A Computer Tool for Modelling CO 2 Emissions in Driving Cycles for Spark Ignition Engines Powered by Biofuels

: A driving cycle is a record intended to reﬂect the regular use of a given type of vehicle, presented as a speed proﬁle recorded over a certain period of time. It is used for the assessment of engine pollutant emissions, fuel consumption analysis and environmental certiﬁcation procedures. Different driving cycles are used, depending on the region of the world. In addition, drive cycles are used by car manufacturers to optimize vehicle drivelines. The basis of the work presented in the manuscript was a developed computer tool using tests on the Toyota Camry LE 2018 chassis dynamometer, the results of the optimization process of neural network structures and the properties of fuels and biofuels. As a result of the work of the computer tool, the consumption of petrol 95, ethanol, methanol, DME, CNG, LPG and CO 2 emissions for the vehicle in question were analyzed in the following driving tests: Environmental Protection Agency (EPA US06 and EPA USSC03); Supplemental Federal Test Procedure (SFTP); Highway Fuel Economy Driving Schedule (HWFET); Federal Test Procedure (FTP-75–EPA); New European Driving Cycle (NEDC); Random Cycle Low ( × 05); Random Cycle High ( × 95); Mobile Air Conditioning Test Procedure (MAC TP); Common Artemis Driving Cycles (CADC–Artemis); Worldwide Harmonized Light-Duty Vehicle Test Procedure (WLTP).


Introduction
The dynamic development of technology, which the automotive industry has seen for many years, includes both achieving an appropriate level of vehicle performance and meeting appropriate environmental protection requirements [1][2][3][4]. Keeping exhaust gas emissions under the permissible limits is the basic criterion that determines the directions of further development of engines used to drive motor vehicles [5][6][7][8]. Increasingly restrictive legal regulations are introduced to protect the climate [9][10][11][12]. The European Union (EU) has long been setting ambitious climate goals, which will not be achievable without reducing greenhouse gas emissions in transport-which consumes a third of the energy in the EU [13][14][15]. It is the transport sector in the EU that accounts for almost 30% of total CO 2 emissions, 72% of which comes from road transport [16,17]. Passenger cars are responsible for 60.7% of all CO 2 emissions from road transport in Europe [18,19].
Additionally, in the United States, car exhaust gases are the main source of greenhouse gas emissions, thus causing climate change [20][21][22]. The local permissible exhaust emission standards are based on research by a federal US body-the Environmental Protection Agency (EPA) [23,24]. Greenhouse gas emissions from transport account for approximately 28 percent of total US greenhouse gas emissions [25,26].
In China, combustion tests are a mixture of the abovementioned European and American regulations [27,28]. Work is also underway on a new type of test, which will be even more complicated and will much better reflect actual conditions [29,30]. The Chinese transport sector is responsible for around 12% of domestic emissions [31][32][33][34].
