for example, to establish matrices that give an average pollutant emission rate associated with speed and acceleration values ​​of the vehicle in question. However, the data from such tests are not representative of real driving conditions since they essentially characterize stationary operating speeds of the vehicle's engine.

[0009] To address this problem, portable emissions measurement systems have been developed since the 1990s to link emissions to vehicle dynamics in real-life conditions. From this data, more accurate models, known as microscopic, can be established to give the instantaneous emission rate of a vehicle as a function of its dynamics. These can, for example, be neural networks, or linear regression models.

[0010] To go further, some of these models previously calibrated from experimental data were also validated in real driving.

[0011] Portable emission measurement systems are suitable for the characterization of a few specific vehicles, but not for very large-scale measurements, in particular due to the cost associated with their use. An alternative solution could be to indirectly measure traffic emissions using air quality sensors, but it is difficult to associate pollution with its source. Thus, when it comes to estimating pollutant emissions from a vehicle for which this type of data is not available, an alternative is to calculate them using a microscopic physical model, which additionally takes as input the characteristics of the vehicle in question (weight, engine, etc.), as described in Gartner, U., Hohenberg, G., Daudel, H., & Oelschlegel, H. (2004). Development and application of a semi-empirical NOX model to varions HD diesel engines.In Thermo- and Fluid Dynamic Processes in Diesel Engines 2 (pp. 285-312). Springer, Berlin, Heidelberg, or in the widely used Comprehensive Modal Emission Model (CMEM) from the University of California, which is a combination of parameterizable physical models (as described in Barth, M., An, F., Younglove, T., Scora, G., Levine, C., Ross, M., & Wenzel, T. (2000). Comprehensive Modal Emission Model (CMEM), version 2.0 user's guide. University of California, Riverside, 4.).

[0012] These microscopic models, whether data-based or physically based, are however expensive in terms of computational time and computing resources (memory and processor) for calculations and are therefore generally mainly designed for offline studies. In addition, they require precise dynamic speed profiles which are in practice only rarely measured. Thus, another category of models, called macroscopic, has been developed to estimate emissions on a large scale when the numerical complexity becomes too great, or when individual vehicle trajectories are not available. Among macroscopic emission models, the most widespread approach is to introduce emission factors (EFs) (as described in Ntziachristos, L., Gkatzoflias, D., Kouridis, C., & Samaras, Z. (2009). COPERT: a European road transport emission inventory model.In Information technologies in environmental engineering (pp. 491-504) et Yu, L., Xu, Y., Song, G., Hao, Y., Guo, S., & Shi, Q. (2009). Development and application of macroscopie émission model for China. Transportation research record, 2123(1), 66-75. Springer, Berlin, Heidelberg.). Les . EF emission factors correspond to average values ​​of pollutant emissions per vehicle and per distance traveled. In most cases, this approach is based on "average" vehicles with an "average" driving style in order to be representative of the level of emissions when scaling up. The so-called "COPERT" (from the English "COmputer Program to calculate Emissions from Road Transports", which can be translated as a computer program for calculating emissions from road transport and described in the document Ntziachristos, L., Gkatzoflias, D., Kouridis, C., & Samaras, Z. (2009). COPERT: a European road transport emission inventory model. In Information technologies in environmental engineering (pp. 491-504). Springer, Berlin, Heidelberg.) and "HBEFA" (Hausberger, S. (2009). Emission Factors from the Model PHEM for the HBEFA Version 3.) are widely used macroscopic models based on EF emission factors. This type of model is very suitable for carrying out vehicular emission balances at regional or national scales. It can also be used to estimate emissions associated with long journeys quite accurately. However, the EF emission factor approach is by definition not sufficiently accurate when estimating real emissions on a small scale, because it does not take into account the impact of local infrastructure and driving style; thus, their main source of error comes from the impact of driving style and gradient.

[0013] It is also possible to calculate unit emissions more precisely over territories with mesoscopic models which take as input, for example, the number of stops per unit of distance and the duration of these stops, in addition to the average speed. It is also possible to predict the consumption of a vehicle mesoscopically, from which the emissions of certain pollutants can be derived. For example, patent application FR3122011Al (US 2022 / 0335822) describes a method for determining a quantity of pollutants emitted at the mesoscopic scale. However, this model can sometimes be imprecise, because it does not take into account variations linked to the movements of vehicles. Summary of the invention

[0014] The method according to the invention aims to precisely determine a quantity of polluting emissions emitted by a set of vehicles on a road network by taking into account the impact of congestion, signage and the topography of the road network (slope, road layout, etc.), as well as the impact of variations linked to the movements of vehicles within the road network, all with limited computing requirements (processors and memories). In addition, the method according to the invention aims to be adapted to any size of the road network considered. For these purposes, the invention relates to a method for determining the quantity of pollutants emitted by at least one vehicle on a section of a road network, in which an average speed of the vehicles on the sections of the road network is acquired, and a flow model is used to determine the flow of vehicles on the sections of the road network, and a unit model of pollutant emissions (microscopic or mesoscopic) is used to determine a quantity of pollutant emissions from a single vehicle. Finally, for all vehicles on the road network, an aggregate quantity of pollutants emitted is determined using the determined flow rate, the determined unit quantity, as well as the distribution of driving styles and the thermal state of the vehicles (i.e. the thermal state of an after-treatment system fitted to the vehicle, in other words an indication of whether the after-treatment system is hot or cold, which reflects its effectiveness, particularly for Nox).A feature of the invention is to provide a unitary model of pollutants for each driving style and for each thermal state, thus, the determination of the aggregated quantity of pollutants is more precise than a macroscopic model, while using more limited computing resources than a microscopic model.

