Description:Complete Specification:
The following specification describes and ascertains the nature of this invention and the manner in which it is to be performed.
Reference: The present invention comprises an improvement in, or a modification of, the invention claimed in the specification of the main patent application 202241037775 dated 30 June 2022.

Field of the invention:
The present invention relates to a computing device and method to estimate carbon dioxide (CO2) emission from vehicles.

Background of the invention:
The climate change is one of the most serious environmental concerns confronting humanity today. Carbon Dioxide, Nitrous Oxide, and Methane emissions have caused global temperatures to rise by 1 degree C since the industrial revolution. CO2 is the most significant contributor to the greenhouse effect, accounting for most emissions from the electricity, transportation, and industry sectors For a long time, transportation has been a significant source to CO2 emissions In India, transportation accounts for 23.6% of total CO2 emissions, with road transport accounting for 65.4%. The introduction of zero emission electric cars is a hopeful step towards a more sustainable strategy to reduce surplus CO2 in the environment. In order to manage or restrict the influence on the climate, it is currently required to monitor the amount of CO2 produced by vehicles.

The introduction of zero emission electric cars is a hopeful step towards a more sustainable strategy to reduce surplus CO2 in the environment. In order to manage or restrict the influence on the climate, it is currently required to monitor the amount of CO2 produced by vehicles. A high spatiotemporal resolution emission monitoring is a topic of debate among researchers. To do this, a few techniques combine infrequent air pollution monitoring with assessments of heavy traffic flow or vehicle density. The latter extrapolates the cars officially reported CO2 readings to forecast emissions. However, there is a 30-40% discrepancy between theoretical values and actual emissions, hence this is not a reliable monitoring technique. The major cause of the disparity is the mismatch between actual driving behavior and that which was predicated in the test methods

In the literature, several sorts of models have been utilized. A white box models are detailed physical models that are typically created by automobile or engine manufacturers. They are extremely precise and clear. They do, however, need comprehensive information that is typically unavailable, such as engine friction or pumping losses. A grey box models combine basic physical models with data from controlled tests. The Comprehensive Modal Emissions Model (CMEM) which is frequently used in traffic simulation, is the most prominent grey box model. The accuracy of CMEM is mostly determined by vehicle speed and data integrity When using average speed profiles, for example, the estimated fuel consumption might be less than half of the actual. A black box models rely only on data, which is often collected via realistic driving experiences. The models are divided into two categories average and immediate fuel usage explored the prediction of average fuel usage using Back Propagation (BP), Neural Networks (NN) and Radial Basis (RB) neural networks. As input variables, both NNs utilized the car's make, weight, engine style, vehicle, and transmission type.

Several fuel consumption and emissions models have been developed in the past that include the impact of road infrastructure (e.g. gradient, roughness, and macrotexture). However, most of these models base their estimates on standard drive cycles, are calibrated only for specific vehicles, or offer only a simplified mechanistic model.

According to a prior art 202241026881 a design system of IoT driven to estimating CO2 gas emissions using machine learning for smart city is disclosed. Because of the fast spread of the Internet of Things (IoT) in the 5G era, the establishment of smart cities throughout the world is an important avenue for global low-carbon city research. Carbon dioxide emissions in small cities are frequently greater than in large and medium cities. Due to enormous disparities in data settings between small and medium-large cities, insufficient IoT hardware, and high input costs, the development of a small city smart carbon monitoring platform has yet to be completed. CO2 emissions from fossil fuel burning are a substantial contributor to atmospheric changes and climate change. General CO2 estimations can be estimated by considering the weight moved, distance travelled, time spent, and emissions factor of each individual vehicle, and (ii) by considering the weight moved, distance travelled, time spent, and emissions factor of each unique vehicle. These approaches are inefficient for measuring the influence of automobiles on CO2 in cities since vehicles emit CO2 in a variety of ways depending on their efficiency, producer, fuel, weight, driver style, road conditions, seasons, and other factors. Thanks to today's technology, it is now possible to collect real-time traffic data to obtain critical information that may be used to measure changes in carbon emissions. The research investigated the sources of CO2 emissions in the air under different traffic conditions. We present a model and method for computing CO2 emissions from traffic flow data in both non-congested and congested environments. These traffic scenarios each contribute a different amount of CO2 to the atmosphere, resulting in a different emissions factor. The system was tested in the city of Florence, where CO2 levels are measured by sensors at a few places with over 100 traffic flow sensors (data accessible on the Snap4City platform). The solution allowed for the assessment of seasonal fluctuations in the emissions factor and accuracy, as well as the computation of CO2 from the traffic flow. The provided model and solution might be used to calculate the CO2 distribution in the city.