Each new passenger car must meet exhaust gas toxicity standards before it is introduced to the market [35][36][37]. The test conditions depend on the vehicle class and the Esters (FAEE) and alcohols, mainly primary: methanol and ethanol; secondary, alcohol derivatives (mainly ethers); and liquid products of biomass processing Biomass to Liquid (BTL) [162][163][164].
Among the abovementioned renewable liquid fuels, ethanol and methanol warrant special attention.
Ethanol is obtained from plant products through the process of the fermentation of sugar. The largest disadvantage of ethanol is its low calorific value (30.4 kJ/g). In relation to a liter, this value is 1/3 lower than for petrol, i.e., 10 L of petrol corresponds to approx. 15 L of ethanol (the calorific value of petrol is 45.0 kJ/g). The octane number of this fuel can exceed 108. This enables an increase in the compression ratio or the boost pressure. Commercially, ethanol fuels are sold with the E prefix (e.g., E85 contains 85% ethanol and 15% petrol) [165][166][167].
Methanol is a technical alcohol that is obtained by the dry distillation of wood or evaporation of coal. Its properties are similar to ethanol, but it has a lower calorific value (20.1 kJ/g). The octane number of methyl alcohol exceeds even 110. A large part of its mass is occupied by oxygen, one atom of which is present in each methanol molecule. This means that its calorific value is much lower than that of petrol or ethanol. Methanol is also used to power speedway motorcycles equipped with engines with compression ratios exceeding 16 [168-172].
For many years, efforts have been made to develop dedicated tools for computer simulations of the analysis of the amount of pollutants emitted from motor vehicles.
An example of such a tool is the Vehicle Energy Consumption Calculation Tool (VECTO) [173][174][175]. The simulation tool launched by the European Commission is used to calculate the amount of fuel consumed and carbon dioxide emitted by brand new trucks. The tool calculates driving behavior, load capacity, vehicle configurations, axle configurations, vehicle weight, engine characteristics (engine capacity, fuel map and full load curve), aerodynamic drag and tire rolling resistance. The VECTO calculates the fuel consumption in liters per 100 km and the fuel consumption per ton-kilometer transported, as well as the CO 2 emissions. The program can affect the fuel efficiency of the fleet, due to its thorough analysis of fuel consumption in various vehicle configurations [176][177][178][179].
Another tool used as a fuel consumption simulator for passenger cars and delivery vans was CO 2 Mpas. It enabled a simulation run that showed the results that a given vehicle with WLTP tests would achieve in the NEDC test. The tool used correlation methods [180][181][182].
The literature describes tools for the analysis of pollutant emissions from bus fleets in urban areas [183]. The proposed solution uses the results of measurements made with on-board instrumentation and the calculation method to estimate the emissions and fuel consumption as a function of vehicle parameters and the operating cycle.
The aim of this work was to build a computer tool for simulating driving tests as a function of the consumption of selected fuels and biofuels and CO 2 emissivity. The developed tool is dedicated to vehicles with a spark ignition engine.