[0015] The invention relates to a method for determining a quantity of at least one pollutant emitted by a set of vehicles within a road network, said road network comprising a set of strands characterized by macroscopic data including the topography of said strands, the movement of said vehicles within said road network being characterized by at least two driving styles and by at least two thermal states of the vehicles. For this method, the following steps are implemented: a. An average vehicle speed is acquired for each strand of said road network; b. A vehicle flow rate is determined for each strand of said road network from said average speed acquired by means of a vehicle flow model which links the vehicle flow rate, the average vehicle speed and said macroscopic data of said strand; c. For each strand, a unit quantity of said at least one pollutant emitted by a single vehicle is determined for each driving style and for each thermal state of the vehicles, based on said average speed acquired and a unit model of pollutant emissions; and d. An aggregate quantity of said at least one pollutant for all vehicles is determined for each strand of said road network, based on said determined vehicle flow rate, said unit quantities of said at least one pollutant emitted by a single vehicle, and based on a distribution of the driving styles of the vehicles within the road network, and a distribution of said thermal states of the vehicles within the road network.

[0016] According to one embodiment, said aggregated quantity of at least one pollutant emitted is determined for all vehicles by further considering data from a vehicle fleet associated with said road network in question.

[0017] According to one implementation, said unitary pollutant emissions model is a unitary mesoscopic pollutant emissions model, said unitary mesoscopic pollutant emissions model being a multivariate regression of the quantity of pollutant emissions by driving style and by thermal state of the vehicle.

[0018] Advantageously, said unitary mesoscopic model of pollutant emissions is constructed by a machine learning or deep learning method trained on a learning base formed on a learning road network comprising the average vehicle speed on the strands of said learning road network, macroscopic data of said strands of said learning road network, and data acquired from a unit quantity of emissions of said at least one pollutant on said learning road network.

[0019] According to one embodiment option, a unit quantity of said at least one pollutant emitted by a single vehicle is determined by means of the following steps: i. For each strand, for each driving style, a speed profile of said single vehicle on said strand is determined from said average speed and said macroscopic data of said strand; ii. For each strand, for each driving style and for each thermal state of said vehicle, said unit quantity of said at least one pollutant is determined by means of a unitary microscopic model of pollutant emissions applied to said speed profile of said single vehicle.

[0020] Advantageously, said speed profile is constructed by means of a speed profile model constructed by a machine learning or deep learning method trained on a learning base formed on a learning road network, said learning base comprising the average vehicle speed on the strands of said learning road network, macroscopic data of said strands of said learning road network, and acquired instantaneous speed data of at least one vehicle on said strands of said learning road network.

[0021] Advantageously, said unitary microscopic model of pollutant emissions is constructed by a machine learning or deep learning method trained on a learning base formed on a learning road network comprising the speed profile of vehicles within said learning road network, macroscopic data of said strands of said learning road network, and acquired data of unit quantity of pollutant emissions. on the said strands of the said learning road network.

[0022] According to one configuration, said macroscopic data of a road network are chosen from the topology of the strand of the road network, the number of lanes of the strand of the road network, the maximum authorized speed of the strand of the road network, the slope of the strand of the road network, the signaling of the strand of the road network, the length of the strand of the road network, preferably said macroscopic data of said road network being provided by a geographic information system.

[0023] According to one embodiment, said average speed of the vehicles is acquired by measurement and / or by means of a geographic information system.

[0024] According to one implementation, said flow model is constructed by a machine learning or deep learning method trained on a learning base trained on a learning road network comprising the flow of vehicles within said learning road network, macroscopic data of said strands of said learning road network, and the average speed of vehicles on said strands of said learning road network.

[0025] According to one aspect, said quantity of at least one pollutant emitted by the at least one vehicle on said at least one strand of said road network is displayed on a road map, preferably by means of a smartphone or a computer system.

[0026] Other characteristics and advantages of the method according to the invention will appear on reading the following description of non-limiting examples of embodiments, with reference to the appended figures described below. List of figures

[0027] [Fig.l]

[0028] [Fig.l] illustrates the steps of the method according to a first embodiment of the invention.

[0029] [Fig.2]

[0030] [Fig.2] illustrates the steps of the method according to a second embodiment of the invention.

[0031] [Fig.3]

[0032] [Fig. 3] illustrates the steps of the method according to a third embodiment of the invention.

[0033] [Fig.4]

[0034] [Fig.4] illustrates part of the steps of the method according to an alternative embodiment of the invention.

[0035] [Fig.5]

[0036] [Fig.5] illustrates, for an example, curves of the quantity of CO2 emitted by vehicles on a road section as a function of the average speed of the vehicles, a curve being determined by a method according to a prior art, and the two other curves being determined by means of an embodiment of the method according to the invention as a function of two limit speeds.

[0037] [Fig.6]

[0038] Figures 6A to 6D illustrate, for one example, a representation of the quantities of pollutants emitted on a road network, Figures 6A, 6B and 6C corresponding to different driving styles, and Figure 6D corresponding to a determination according to a prior art. Description of the embodiments

[0039] The present invention relates to a method for determining the quantity of at least one pollutant of at least one vehicle on at least one strand of a road network. Thus, in general, the method makes it possible to determine the quantity of pollutants of a vehicle or of a plurality of vehicles which circulate on a portion of a road network. The quantity of pollutants (also called quantity of pollutant emissions) can, for example, be expressed in mass per distance, or in concentration, or in volume per distance. The road network consists of all the roads for a given territory, for example for a country or for a region, or for a city, a district of a city, etc. A strand of the road network is an elementary subdivision of the road network between two consecutive nodes of the road network.For example, a strand of the road network can be a road between two consecutive intersections, between two consecutive signs, between an intersection and a sign, or a section of motorway between two consecutive exits, etc. Thus, we have a fine division of the road network, and a model that is adapted to the road network without microscopic data. Thus, this division makes it possible to obtain the most representative possible determination of pollutant emissions at a fine spatial scale.