Brief description of the accompanying drawings:
An embodiment of the disclosure is described with reference to the following accompanying drawings,
Fig. 1 illustrates a block diagram of a computing device to estimate carbon dioxide (CO2) emission from vehicles, according to an embodiment of the present invention;
Fig. 2 illustrates a block diagram for the estimation model in the computing device, according to an embodiment of the present invention, and
Fig. 3 illustrates a method for estimating carbon dioxide (CO2) emission from vehicles, according to the present invention.

Detailed description of the embodiments:
Fig. 1 illustrates a block diagram of a computing device to estimate carbon dioxide (CO2) emission from vehicles, according to an embodiment of the present invention. The computing device 120 configured to receive input signals comprising operational parameters 102 from an internal network 104 such as a Controller Area Network (CAN) bus of the vehicle 100 or sensors directly interfaced with the computing device 120. The computing device 120, characterized in that, configured to process the operational parameters 102 using an estimation model 116. The estimation model 116 is based on a Long Short-Term Memory (LSTM) network. The computing device 120 then estimates the CO2 emission from the vehicle 100. The computing device 120 further configured to compare the estimated CO2 with a threshold CO2 level 118 stored in a memory element 106. The computing device 120 then triggers an action 114 if the estimated CO2 exceeds the threshold CO2 level 118.

According to an embodiment of the present invention, the operational parameters 102 is selected from engine load, wheel speed, accelerator pedal position and engine speed. Moreover, each of the operational parameters 102 are measured using respective sensors in the vehicle 100 or derived using measured parameters.

According to an embodiment of the present invention, the computing device 120 applies LSTM network to avoid vanishing gradient issue found in Recurrent Neural Network (RNN). The vehicle 100 comprises cars, trucks, buses, off-road vehicles 100 for agriculture, etc. The LSTM based estimation model 116 predicts/estimates the present CO2 emissions which outperforms several other prediction techniques.

In accordance to an embodiment of the present invention, the computing device 120 is at least one selected from a group comprising an internal device comprising an Electronic Control Unit (ECU) 110 of the vehicle 100, and an external device comprising a cloud 108 based device and a communication device 112 and a combination thereof. The internal device denotes that the computing device 120 is internal or part of the vehicle 100. Similarly, the external device denotes that the computing device 120 is externally interfaced with the vehicle 100 and is generally not part of the vehicle 100. The ECU 110 (or controller) is at least one of an Engine Management System (EMS) controller, a Tire Pressure Monitoring System (TPMS) controller, a Telematics Control Unit (TCU) controller, Anti-lock Braking System (ABS) ECU 110, a Body Control Unit (BCU), a Human-Machine Interface (HMI) cluster unit, other vehicular controllers, and a combination thereof. The communication device 112 corresponds to electronic computing devices such as smartphone, wearable electronics such as smart watch, intelligent HMI cluster (or connected cluster) in the vehicle 100 etc. The cloud 108 based device corresponds to cloud computing architecture having single or network of servers, databases connected with each other and vehicle 100 for processing of inputs and providing outputs.

According to the present invention, the computing device 120 is provided with necessary signal detection, acquisition, and processing circuits. The computing device 120 is a controller/ control unit which comprises memory element 106 such as Random Access Memory (RAM) and/or Read Only Memory (ROM), Analog-to-Digital Converter (ADC) and a Digital-to-Analog Convertor (DAC), clocks, timers, counters and at least one processor (capable of implementing machine learning) connected with each other and to other components through communication bus channels. The memory element 106 is pre-stored with logics or instructions or programs or applications or modules/models and/or threshold values (threshold CO2 level 118), which is/are accessed by the at least one processor as per the defined routines. The internal components of the computing device 120 are not explained for being state of the art, and the same must not be understood in a limiting manner. The computing device 120 may also comprise communication units to communicate with an external computing device such as the cloud 108, a remote server, etc., through wireless or wired means such as Global System for Mobile Communications (GSM), 3G, 4G, 5G, Wi-Fi, Bluetooth, Ethernet, serial networks, and the like. The computing device 120 is implementable in the form of System-in-Package (SiP) or System-on-Chip (SOC) or any other known types.