Materials and Methods
The list below contains a set of the most important quantities used in the calculations with the appropriate symbols and units ( Table 1).
The development of the simulation model for driving tests was based on the research of the Toyota Camry LE 2018 and published [184]. Table 2 below presents the most important technical parameters of the vehicle and the factors necessary to be used in driving tests and programs generating the required waveforms: vehicle speed, gear number, clutch engagement and pedal position. The values of the Ratio n/v coefficient for individual runs were calculated on the basis of the dependencies, including the data contained in [185]:

Building a Quantitative Model
In order to construct a quantitative model that would enable the calculation of the instantaneous value of the fuel flow as a function of engine speed, engine torque, transmission gear number and vehicle speed, published data were used, which were obtained during Toyota Camry LE 2018 tests on a chassis dynamometer [184]. Figure 1 presents the set of points obtained during vehicle tests on a chassis dynamometer, converted to the value of hourly fuel consumption as a function of engine speed and torque generated by the engine.
Energies 2021, 14, 1400 6 of 33 speed, engine load torque, vehicle speed, transmission gear number, fuel consumption, etc. These were recorded during the EPA's surveys every 0.1 s. A total of about 350,000 measurement points were used to build the neural model and verify its performance. To build the neural model, about 80% of the available data were used as a learning set, while about 20% of the data were used in the process of verifying the performance of the developed simulation. In order to construct a quantitative model of instantaneous fuel consumption as a function of engine rotational speed and its generated torque, structures of the "Multilayer Feedforward Backpropagation Network" neural networks with approximating properties were used. The neural network structure itself used (in the hidden layers) a non-linear F1(x) activating function determined by the dependency, and a linear F2(x) activating function (in the output layer), in the following form: In the learning process of the network, the Levenberg-Marquardt algorithm was used, the basis of which is the optimization process through finding the minimum value of the objective function defined as the average value of the sum of squared differences between the current values of the network outputs and the assigned values, in the following form: Figure 2 below shows a general scheme of the neural network structure that complies with the abovementioned relationships. The "Neural Network Module Version 3.0" library was applied within the Scilab 6.1.0 [187,188] numerical software environment in order to build the neural model. The EPA published data included measurement points from actual measurements of the vehicle under consideration on a chassis dynamometer for 6 road tests (UDDS, HWFET, US06, LA92, WLTC and NEDC), for which multiple test repetitions were also provided. These data in spreadsheet form contained instantaneous values of engine speed, engine load torque, vehicle speed, transmission gear number, fuel consumption, etc. These were recorded during the EPA's surveys every 0.1 s. A total of about 350,000 measurement points were used to build the neural model and verify its performance. To build the neural model, about 80% of the available data were used as a learning set, while about 20% of the data were used in the process of verifying the performance of the developed simulation.
In order to construct a quantitative model of instantaneous fuel consumption as a function of engine rotational speed and its generated torque, structures of the "Multilayer Feedforward Backpropagation Network" neural networks with approximating properties were used. The neural network structure itself used (in the hidden layers) a non-linear F 1 (x) activating function determined by the dependency, and a linear F 2 (x) activating function (in the output layer), in the following form: In the learning process of the network, the Levenberg-Marquardt algorithm was used, the basis of which is the optimization process through finding the minimum value of the objective function defined as the average value of the sum of squared differences between the current values of the network outputs and the assigned values, in the following form: Figure 2 below shows a general scheme of the neural network structure that complies with the abovementioned relationships. The "Neural Network Module Version 3.0" library was applied within the Scilab 6.1.0 [187,188] numerical software environment in order to build the neural model. In order to obtain a neural model with the highest possible extent of adjustment to the research data published by EPA [184], an optimization process of the selection of the neural network structure was carried out, which included the change in the number of input parameters, engine rotational speed, engine torque, vehicle gear number, vehicle speed and the change in the number of hidden neurons. In the optimization process, a scalar objective function was used, according to the following dependence: Figure 3 presents selected results of the optimization process for different network structures in question, which differ in the number of input parameters and the number of neurons in the hidden layer, which, in many iterations, obtained the greatest degree of adjustment to the research data.
For the subsequent stages of building a vehicle simulation in road tests, a neural network structure was selected with two inputs for the input signals, engine rotational speed and engine torque, as well as three neurons in the hidden layer.
The selected neural network structure, which was characterized by achieving the smallest relative error value during the learning process for the learning data set, was verified using verification data, which represented approximately 20% of the actual vehicle test data on the chassis dynamometer for the considered tests published by the EPA. Again, the relative error between the simulation fuel consumption result and the realworld test fuel consumption, calculated from Equation (5), did not exceed 0.4%. In order to obtain a neural model with the highest possible extent of adjustment to the research data published by EPA [184], an optimization process of the selection of the neural network structure was carried out, which included the change in the number of input parameters, engine rotational speed, engine torque, vehicle gear number, vehicle speed and the change in the number of hidden neurons. In the optimization process, a scalar objective function was used, according to the following dependence: Figure 3 presents selected results of the optimization process for different network structures in question, which differ in the number of input parameters and the number of neurons in the hidden layer, which, in many iterations, obtained the greatest degree of adjustment to the research data.
For the subsequent stages of building a vehicle simulation in road tests, a neural network structure was selected with two inputs for the input signals, engine rotational speed and engine torque, as well as three neurons in the hidden layer.
The selected neural network structure, which was characterized by achieving the smallest relative error value during the learning process for the learning data set, was verified using verification data, which represented approximately 20% of the actual vehicle test data on the chassis dynamometer for the considered tests published by the EPA. Again, the relative error between the simulation fuel consumption result and the real-world test fuel consumption, calculated from Equation (5), did not exceed 0.4%.