[0040] Each strand of the road network is characterized by at least one macroscopic data, in particular the topology (i.e. the slope, the turns, the intersections, the signage, etc.). According to one aspect of the invention, the macroscopic data of the road network may be the topology (i.e. the slope, the turns, the intersections, the signage, etc.), the number of lanes of the strand of the road network, the maximum authorized speed of the strand of the road network, the slope of the strand of the road network, the signage of the strand of the road network, the length of the strand of the road network, preferably said macroscopic data of said road network being provided by a geographic information system. Preferably, the macroscopic data of the road network may be provided by a geographic information system (GIS).Here Maps™, Google Maps™, OpenStreetMap™ are examples of geographic information systems. The . macroscopic data is real data that is always available and from any location.

[0041] At least two driving styles are associated with the road network to characterize the behavior of vehicles within the road network. Since pollutant emissions are directly linked to the driving style, taking them into account allows for better accuracy of the quantity of pollutants emitted. Indeed, taking a single driving style into account underestimates the impact of the most polluting driving styles. For example, the driving style can be chosen from smooth driving (with low accelerations and decelerations), average driving (with medium accelerations and decelerations), aggressive driving (with significant accelerations and decelerations). Thresholds can be defined for accelerations and decelerations to characterize each driving style. In addition, a distribution of driving styles within the road network is associated with the road network.For example: for a road network, we can associate 25% of smooth driving, 50% of average driving and 25% of aggressive driving. According to another example, we could have a variable distribution per road section, or per portion of the road network. Advantageously, these driving styles and this distribution can be determined from measurements of instantaneous speeds of vehicles within the road network. Alternatively, these driving styles and this distribution can be predetermined.

[0042] At least two thermal states of the vehicles are associated with the road network to characterize the polluting behavior of the vehicles within the road network. The thermal state of a vehicle is called the thermal state of a post-treatment system equipping the vehicle, when it is a vehicle with a thermal engine. In other words, the thermal state of the vehicle is representative of the temperature of the post-treatment system equipping the vehicle. Indeed, when a vehicle starts, the post-treatment system is generally at a temperature lower than its optimal operating temperature, consequently, the vehicle will emit more pollutants. On the contrary, when a vehicle has been in operation for a certain time, the post-treatment system is generally at a temperature greater than or equal to its optimal operating temperature, consequently, the pollutant emissions of the vehicle are limited.Taking these thermal states into account therefore allows for an accurate estimation of the quantity of pollutants emitted. For example, the thermal state of the vehicle can be chosen from a cold vehicle (with an aftertreatment system temperature below a predetermined threshold) and a hot vehicle (with an aftertreatment system temperature above a predetermined threshold). In addition, a distribution of thermal states within the road network is associated with the road network. For example: for a road network, 15% of can be associated. cold vehicles and 85% hot vehicles. According to another example, we could have a variable distribution by road section, or by portion of the road network. Advantageously, these thermal states and this distribution can be determined from measuring the travel times of vehicles within the road network. Alternatively, these thermal states and this distribution can be predetermined, in particular depending on the type of road network considered: on motorways, the thermal state of vehicles is generally similar, with post-treatment systems at a high temperature, whereas in built-up areas, several thermal states are present.

[0043] The method according to the invention can make it possible to determine the polluting emissions of at least one of the pollutants (including gaseous compounds and / or particles) chosen from: nitrogen oxides NOX, carbon dioxide CO2, carbon monoxide CO, fine particles PM, unburned hydrocarbons, etc.

[0044] Preferably, the vehicle is a motorized vehicle traveling within the road network, such as a motor vehicle, a two-wheeler, a heavy goods vehicle, a coach, a bus.

[0045] Advantageously, the method according to one embodiment of the invention can determine the quantity of pollutants for a plurality of vehicles moving on the road network. Thus, the invention makes it possible to measure the quantity of pollutants for a set of vehicles, thus ensuring a determination of the quantity of pollutants for at least one strand of the road network, preferably on a plurality of strands of the road network.

[0046] In the present application, the expression “unit quantity of pollutants” designates a quantity of pollutants emitted by a single vehicle on a section of the road network per unit of distance (it can be expressed in mg / km), the expression “aggregate quantity of pollutants” designates a quantity of pollutants emitted by all the vehicles on a section of the road network, per unit of time (it can be expressed in mg / s).

[0047] According to the invention, the following steps are implemented:

[0048] A- Acquisition of average speeds

[0049] B- Determination of a vehicle flow rate

[0050] C- Determination of a unit quantity of pollutants

[0051] D- Determination of an aggregate quantity of pollutants

[0052] These steps can be implemented by computer means (server, computer, etc.). Steps B and C can be carried out in this order, simultaneously or in reverse order. These steps will be detailed in the remainder of the description.

[0053] [Fig.l] illustrates schematically and in a non-limiting manner, the steps of the method for determining a quantity of pollutant emissions according to a first embodiment of the invention. In a first step, an average speed VIT of the vehicles within the road network is acquired. From this speed average VIT, on the one hand, a vehicle flow rate Db is determined for each strand of the road network using a flow rate model MDV, and on the other hand, a unit quantity of pollutants emitted Qu is determined using a unit model of the quantity of pollutants emitted. An aggregate quantity of pollutants emitted Q is then determined, using the flow rate Db, the unit quantity Qu, as well as a distribution of the styles of conduits DSC and a distribution of the thermal states DTH. A- Acquisition of average speeds

[0054] During this step, an average speed of the vehicles is acquired for each strand of the road network. Using this average speed, the quantity of pollutants emitted for each strand of a road network can be estimated.