According to an embodiment of the present invention, when the external device is used as the computing device 120, the input signals are received either through Telematics Control Unit (TCU) of the vehicle 100 or through wireless transceiver connected to an On-Board Diagnostic (OBD) port of the vehicle 100 or both.

According to the present invention, the action 114 comprises sending an alert to a display screen in the vehicle 100, or to the communication device 112 of a concerned stakeholder, or to government/regulatory offices or private emission management servers of Original Equipment Manufacturers (OEMs) and the like. The action 114 may also comprise curtailing the CO2 emission from the vehicle 100.

According to the present invention, a working of the computing device 120 is envisaged. Consider the OBD II interface is used as the means of accessing the data of the vehicle 100. The data is collected and transferred through a Bluetooth wireless dongle based connection to the communication device 112, where the data is processed and registered. The sensor data are observed once every second or other frequency. Consider the estimation model 116 is located in the communication device 112. The estimation model 116 processes the input signals of the operational parameters 102 based on the window length and provides instantaneous output.

Fig. 2 illustrates a block diagram for the estimation model in the computing device, according to an embodiment of the present invention. Following is a breakdown of each of the architectural components displayed in the Figure. “C" stands for cell and "t” stands for time. As at this moment, there are two inputs, x(t) is the current input and h(t-1) is the output of the prior input at (t-1) time. In the Fig. 2, symbols are used instead of reference numerals to explain the working of LSTM network and the same must not be understood in limiting manner. “σ” stands for Sigma for sigmoid activation function, and “tanh” refers to hyperbolic tangent function. The sigmoid and tanh functions collaborate to control the flow of input through the LSTM network, enabling it to selectively update its memory while preserving cell state values within a tolerable range. This makes the LSTM network particularly useful for activities requiring long-term memory as well as the capacity to selectively forget or retain information over time.

The Internal state c(t) is the most important component of an LSTM network. The major flow of information occurs through the cell’s internal state. The state is the LSTM system's memory, and it includes learning from all prior data. The state may alternatively be seen as a stream of information from which information is withdrawn or added. The input and outflow of information is regulated through gates. The gates are a unique element of LSTM network that determine what information and how much should travel through. An LSTM gate is essentially a sigmoidal unit triggered by the current input x(t) and the hidden layer from the previous time step.

A forget gate (ft) is now described. First, the LSTM determines which information should be deleted from memory. This choice is made by the sigmoidal unit known as the forget gate. It takes in x(t) and h(t-1) and returns a value between 0 and 1 for each state. This is multiplied by the internal state of the cell to delete information.

An Input Gate (It) and Node is described. In order to continue learning, new information must be introduced to the cell state. The Input Gate does this. A sigmoidal unit is used by an input gate to determine how much information to store. In the meanwhile, an Input Node constructs a vector of the new candidate states. After being multiplied by the output of the Input gate, these states are added to the cell state.

An Output Gate Ot is described. A selection of what information will be delivered as output for the following cell is to be selected. For this, the output gate Ot is utilized. After activating the cell state using tanh activation, the result of the output gate is multiplied by the cell's final output.

The LSTM network may thus carry an important characteristic over a lengthy period. Given that it is a time series based estimate and that the estimation model 116 may train the estimation more effectively by remembering prior inputs, LSTM is thus perfect for the prediction of a vehicle's 100 instantaneous fuel consumption and calorific value (CV).

According to an embodiment of the present invention, a training of the estimation model 116 is disclosed. The training is divided into three stages, feature sequence formation, feature selection, and LSTM estimation. The feature sequence formation/generation step involves data pre-processing and sequencing. The feature selection stage comprises performing optimal feature subset selection. Finally, utilizing the data sequences, LSTM is trained to predict instantaneous vehicle fuel consumption and emissions (i.e. CV). The stages/phases specifics are briefly explored. The proposed estimation model 116 is capable of being deployed on the cloud 108, with the vehicle 100 sending the data at fixed intervals (equal to the window size of the estimation model 116), and the cloud 108 making use of the forward prop of the LSTM to respond with an estimated instantaneous vehicle fuel consumption and emissions (CV) of the vehicle 100. One sample per second is the sampling rate that such an architecture will have (Interval Length).