Theoretical Assumptions of the Driving Test Simulator
The published test results on the chassis dynamometer were obtained with the use of standard commercial 95 octane petrol fuel. The presumptions of the work conducted on the vehicle simulation in driving tests were to introduce a functionality that would enable the definition of the consumption of other fuels used to power spark ignition engines. With the use of the neural model (fNet), on the basis of the instantaneous values of the torque generated by the engine (Tengine) and the engine speed (ηengine), the instantaneous values of the fuel flow for petrol 95 are obtained from the following dependence: Then, the simulation calculates the calorific value, in the case of using a fuel other than petrol 95 or fuel mixtures from the relationship:

Theoretical Assumptions of the Driving Test Simulator
The published test results on the chassis dynamometer were obtained with the use of standard commercial 95 octane petrol fuel. The presumptions of the work conducted on the vehicle simulation in driving tests were to introduce a functionality that would enable the definition of the consumption of other fuels used to power spark ignition engines. With the use of the neural model (f Net ), on the basis of the instantaneous values of the torque generated by the engine (T engine ) and the engine speed (η engine ), the instantaneous values of the fuel flow for petrol 95 are obtained from the following dependence: Energies 2021, 14, 1400 9 of 33 Then, the simulation calculates the calorific value, in the case of using a fuel other than petrol 95 or fuel mixtures from the relationship: It was assumed in the calculations that, for the instantaneous load value arising from the rotational engine speed and the engine-generated torque, a stream of another fuel must provide the same amount of energy over time as in the case of petrol 95. The efficiency of operation in the case of an engine powered by other fuels remains the same as for petrol 95, for each given calculation point. In this case, the instantaneous stream of fuels other than petrol 95 is calculated from the following dependence: Table 3 presents the basic parameters of the fuels used in the simulation: Table 3. Basic parameters of the fuels used in the simulation [189][190][191][192][193]. The presented properties of CNG fuel refer to the mixture which is used to power vehicles in a compressed form to the value of about 20MPa, containing 96-98% of methane with a minimum amount of other polluting gases and water vapor. Figure 4 shows the waveforms of the instantaneous value of the specific fuel consumption as a function of the engine rotational speed and the engine-generated torque for 6 types of fuels used in the simulation (petrol 95, ethanol, methanol, DME, CNG and LPG).

Parameter
It was assumed in the calculations that, for the instantaneous load value arising from the rotational engine speed and the engine-generated torque, a stream of another fuel must provide the same amount of energy over time as in the case of petrol 95. The efficiency of operation in the case of an engine powered by other fuels remains the same as for petrol 95, for each given calculation point. In this case, the instantaneous stream of fuels other than petrol 95 is calculated from the following dependence: Table 3 presents the basic parameters of the fuels used in the simulation: Table 3. Basic parameters of the fuels used in the simulation [189][190][191][192][193]. The presented properties of CNG fuel refer to the mixture which is used to power vehicles in a compressed form to the value of about 20MPa, containing 96-98% of methane with a minimum amount of other polluting gases and water vapor. Figure 4 shows the waveforms of the instantaneous value of the specific fuel consumption as a function of the engine rotational speed and the engine-generated torque for 6 types of fuels used in the simulation (petrol 95, ethanol, methanol, DME, CNG and LPG).   In order to calculate the CO2 emissivity, the mass content of carbon in the analyzed fuel was calculated. This was performed on the basis of the available information on the chemical compositions of the individual mixture components, the mass content of the fuel in the mixture and the instantaneous fuel stream that resulted from the conditions of the engine operation conditions using the following relationship:

Driving Test Generator
The most labor intensive process was teaching neural network structures. In this study, in order to obtain an optimal neural model for determining the instantaneous value of fuel consumption as a function of the engine speed and torque, structures with 2 and 3 inputs and a variable number of neurons in the hidden layer (1-4) were used. The structure learning process for fixed inputs and number of neurons in the hidden layer were repeated at least 100 times with a fixed minimum number of learning epochs of 1000. In total, the process of learning neural network structures to select the most fitting model took about 13 h. However, the process of simulation by the selected neural model of selected driving tests took about several minutes. In the developed simulation, no correlation was made between the simulation time and the actual duration of the driving test. In order to calculate the CO 2 emissivity, the mass content of carbon in the analyzed fuel was calculated. This was performed on the basis of the available information on the chemical compositions of the individual mixture components, the mass content of the fuel in the mixture and the instantaneous fuel stream that resulted from the conditions of the engine operation conditions using the following relationship:

Driving Test Generator
The most labor intensive process was teaching neural network structures. In this study, in order to obtain an optimal neural model for determining the instantaneous value of fuel consumption as a function of the engine speed and torque, structures with 2 and 3 inputs and a variable number of neurons in the hidden layer (1-4) were used. The structure learning process for fixed inputs and number of neurons in the hidden layer were repeated at least 100 times with a fixed minimum number of learning epochs of 1000. In total, the process of learning neural network structures to select the most fitting model took about 13 h. However, the process of simulation by the selected neural model of selected driving tests took about several minutes. In the developed simulation, no correlation was made between the simulation time and the actual duration of the driving test.
Based on the collected data of operational parameters of the vehicle in question and using the "Gearshift calculation tool" [194,195] application, runs for simulation control were created for the following drive tests: • US 06-The US06 (SFTP) [196,197] [218][219][220][221].
Upon entering the complete information about the vehicle, the program is ready to generate the necessary waveforms in the time domain, which in turn enable the determination of the instantaneous operating parameters of the program in question. These waveforms were then exported to text files. The instantaneous waveforms of the following quantities were used in the further stages of the simulation: simulation time

Simulator
A driving test simulator was developed in OpenModelica v1.16.2, based on the analysis of the data created with the use of the "Gearshift calculation tool" programme, the results of the process of neural network structure optimization and the properties of the tested biofuels [222]. The simulator is made up of blocks that are responsible for individual functionalities, and its connection diagram is presented in Figure 5 below: • Drive tests generator (text files)-responsible for loading files with data that control the selected driving test process from a text file created with the use of the "Gearshift calculation tool" application. It is also responsible for converting the read data to other formats compatible with OpenModelica v1.16.2. The following parameters are then relayed to the following calculation modules of the simulation: engine speed, engine torque, vehicle speed; • Model of specific consumption (neural)-this block calculates the instantaneous values of petrol 95 mass flow and relays this parameter to the next block, based on the quantities which characterize the engine operating parameters: engine speed, engine torque and the prepared neural network structure; • Calculations of fuel and CO 2 mass flows-this block is responsible for calculating the streams of the tested biofuels which are necessary to power the engine in the driving test. This is achieved using the petrol 95 mass flow parameter and the fuel calorific value characteristic for the fuel in question calculated in the previous block. This block also calculates the CO 2 emission stream with the use of the carbon mass content property and the instantaneous fuel stream; • Calculation of driving test parameters-on the basis of the driving test parameters, this block calculates the distance covered by the vehicle during the test, the power generated by the engine and the mechanical energy generated during the test. • Calculation of driving test parameters-on the basis of the driving test parameters, this block calculates the distance covered by the vehicle during the test, the power generated by the engine and the mechanical energy generated during the test.

Results
Presented below are the processes of independent simulations of the selected Toyota Camry LE 2018 vehicle in the applied driving tests with changing fuels (petrol 95, ethanol, methanol, DME, CNG and LPG): • the results of the simulation work for the processed EPA test data, which are learning models for the neural network; • the results of the driving test simulator for the prepared drive tests (the "Gearshift Calculation Tool" application) in the form of vehicle speed graphs, distance travelled, engine speed, engine torque, engine power and mechanical energy used during the test; • the simulation results for the stream and final fuel consumption; • the simulation results for the stream and CO2 emissions for selected driving tests and selected fuels for the 2018 Toyota Camry LE vehicle; • the results of fuel consumption and carbon dioxide emissivity per 1 km of the distance travelled by the vehicle in the tests and per 1 kWh of the mechanical energy generated in the test.

Simulation Work Results for the Processed EPA Test Data
The published data from actual vehicle tests carried out by the EPA were used in order to verify the correct operation of the driving test simulator. The input data were so transformed that they could be fed into the simulator. As a result of the simulator's work, the instantaneous values of the key simulation parameters were obtained, which are hereby presented in the figures below ( Figure 6) [184].