[0055] According to an alternative embodiment, this average speed can be acquired by measuring the instantaneous speeds of the vehicles within the road network. These instantaneous speeds can be measured in particular by means of a smartphone, or by means of a geolocation system (such as a satellite positioning sensor, such as the GPS system - from the English Global Positioning System -, the Galileo system, etc.), or by means of instrumentation of at least one vehicle, or any similar means. Measurement by means of a smartphone allows large-scale measurement without the need for instrumentation of a plurality of vehicles.

[0056] According to an implementation of the invention, this step can implement an FCD database (from the English "floating car data" which can be translated as data of journeys made and measured). As a non-limiting example, the FCD database used can contain approximately two million instantaneous speed measurements for approximately two hundred thousand strands of the road network.

[0057] According to another embodiment, this average speed can be acquired from a geographic information system (GIS), where appropriate, one which includes the macroscopic data of the road network. Thus the average speed can be available at any time and for all strands.

[0058] Cumulatively to these achievements, at least one instantaneous speed can be generated by a microscopic traffic simulator or a macroscopic traffic simulator, the simulator is duly calibrated with measurements of journeys taken. The so-called "SUMO" simulator (for "Simulation of Urban Mobility" which can be translated as urban mobility simulation) is an example of an open-source microscopic traffic simulator. This simulated instantaneous speed then makes it possible to deduce an average speed, which makes it possible to implement the method according to the invention for road networks with a limited number of measurements, or makes it possible to predict pollutant emissions for a new road network architecture or for a modification of a road network architecture. B- Determination of vehicle flow

[0059] In this step, a vehicle flow rate is determined for each strand of the road network. This step implements a vehicle flow rate model and the average speed acquired in step A. The vehicle flow rate model is a model that links the vehicle flow rate, the average vehicle speed and the macroscopic data of the strand considered, the vehicle flow rate model also being time-dependent (hour, day, season, etc.). Thus, the flow rate model has as input an average speed of each strand of the road network and macroscopic data, an instant (hour, day, season, etc.) and as output a vehicle flow rate of the strand considered.

[0060] Since the traffic data are variable over time (for example to translate the start and end of day congestion times, or to translate the day of the week, or to translate the month of the year, school holiday periods, public holidays, etc.), the vehicle flow model thus constructed determines a flow rate which depends on the time considered, making the determination of the flow rate precise at the time scale considered.

[0061] According to one embodiment of the invention, the flow model can be constructed by a machine learning or deep learning method trained on a learning base formed on a learning road network comprising the flow of vehicles within said learning road network, macroscopic data of the strands of said learning road network, and the average speed of the vehicles on said strands of the learning road network. The learning road network is a road network for which the flow rate is known for constructing the flow model. The learning road network may be different from or identical to the road network considered for implementing the method.

[0062] For this embodiment, the maximum flow rate model can be constructed by machine learning, preferably by supervised machine learning, preferably by supervised regression machine learning. Since the proposed invention considers the flow rate measured at certain points and the macroscopic data of the road network for machine learning, the vehicle flow rate model is accurate and representative.

[0063] According to an exemplary embodiment, the vehicle flow model DM can be written:

[0064] DM(t)^f (Top, slope, Vmoy, t) with t the time considered (hours, day, month, etc.), Top the topology of the road section, slope the slope of the road section, Vmoy the average speed on the road section. Top, slope are examples of macroscopic data. In general, the function f can depend on other macroscopic data of the road network such as those listed previously (for example, the type of road section, number of lanes of the road section, maximum speed and length of the road section).

[0065] The function f can be determined from a supervised regression learning algorithm, such as a support vector machine (SVM), a neural network, a random forest or a combination of these methods. As a non-limiting example, the method can implement a neural network with two hidden layers of twenty and fifteen neurons respectively, or can implement a random forest with twenty estimators (or trees) of depth equal to 15, or any equivalent algorithm.

[0066] According to one aspect, this step may include a validation method, preferably a cross-validation method, in particular a k-fold cross-validation method. This cross-validation method makes it possible to reduce overfitting problems and improve the accuracy of the model.

[0067] Advantageously, the flow model can take as input topographical, temporal, population, and congestion data in order to estimate the vehicular flow on each of the strands of the study area, as well as measurements with fixed sensors at certain points of the road network. The congestion data typically correspond to the average speed observed. For this implementation, the different variants described in patent application FR3129021 can be implemented.

[0068] According to one implementation, it is possible to use additional data such as the number of Floating Car Data (FCD) counts on each strand. As output, the model gives the flow predictions on each strand of the study area. This traffic model is capable of capturing daily variations with, for the same strand, a high flow and a low speed during peak hours and a low flow and a high speed during off-peak hours. For this implementation, the different variants described in patent application FR3134472 can be implemented.

[0069] C- Determination of a unit quantity of pollutants

[0070] During this step, a unit quantity of pollutants emitted by a single vehicle is determined for each strand of the road network, for each driving style and for each thermal state of the vehicle. From the average speed acquired in step A, this step implements a plurality of unit pollutant models: one model for each driving style and each thermal state of the vehicle. The vehicle flow model is a model that links the vehicle flow, the average vehicle speed and the macroscopic data of the strand considered. Thus, each unit pollutant model has as input an average speed of each strand of the road network and macroscopic data of the strand and as output a unit quantity of pollutants.

[0071] According to one embodiment, this step can optionally implement (for each driving style and for each vehicle thermal state): - A mesoscopic unitary model of pollutant emissions, and / or - A velocity profile model, and a microscopic unitary model of pollutant emissions.

[0072] A microscopic model is a precise model taking as input an instantaneous speed and having as output an instantaneous unit quantity of pollutants emitted, which may be based on physics. This may be, for example, the CMEM model of the prior art. Such a model requires significant computing resources for use over a large territory, but remains suitable for a territory of limited size (for example, a few streets). A mesoscopic model is a model taking as input an average speed and macroscopic data from the road network, and having as output an average unit quantity of pollutants. Such a mesoscopic model is more precise than a macroscopic model, such as the COPERT model. The speed profile is the variation in the speed of the vehicle along a road in a road network.