According to the present invention, in the pre-processing stage, the data is standardized prior to processing in order to maintain a uniform scale throughout. Regex-based parsers are used to transform text input to float numbers and remove commas and units. If the LSTM network's training is not normalized, the gradients ends up disappearing. Using Min-Max normalization, the data is standardized.

In sequence formation stage, a model cannot be immediately trained on telematics data of the vehicle 100 since it is a temporal series of data. The estimation model 116 is trained on a succession of such instantaneous characteristics instead, which allows it to understand the sequential correlations between data. A 64 width (just for example) moving window with a 50% overlap is utilized to create the sequence. The feature length and model performance are affected by the window's width. Experimental study determines the width of 64. Normalizing feature sequences helps to prevent the vanishing gradient issue while training an LSTM network.

In feature selection stage, the top four features have been chosen as inputs to the proposed estimation model 116 engine load, wheel speed, acceleration pedal position, and engine speed were chosen as features.

Normal machine learning approaches like Support Vector Machine (SVM) and traditional feed forward Neural Networks cannot be employed for prediction since the produced input samples are in the form of sequences. LSTM features connections for feedback, which allows it to digest sequential inputs very effectively, unlike typical neural networks, which are feed forward in nature. To predict the instantaneous vehicle fuel consumption and emissions (i.e., CV) of an input, a three layer LSTM network is disclosed. Performance and complexity of the estimation model 116 are strongly related to its hyper parameters like the number of layers and nodes in the model. Using training and validation data, a random search is conducted to discover the best setting. A three layer model with 120, 240 and 500 neurons in each layer has been determined to be the most efficient LSTM network design. In order to ensure that all nodes are trained equally, dropout layers with a dropout rate of 0.2 have been placed between the LSTM layers an RMS Prop Optimizer with MSE loss (mean squared error) is used to train the model over 100 iterations. With its insensitivity to latencies and gaps of undetermined length, LSTM networks are ideal for time series categorization. In general, LSTM networks are favored over RNNs since RNN is susceptible to the vanishing gradient issue. In comparison to RNN and Markov models, an LSTM has a better ability to deal with gaps in data.

The suggested model obtains Mean of Absolute Errors (MAE) values of 1.52 and Root Mean Square Errors (RMSE) values of 1.49 while also having an R2 value of 0.97. The estimation model 116 was tested on a brand new vehicle, which it had never encountered, and the results obtained are really encouraging. The estimation model 116 is usable for a wide variety of automobiles, regardless of the brand or model of those automobiles. Because a single model can anticipate vehicle fuel consumption and emissions using just generic parameters of the vehicle, such as engine capacity, size, etc., an application of this on a broad scale is thus feasible. The usefulness of the developed methods was verified by performance comparisons with state of the art methods (i e DT,RF and SVR etc.).

According to an embodiment of the present invention, the estimation model 116 utilizes below mathematical formula for calculating CO2 emission in Kg.

〖Vehicle_Emission〗_(〖CO〗_2 ) (kg)=(odometer (km))/(Fuel Consumption(km/l) )*Fuel_density(kg/l)*Carbon_content(%) *(Atomic_weight_of_〖CO〗_2)/(Atomic_weight_of_Carbon)
So for Diesel fuel:
Fuel_density(kg/l) =0.835, Carbon_content(%) =0.862, 〖Atomic_weight_of_CO〗_2=44, Atomic_weight_of_Carbon=12
Similarly, the same is applicable for gasoline and other types of fuel by using respective information.

The complexity of the LSTM network is impacted by the number of layers that it has. Adding more layers might have a negative impact on the performance of the estimation model 116 since the model only functions at its best when it is at the optimum depth. In the present invention, several LSTM designs, including Single Layer LSTM, Double Layer LSTM, Triple Layer LSTM, and 4 Layer LSTM, were analyzed and contrasted with one another. The comparison is going to be based on how well the models did when they applied to the validation set. On the exam set is where all the training is done Based on the results of several experiments, the two layer design provides the greatest results, as determined by the random search. In addition to that, the quality of the estimation model 116 performance is assessed using noisy data and is found to be superior to the performance of the traditional machine learning algorithms.