Results
Presented below are the processes of independent simulations of the selected Toyota Camry LE 2018 vehicle in the applied driving tests with changing fuels (petrol 95, ethanol, methanol, DME, CNG and LPG): • the results of the simulation work for the processed EPA test data, which are learning models for the neural network; • the results of the driving test simulator for the prepared drive tests (the "Gearshift Calculation Tool" application) in the form of vehicle speed graphs, distance travelled, engine speed, engine torque, engine power and mechanical energy used during the test; • the simulation results for the stream and final fuel consumption; • the simulation results for the stream and CO 2 emissions for selected driving tests and selected fuels for the 2018 Toyota Camry LE vehicle; • the results of fuel consumption and carbon dioxide emissivity per 1 km of the distance travelled by the vehicle in the tests and per 1 kWh of the mechanical energy generated in the test.

Simulation Work Results for the Processed EPA Test Data
The published data from actual vehicle tests carried out by the EPA were used in order to verify the correct operation of the driving test simulator. The input data were so transformed that they could be fed into the simulator. As a result of the simulator's work, the instantaneous values of the key simulation parameters were obtained, which are hereby presented in the figures below ( Figure 6)

Simulation Work Results for the Driving Tests Performed
On the basis of the prepared input data, using the "Gearshift Calculation Tool" software, simulations of selected driving tests were carried out for the vehicle in question. Figure 7 below shows the waveforms of the instantaneous vehicle speed values in the test.

Simulation Work Results for the Driving Tests Performed
On the basis of the prepared input data, using the "Gearshift Calculation Tool" software, simulations of selected driving tests were carried out for the vehicle in question. Figure 7 below shows the waveforms of the instantaneous vehicle speed values in the test. These waveforms indicate a large variability of this parameter in the simulated tests, including mean values, the dynamics of changes and the changes in the distribution of the values over time. The simulated tests were also characterized by high variability of the execution time.  The instantaneous values of the power generated by the engine and the mechanical energy used during the test were calculated in the developed driving test simulator. Figure 9a,b show the waveforms of these parameters.   The instantaneous values of the power generated by the engine and the mechanical energy used during the test were calculated in the developed driving test simulator. Figure 9a,b show the waveforms of these parameters. The instantaneous values of the power generated by the engine and the mechanical energy used during the test were calculated in the developed driving test simulator. Figure 9a,b show the waveforms of these parameters.

Simulation Results for the Stream and Final Fuel Consumption for the Selected Driving Tests and Fuels
The instantaneous values of fuel streams and their mass consumption for the tests in question were calculated on the basis of the values of the petrol 95 stream calculated in the simulator, taking into account the calorific values of the other considered fuels. The figures below ( Figure 10) present a summary of the obtained waveforms of the instantaneous values of fuel flows and the mass consumption of fuels in the given driving test.

Simulation Results for the Stream and Final Fuel Consumption for the Selected Driving Tests and Fuels
The instantaneous values of fuel streams and their mass consumption for the tests in question were calculated on the basis of the values of the petrol 95 stream calculated in the simulator, taking into account the calorific values of the other considered fuels. The figures below ( Figure 10) present a summary of the obtained waveforms of the instantaneous values of fuel flows and the mass consumption of fuels in the given driving test.

Simulation Results for the Stream and Final Fuel Consumption for the Selected Driving Tests and Fuels
The instantaneous values of fuel streams and their mass consumption for the tests in question were calculated on the basis of the values of the petrol 95 stream calculated in the simulator, taking into account the calorific values of the other considered fuels. The figures below ( Figure 10) present a summary of the obtained waveforms of the instantaneous values of fuel flows and the mass consumption of fuels in the given driving test.

The Results of the Simulation of Carbon Dioxide Flux and Emissions for Selected Driving Tests and Fuels
As a result of the vehicle simulation processes performed for selected driving tests, including various fuels, the instantaneous values of the carbon dioxide flux and its emissivity during the test were obtained. The figures below ( Figure 11) show the results of the simulator work in the form of the waveforms of carbon dioxide streams and its emissivity while taking into account the fuels considered for individual simulated tests.