[0073] When this step has the capacity to implement the mesoscopic unitary model of pollutant emissions and the unitary model of pollutant emissions, the choice between the microscopic model and the microscopic model may be dependent on the size of the road network considered. For example, for a road network of large size (representing for example an entire city) the method may implement the mesoscopic model, and for a network of limited size (for example a district of a city) the method may implement the microscopic model. A dimensional threshold may be predetermined for the automatic choice of the type of model. Thus, the method may be implemented at different road network scales, and may optimize the computing resources (memory and processors) by using the more precise model if the size of the road network is limited, in order to improve the accuracy of the unitary quantity of pollutant emissions determined.

[0074] [Fig. 3] illustrates, schematically and in a non-limiting manner, the stages of the method according to this embodiment of the invention. The steps identical to the embodiment of [Fig.l] are not re-described in detail. For this embodiment, the determination of the unit quantity Qu of the quantity of pollutants emitted by the unit emission model MUE is implemented by a mesoscopic unit model MUM of the pollutant emissions, and / or by a speed profile model PRO and a microscopic unit model MUp of the pollutant emissions. The type of unit model used may depend in particular on the size of the road network considered, this step of selecting the type of unit model is illustrated by a diamond. In addition, this figure illustrates the consideration of a PAU vehicle fleet, this optional implementation will be detailed with step D.

[0075] [Fig.4] illustrates, schematically and in a non-limiting manner, the step of determining pollutant emissions using the MUM mesoscopic model of pollutant emissions. For the illustrated embodiment, three styles of driving are considered: flexible, medium and aggressive and two thermal states: hot or cold post-treatment system. Thus, this step includes six models: - A MUMijF model for flexible driving style and cold aftertreatment system; - A MUM] C model for smooth driving style and hot aftertreatment system; - A MUM2jF model for average driving style and cold aftertreatment system; - A MUM2jC model for average driving style and hot aftertreatment system; - A MUM3jF model for aggressive driving style and cold aftertreatment system; - A MUM3jC model for aggressive driving style and hot aftertreatment system.

[0076] Each of these models has as input the average speed VIT and as output a unit quantity Qu of pollutants emitted by a single vehicle.

[0077] For this implementation, the mesoscopic unitary model can be constructed by a multivariable regression per vehicle, per pollutant, per driving style and, for a thermal state which gives the unit emissions per strand. Each mesoscopic unitary pollutant model has as input an average speed of each strand of the road network and macroscopic data of the strand and as output a unit quantity of pollutants. This regression can be carried out using analytical methods, machine learning or deep learning, preferably by supervised machine learning, preferably by supervised regression machine learning.This machine or deep learning is trained on a learning base formed on a learning road network comprising a unit quantity of pollutants emitted within said learning road network, macroscopic data of said strands of said learning road network, and the average speed of vehicles on said strands of said learning road network. The learning road network is a road network for which a unit quantity of pollutants emitted is known for the construction of the mesoscopic unit model. The learning road network may be different or identical to the road network considered for the implementation of the method. Where appropriate, the learning network may be identical to the learning network implemented in step B.

[0078] According to an exemplary embodiment, the mesoscopic unitary model of QM pollutant emissions can be written:

[0079] Q = g {Top, slope, Vmoy, Vlirn ) with Top the topology of the road strand, slope the slope of the road strand, Vmoy the average speed on the road strand and Viim the speed limit on the strand. Top, slope are examples of macroscopic data. Generally speaking, the function g may depend on other macroscopic data of the road network such as those listed previously (for example, the type of road strand, the number of lanes of the road strand, the maximum speed, the curvature of the road strand - in particular for the determination of the quantity of particulate emissions, and the length of the road strand).

[0080] The function g can be determined from a supervised regression learning algorithm, such as a support vector machine (SVM), a neural network, a random forest or a combination of these methods. As a non-limiting example, the method can implement a neural network with two hidden layers of twenty and fifteen neurons respectively, or can implement a random forest with twenty estimators (or trees) of depth equal to 15, or any equivalent algorithm.

[0081] According to one aspect, this step may include a validation method, preferably a cross-validation method, in particular a k-fold cross-validation method. This cross-validation method makes it possible to reduce overfitting problems and improve the accuracy of the model.

[0082] For example, the determination of the mesoscopic unitary model can implement the steps described in patent application FR3122011 for each driving style and for each thermal stage of the vehicle.

[0083] For the embodiment implementing the determination of a speed profile and a microscopic unitary model of pollutant emissions, the speed profile can be determined by a multivariable regression per vehicle, per pollutant, per driving style and, for a thermal state. This regression can be carried out using analytical methods, machine learning or deep learning, preferably by supervised machine learning, preferably by supervised regression machine learning. This machine or deep learning is trained on a learning base formed on a learning road network comprising an instantaneous speed of the vehicles within said learning road network, macroscopic data of said strands of said learning road network, and the average speed of the vehicles on said strands of said learning road network.The learning road network is a road network for which instantaneous speeds are known. of vehicles for the construction of the mesoscopic unit model. The learning road network may be different or identical to the road network considered for the implementation of the method. If necessary, the learning network may be identical to the learning network implemented in step B.

[0084] According to an exemplary embodiment, the speed profile model Vpro can be written:

[0085] Vpm( ​​t ) = h( Top, slope, Vmoy, t) with Top the topology of the road strand, slope the slope of the road strand, Vmoy the average speed on the road strand and t the time. Top, slope are examples of macroscopic data. Generally speaking, the function h can depend on other macroscopic data of the road network such as those listed previously (for example, the type of road strand, the number of lanes of the road strand, the maximum speed and the length of the road strand).

[0086] The function h can be determined from a supervised regression learning algorithm, such as a support vector machine (SVM), a neural network, a random forest or a combination of these methods. As a non-limiting example, the method can implement a neural network with two hidden layers of twenty and fifteen neurons respectively, or can implement a random forest with twenty estimators (or trees) of depth equal to 15, or any equivalent algorithm.

[0087] According to one aspect, this step may include a validation method, preferably a cross-validation method, in particular a k-fold cross-validation method. This cross-validation method makes it possible to reduce overfitting problems and improve the accuracy of the model.

[0088] For example, the determination of the speed profile can implement the steps described in patent application FR3096822 (US2022 / 0215749), or FR3115256 (US2022 / 0118986), or FR3130432 for each driving style and for each thermal state of the vehicle.

[0089] Advantageously, the microscopic unitary model of pollutant emissions can be constructed from a modeling of the physics of a vehicle (for example, type of engine, post-treatment system, etc.). This microscopic unitary model links a speed profile of a vehicle to a quantity of pollutants emitted by the vehicle. Thus, each microscopic unitary model of pollutants has as input a speed profile of each strand of the road network and macroscopic data of the strand and as output a unitary quantity of pollutants.

[0090] Alternatively, the speed profile can be determined using machine learning or deep learning methods, preferably by supervised machine learning, preferably by supervised regression machine learning. This machine or deep learning is trained on a learning base trained on a network learning road network comprising an instantaneous speed of vehicles within said learning road network, macroscopic data of said strands of said learning road network, and a quantity of pollutant emissions emitted on said strands of said learning road network. The learning road network is a road network for which instantaneous vehicle speeds are known for the construction of the mesoscopic unit model. The learning road network may be different from or identical to the road network considered for the implementation of the method. Where appropriate, the learning network may be identical to the learning network implemented in step B.

[0091] According to an exemplary embodiment, the unitary model of pollutant emissions QMU[i can be written:

[0092] Mmup = j{Top, slope, Vpro(t)) with Top the topology of the road strand, slope the slope of the road strand, Vpro the speed profile on the road strand and t the time. Top, slope are examples of macroscopic data. Generally, the function h can depend on other macroscopic data of the road network such as those listed previously (for example, the type of road strand, the number of lanes of the road strand, the maximum speed and the length of the road strand).

[0093] The function j can be determined from a supervised regression learning algorithm, such as a support vector machine (SVM), a neural network, a random forest or a combination of these methods. As a non-limiting example, the method can implement a neural network with two hidden layers of twenty and fifteen neurons respectively, or can implement a random forest with twenty estimators (or trees) of depth equal to 15, or any equivalent algorithm.

[0094] According to one aspect, this step may include a validation method, preferably a cross-validation method, in particular a k-fold cross-validation method. This cross-validation method makes it possible to reduce overfitting problems and improve the accuracy of the model.

[0095] For example, the determination of the microscopic unit quantity of pollutant emissions may implement the CMEM method of the prior art, or correspond to the steps of the method described in patent application FR3122011 (US 2022 / 0335822). Such a model may be calibrated with real-life driving conditions to give a reliable estimate of the instantaneous quantity of pollutants emitted on the road section considered.

[0096] D - Determination of an aggregate quantity of pollutants

[0097] During this step, an aggregate quantity of said at least one pollutant emitted by all vehicles is determined for each strand of the road network, based on: - From the flow of vehicles determined in step B, - From the unit quantity of pollutant emissions determined in step C, - From the distribution of the thermal states of the vehicles, - On the distribution of driving styles.

[0098] In other words, the unit quantity of pollutants emitted being determined in step C for each thermal state and each driving style, each unit quantity is multiplied by the representative part of the driving style and the thermal state to obtain a unit quantity representative of all the driving styles and all the vehicle thermal states. Then this representative unit quantity is multiplied by the vehicle flow rate determined in step B, to take into account the number of vehicles, and obtain an aggregated quantity for all the vehicles.

[0099] According to an embodiment option, the determination of the aggregated quantity of pollutants emitted by the set of vehicles can also take into account the vehicle fleet used for the road network considered. The vehicle fleet can be the current fleet of vehicles crossing the road network considered, in this case the learning road network. It can be defined by the user based on the recording histories and / or prior knowledge of the vehicle fleet in the area considered (i.e. the road network). Thus, the predefined fleet is a distribution, in number or percentage of vehicles, of each predetermined vehicle traveling on the portion of the road network. Within the vehicle fleet, the vehicles are categorized, the vehicle category can include in particular a European standard for pollutant emissions, a cylinder capacity, a type of engine (gasoline, diesel, electric, etc.), and a post-processing technology.This breakdown of the vehicle fleet can be carried out for private vehicles, heavy goods vehicles, light utility vehicles, two-wheelers, etc. By taking the vehicle fleet into account, the determination of pollutant emissions is more representative of real conditions.

[0100] As a non-limiting example, it is possible to create for the vehicle fleet of the road network an average vehicle which is the linear combination of known vehicles. These vehicles can in particular be divided into four categories: light vehicles, light utility vehicles, heavy goods vehicles and two-wheelers. Within these categories the vehicles are sorted into subcategories in particular according to their engine, their emission standard (for example, Euro 4, Euro 5 or Euro 6), their weight and / or their post-treatment device. From these categories, subcategories and the distribution of the vehicles (portion of vehicles in each category and subcategory), it is possible to define an average vehicle which is representative of the entire fleet. This average vehicle can evolve according to policies on access to old vehicles, heavy goods vehicle bans. Then, the characteristics are applied (weight, emissions standard, etc.) of this average vehicle to the multiplication of the unit quantity of pollutant emissions by the vehicle flow rate, to make the aggregate quantity of pollutant emissions as accurate as possible.

[0101] [Fig.2] illustrates, schematically and in a non-limiting manner, the steps of the method according to this implementation. The steps identical to those of the embodiment of [Fig.l] are not detailed again. For this configuration, the step of determining DET the aggregated quantity of pollutants emitted takes into account a distribution of the PAU vehicle fleet.