The Results of the Simulation of Carbon Dioxide Flux and Emissions for Selected Driving Tests and Fuels
As a result of the vehicle simulation processes performed for selected driving tests, including various fuels, the instantaneous values of the carbon dioxide flux and its emissivity during the test were obtained. The figures below ( Figure 11) show the results of the simulator work in the form of the waveforms of carbon dioxide streams and its emissivity while taking into account the fuels considered for individual simulated tests.

Discussion
The developed tool and the methodology used to build quantitative models of fuel consumption and CO2 emissivity of the selected vehicle as a function of engine load and vehicle speed might constitute the basis for the construction of road simulators. In this case, the simulation tool can be adapted to the operational parameters of a large set of vehicles that represent a given car market. Road simulators developed on the basis of the described tool will make it possible to obtain more precise emissivity values in road traffic than the adopted environmental estimates. Figure 12 presents the results of the simulator work for the considered fuels and driving tests in the form of the fuel consumption parameter per one kilometer driven in the test. For CNG fuel, the minimum value was achieved at the level of 32 g/km for the US

Discussion
The developed tool and the methodology used to build quantitative models of fuel consumption and CO 2 emissivity of the selected vehicle as a function of engine load and vehicle speed might constitute the basis for the construction of road simulators. In this case, the simulation tool can be adapted to the operational parameters of a large set of vehicles that represent a given car market. Road simulators developed on the basis of the described Energies 2021, 14, 1400 22 of 33 tool will make it possible to obtain more precise emissivity values in road traffic than the adopted environmental estimates. Figure 12 presents the results of the simulator work for the considered fuels and driving tests in the form of the fuel consumption parameter per one kilometer driven in the test. For CNG fuel, the minimum value was achieved at the level of 32 g/km for the US highway test, while the maximum value was obtained at the random cycle high (×95) ( Figure 13 presents the results of the simulator work for the considered fuels and driving tests in the form of the CO2 emission parameter per one kilometer driven in the test. For petrol 95, the minimum value was reached at 116 g/km for the US highway test, while the maximum value was obtained at the random cycle high (×95) (187 g/km).  Figure 13 presents the results of the simulator work for the considered fuels and driving tests in the form of the CO 2 emission parameter per one kilometer driven in the test. For petrol 95, the minimum value was reached at 116 g/km for the US highway test, while the maximum value was obtained at the random cycle high (×95) (187 g/km).   Figure 13 presents the results of the simulator work for the considered fuels and driving tests in the form of the CO2 emission parameter per one kilometer driven in the test. For petrol 95, the minimum value was reached at 116 g/km for the US highway test, while the maximum value was obtained at the random cycle high (×95) (187 g/km).    Figure 14 below shows the data obtained from the performed simulations of driving tests, including biofuels, in the form of the parameter of mass consumption of a given fuel per unit of mechanical energy produced (1 KWh). For petrol 95, the minimum value was achieved at the level of 486 g/kWh for the US 06 driving test, while the maximum value was obtained for the US SC03 test (1630 g/kWh).  Figure 15 presents the results of the simulator work for the considered fuels and driving tests in the form of the carbon dioxide emission parameter per unit of mechanical energy produced (1 kilowatt-hour). For petrol 95, the minimum value was achieved at the level of 1538 g/kWh for the US 06 driving test, while the maximum value was obtained for the US SC03 (5182 g/kWh).  Figure 15 presents the results of the simulator work for the considered fuels and driving tests in the form of the carbon dioxide emission parameter per unit of mechanical energy produced (1 kilowatt-hour). For petrol 95, the minimum value was achieved at the level of 1538 g/kWh for the US 06 driving test, while the maximum value was obtained for the US SC03 (5182 g/kWh).
ies 2021, 14, x FOR PEER REVIEW 23 of 33 Figure 14 below shows the data obtained from the performed simulations of driving tests, including biofuels, in the form of the parameter of mass consumption of a given fuel per unit of mechanical energy produced (1 KWh). For petrol 95, the minimum value was achieved at the level of 486 g/kWh for the US 06 driving test, while the maximum value was obtained for the US SC03 test (1630 g/kWh).  Figure 15 presents the results of the simulator work for the considered fuels and driving tests in the form of the carbon dioxide emission parameter per unit of mechanical energy produced (1 kilowatt-hour). For petrol 95, the minimum value was achieved at the level of 1538 g/kWh for the US 06 driving test, while the maximum value was obtained for the US SC03 (5182 g/kWh).

Conclusions
The paper presents a computer tool for simulating driving tests as a function of the consumption of selected fuels and biofuels and CO 2 emissivity, dedicated to vehicles with spark ignition engines. The basis for the work conducted was chassis dynamometer tests on the Toyota Camry LE 2018 vehicle.

•
Neural network structures characterized by approximation (regression) properties were used to build a model enabling the determination of instantaneous fuel consumption values as a function of engine rotational speed and torque produced by the engine. The process of learning these network structures used data from actual driving tests performed on a selected vehicle on a chassis dynamometer published by the EPA. After selecting the neural network structure that obtained the smallest value of relative error with respect to the data from real measurements, the verification of the obtained neural model was carried out using the verification data of real tests included in the EPA publication. • Based on the operational parameters analyzed with the use of the "Gearshift Calculation Tool" application, the results of the optimization process of the neural network structures and the properties of the biofuels in question, a driving test simulator was developed in the OpenModelica v1.16.2 program. Scilab 6.1.0 numerical software was then used to build the neural model.

•
The developed simulation tool used neural networks, whose learning processes used the Levenberg-Marquardt algorithm. An optimization process was carried out for various investigated network structures differing in the number of input parameters and the number of neurons in the hidden layer. The relative error between the model and actual data did not exceed 1%.

•
Twelve driving tests were analyzed in this study. These tests differed from one another in terms of the duration, speeds achieved by the vehicle and allowances for the use of any additional equipment in the vehicle (e.g., A/C).

•
When analyzing the consumption parameter of a given fuel per one kilometer driven in the test, the best results were achieved for CNG fuel, for which the minimum value was reached at 32 g/km for the US highway driving test, while the maximum value was obtained in the Random Cycle High test (×95) (52.0 g/km). The highest fuel consumption per one kilometer in the test was observed in the case of methanol in the Random Cycle High (×95) (129.4 g/km).

•
When considering the emissions of carbon dioxide per kilometer in the test, the highest values were recorded for petrol 95, where the minimum value was reached at 116 g/km for the US highway driving test, and the maximum value was obtained at a Random Cycle High (×95) (187 g/km). For CNG, the minimum value was reached for the US highway (87.7 g/km).

•
When analyzing the parameter of mass consumption of a given fuel per unit of mechanical energy produced (1 kilowatt hour) in the case of petrol 95, the minimum value was achieved at 486 g/kWh for the US 06 driving test, while the maximum value was obtained for the US SC03 (1630 g/kWh). The highest consumption was recorded for US SC03, also for DME (2507 g/kWh), ethanol (2667 g/kWh) and methanol (3573 g/kWh).

•
The developed computer tool could be the basis for the development of a method of identifying selected aspects of operating conditions and assessing the energy efficiency of vehicles with spark ignition engines powered by fuels and biofuels. • The research method described in the manuscript aims to obtain a simulation model to calculate instantaneous fuel consumption as a function of engine speed and engine torque produced. This method allows the simulation of vehicle operations under different load conditions and will potentially allow the calculation of fuel consumption and carbon emissions. This method can be used for many popular vehicle models in a given market. In the case of estimating carbon dioxide emissions for real facilities where vehicles move, e.g., road tunnels and large parking lots, a very large number of simulations of individual vehicles in real traffic can be used in a single simulation.
The use of such simulations will allow for the more precise selection of ventilation systems for such objects, which will prevent the increase in carbon dioxide content in the air.
Funding: The APC was funded by Institute of Mechanical Engineering, Warsaw University of Life Sciences.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: All data are presented in this article. Data sharing is not applicable to this article.

Conflicts of Interest:
The author declares no conflict of interests.