[0102] It is recalled that [Fig.3] described above also optionally includes this implementation.

[0103] The method may further comprise an optional step of displaying the quantity of pollutants determined for at least one strand of the road network considered. During this optional step, the quantity of pollutants emitted may be displayed on a road map. This display may take the form of a rating or a color code. Where appropriate, a rating or a color may be associated with each strand of the road network. This display may be carried out on board a vehicle: on the dashboard, on a portable autonomous device, such as a geolocation device (GPS type), a mobile phone (smartphone type), or any similar system. It is also possible to display the quantity of pollutant emissions on a website. In addition, the quantity of pollutants emitted may be shared with public authorities (for example, road managers) and public works companies.In this way, public authorities and public works companies can optimize road infrastructure to improve pollutant emissions. Examples

[0104] The characteristics and advantages of the method according to the invention will appear more clearly on reading the application example below.

[0105] For the first example, we consider a section of a road network, on which a Euro 5 diesel vehicle is traveling, and we determine the quantity of CO2 emitted by the vehicle on the section as a function of the average traffic speed: - Using the prior art COPERT macroscopic model (this model does not depend on the limit speed of the road strand considered), - By means of the mesoscopic unitary model of pollutant emissions, according to an embodiment of step C of the method according to the invention, for a limit speed of the road section at 70 km / h, - By means of the mesoscopic unitary model of pollutant emissions, according to an embodiment of step C of the method according to the invention, for a limit speed of the road section at 130 km / h.

[0106] The determination based on the average traffic speed makes it possible to take into account the congestion of the strand.

[0107] [Fig.5] illustrates, for this example, curves of CO2 emissions in g / km as a function of traffic speed V in km / h. The line corresponds to the emissions determined by the COPERT method according to the prior art. The points correspond to the emissions determined by the unitary emissions model with a speed limit of 70 km / h, and the crosses correspond to the emissions determined by the unitary emissions model with a speed limit of 130 km / h.

[0108] It is noted that the mesoscopic model takes congestion into account. Indeed, two strands having the same descriptors apart from their limiting speed, for the strand with the lowest limiting speed, i.e. the least congested strand, the emissions are lower than for the strand with the highest limiting speed. The macroscopic COPERT model, for its part, does not take congestion into account and underestimates emissions, especially when congestion is high. Thus, the method according to the invention allows a precise determination of the aggregated quantity of pollutants emitted.

[0109] The second example concerns the determination of pollutant emissions for a road network by means of the mesoscopic model according to step C, with a distinction of three driving styles: gentle, medium and aggressive, and by means of the macroscopic model COPERT.

[0110] Figures 6A to 6D represent, schematically and in a non-limiting manner, the quantity of pollutants emitted on a road network (in this case the quantity of NOx). In this figure, the darker the road strand, the higher the quantity of NOx emitted. Figure 6A corresponds to the gentle driving style, Figure 6B corresponds to the average driving style, Figure 6C corresponds to the aggressive driving style and Figure 6D corresponds to the quantity determined by the COPERT method. It is noted that the emissions with gentle driving are lower than with average driving, and with average driving lower than with aggressive driving. The COPERT method is close to average driving.

[0111] Table 1 indicates for this example the quantity of pollutants emitted (NOX, CO2 and PMhe particles) for these four configurations.

[0112] [Tables 1] Mild Medium Aggressive COPERT NOx (t / year) 33 43 57 44 CO2 (kt / year) 15 19 23 19 PMhe (t / year) 17 32 36 31

[0113] We also note that the COPERT method is close to a style of average driving.

[0114] For the third example, the method according to the invention is applied to an infrastructure modification for which a street currently passing through an underground road is leveled and new intersections are added. Thus, several strands which had a negative slope for the descent into the underground road and a positive slope for the ascent are now at zero slope. Several traffic lights and give way signs have been added to manage the new intersections. It is considered that the speed behavior in a hopper is different from a speed behavior without a hopper. The behavior of the successor strands of the hopper is applied to the modified strands.

[0115] The influence of slope, traffic lights and speed change can be separated. In the flow model, the speed change influences the flow model.

[0116] Thus, there is a decrease in speed compared to the descent speed in the underground track and an increase in speed compared to the ascent. In this case at equal limit speed; the decrease in speed tends to increase the flow rate and increase emissions, and the increase in speed tends to decrease the flow rate and decrease emissions.

[0117] Similarly for the slope, the transition from a negative slope to a zero slope increases emissions while the transition from a positive slope to a zero slope decreases emissions. The addition of fires also increases emissions. Overall, the reduction in emissions linked to the reduction in flow is greater than the increase in emissions linked to the slope, for gains of the order of 20% in NOx and CO2.

[0118] In addition, to calm the city and limit pedestrian deaths in the event of a collision with a motor vehicle, it was decided to lower the speed limit in town from 50 km / h to 30 km / h in several large French cities. A reduction in the average speed of vehicle traffic is expected.

[0119] As shown in the curves of the state-of-the-art macroscopic model COPERT, a decrease in average speed below 50 km / h will tend to increase emissions (see [Fig.4]). The mesoscopic unitary model perceives the same phenomenon, but this phenomenon is mitigated by the fact that the mesoscopic model takes into account congestion represented by the difference between average speed and speed limit. Therefore, a decrease in speed limit will tend to decrease emissions if congestion via the calming of behavior also occurs.

[0120] The strength of the mesoscopic model according to the invention is that it takes into account the driving style. It is assumed that the reduction in speed will tend to calm the driving style.

[0121] For the modification of the average speed, a strong hypothesis of a 25% reduction in traffic speed is taken. The quantities of pollutant emissions are presented in Table 2 for different distributions of driving styles before and without modification of the speed limit, and in comparison with the COPERT method.

[0122] [Tables2] Configuration NOX [t / year] co2 [kt / year] PMhe [t / year] Before modification Nominal with hopper 192 58 88 COPERT with hopper 116 36 / Limited speed Distribution 1 204 (+6%) 62 (+7%) 82 (-7%) Distribution 2 191 (-0.5%) 59 (+2%) 73 (-17%) Distribution 3 180 (-6%) 55 (-5%) 62 (-29%) Distribution 4 175 (-9%) 54 (-7%) 57 (-35%) Distribution 4 171 (-11%) 53 (-27%) 51 (-42%) COPERT 121 (+4%) 38 (+5.5%) /

[0123] Distribution 1 corresponds to the distribution of the following driving styles: 25% gentle driving, 50% medium driving and 25% aggressive driving.

[0124] Distribution 2 corresponds to the distribution of the following driving styles: 40% gentle driving, 45% medium driving and 15% aggressive driving.

[0125] Distribution 3 corresponds to the distribution of the following driving styles: 60% gentle driving, 30% medium driving and 10% aggressive driving.

[0126] Distribution 4 corresponds to the distribution of the following driving styles: 80% gentle driving, 10% medium driving and 10% aggressive driving.

[0127] We note from this example the importance of taking into account the style of driving style to determine the quantity of pollutants emitted. Indeed, there are significant variations depending on the distribution considered. Thanks to the use of a unitary model of pollutant emissions by driving style, it is therefore possible to precisely determine the quantity of pollutants emitted.

Claims

1.

2.

3. Claims Method for determining a quantity (Q) of at least one pollutant emitted by a set of vehicles within a road network, said road network comprising a set of strands characterized by macroscopic data including the topography of said strands, the movement of said vehicles within said road network being characterized by at least two driving styles and by at least two thermal states of the vehicles, characterized in that the following steps are implemented: a. An average vehicle speed (VIT) is acquired for each strand of the said road network; b. A vehicle flow rate (Db) is determined for each strand of said road network from said average speed (VIT) acquired by means of a vehicle flow model (VFM) which links the vehicle flow rate, the average vehicle speed and said macroscopic data of said strand; c. For each strand, a unit quantity (Qu) of said at least one pollutant emitted by a single vehicle is determined for each driving style and for each thermal state of the vehicles, based on said acquired average speed (VIT) and a unit model of pollutant emissions (MUE); and d. An aggregate quantity (Q) of said at least one pollutant for all vehicles is determined for each strand of said road network, based on said determined vehicle flow rate (DB), said unit quantities (Qu) of said at least one pollutant emitted by a single vehicle, and based on a distribution of driving styles (DSC) of vehicles within the road network, and a distribution of said thermal states of vehicles (DTH) within the road network. Method according to claim 1, in which said aggregated quantity (Q) of at least one pollutant emitted for all vehicles is determined by further considering data from a vehicle fleet (PAU) associated with said road network considered. Method according to one of the preceding claims, in which said unitary pollutant emission model (MUE) is a meso-model mesoscopic unitary pollutant emissions model (MUM), said mesoscopic unitary pollutant emissions model (MUM) being a multivariable regression of the quantity of pollutant emissions by driving style and by thermal state of the vehicle.

4. The method of claim 3, wherein said unitary mesoscopic pollutant emission model (MUM) is constructed by a machine learning or deep learning method trained on a learning base formed on a learning road network comprising the average vehicle speed (VIT) on the strands of said learning road network, macroscopic data of said strands of said learning road network, and acquired data of a unit quantity of emissions of said at least one pollutant on said learning road network.

5. Method according to one of claims 1 or 2, in which a unit quantity of said at least one pollutant emitted by a single vehicle is determined by means of the following steps: i. A speed profile (PRO) of said single vehicle on said strand is determined for each strand, for each driving style, from said average speed and said macroscopic data of said strand; ii. For each strand, said unit quantity of said at least one pollutant is determined for each driving style and for each thermal state of said vehicle by means of a unit microscopic model of pollutant emissions (MUp) applied to said speed profile (PRO) of said single vehicle.

6. Method according to claim 5, wherein said speed profile (PRO) is constructed by means of a speed profile model constructed by a machine learning or deep learning method trained on a learning base trained on a learning road network, said learning base comprising the average vehicle speed on the strands of said learning road network, macroscopic data of said strands of said learning road network, and acquired instantaneous speed data of at least one vehicle on said strands of said learning road network.

7. Method according to claim 5 or 6, in which said unitary microscopic model of pollutant emissions (MUp) is constructed by a machine learning or deep learning method trained on a learning base trained on a learning road network comprising the speed profile of vehicles within said learning road network, macroscopic data of said strands of said learning road network, and acquired data of unit quantity of pollutant emissions on said strands of said learning road network.

8. Method according to one of the preceding claims, wherein said macroscopic data of a road network are chosen from the topology of the strand of the road network, the number of lanes of the strand of the road network, the maximum authorized speed of the strand of the road network, the slope of the strand of the road network, the signaling of the strand of the road network, the length of the strand of the road network, preferably said macroscopic data of said road network being provided by a geographic information system.

9. Method according to one of the preceding claims, in which said average vehicle speed (VIT) is acquired by measurement and / or by means of a geographic information system.

10. Method according to one of the preceding claims, in which said vehicle flow model (VFM) is constructed by a machine learning or deep learning method trained on a learning base trained on a learning road network comprising the flow of vehicles within said learning road network, macroscopic data of said strands of said learning road network, and the average speed of vehicles on said strands of said learning road network.

11. Method according to one of the preceding claims, in which said quantity (Q) of at least one pollutant emitted by the at least one vehicle on said at least one strand of said road network is displayed on a road map, preferably by means of a smartphone or a computer system.

Images