Fig. 3 illustrates a method for estimating carbon dioxide (CO2) emission from vehicles, according to the present invention. The method comprises plurality of steps, of which a step 302 comprises receiving, by the computing device 100, input signals comprising operational parameters 102 from an internal network 104 of the vehicle 100, such as from On-Board Diagnostic (OBD) port of the vehicle 100 or all the sensors directly interfaced with a computing device 120. The method is characterized by a step 304 which comprises processing operational parameters 102 using the estimation model 116. The estimation model 116 is based on the Long Short-Term Memory (LSTM) network. A step 306 comprises estimating the CO2 emission from the vehicle 100 from the estimation model 116. A step 308 further comprises comparing the estimated CO2 with the threshold CO2 level 118. A step 310 comprises triggering the action 114 if the estimated CO2 exceeds the threshold CO2 level 118. In the method, an instantaneous fuel usage is estimated by the estimation model 116 before estimating the CO2 emission.

According to the present invention, the operational parameters 102 is selected from engine load, wheel speed, accelerator pedal position and engine speed. Further, each of the operational parameters 102 are measured using respective sensors in the vehicle 100 or derived using the measured parameters. In addition, the method is performed by the computing device 120. The computing device 120 is at least one of the internal device and the external device. The internal device is the Electronic Control Unit (ECU) 110 in the vehicle 100, and the external device comprises the cloud 108 and the portable/communication device 112 in communication with the ECU 110 of the vehicle 100.

According to an embodiment of the present invention, the only variables that are used as inputs to the estimation model 116 (or a soft sensor) to calculate fuel consumption (in kilometers per liter) and CO2 emission (in kg) are engine load, wheel speed, the location of the accelerator pedal, and engine speed. The computing device 120 provides presents a novel approach to the monitoring of fuel consumption (km/l) that is data based and uses real time information gathered from customers' vehicles equipped with telematics systems. The approach that was intended to be used also offered an indirect way (by applying the equation) for measuring CO2 emissions from vehicles 100. Because there is no need for a separate carbon dioxide sensor to be installed on vehicles 100, this solution is both cost effective and scalable.

According to the present invention, a deep learning based smart sensor to predicts fuel use and carbon emissions using OEM telematics data is disclosed. The computing device 120 and method discloses an innovative soft sensor for monitoring fuel use in vehicles 100 that is based on LSTM network. The present invention discloses the estimation model 116 that is based on immediate data, in contrast to most of the other contributions in the field, which rely on average consumption models. In addition, this demonstrates that the forecast made by the model does not rely on the model or manufacturer of the vehicle 100.

According to the present invention, the LSTM network based CO2 emission prediction using real time vehicle telematics sensors data is provided. In the present invention, the prediction/estimation model 116 is developed which uses vehicle data such as through OBD-II port data. Since, the OBD is a standardized protocol, the data is easily accessible for the vehicle 100 and in the same format everywhere. The computing device 120 exploits the OBD time series data judiciously to obtain causal predictions/estimation for CO2 emissions in real time. For this the computing device 120 is configured in manner to use LSTM Network time series is involved.

In an embodiment, an end-to end telematics solution to enable our customer to track and monitor respective machines, connect with authenticated dealer network, and reduce machine down time there by increasing their Return on Investment (ROI). As part of this solution, a telematics device (such as TCU) is fitted into the machines (or vehicles 100) which sends data continuously tracks machine location, also collects all this data in real time from machines and sends to the cloud 108. The estimation model 116 available in the cloud 108 analyses all these data at almost real time and provides useful and interactive insights to users through web and mobile applications. The computing device 120 and method provides Artificial Intelligence /Machine Learning (AI/ML) based data analytics approach for CO2 emission prediction using sensors data of the vehicle 100.

It should be understood that the embodiments explained in the description above are only illustrative and do not limit the scope of this invention. Many such embodiments and other modifications and changes in the embodiment explained in the description are envisaged. The scope of the invention is only limited by the scope of the claims.
, Claims:We claim:
1. A computing device (120) to estimate carbon dioxide (CO2) emission from vehicles (100), said computing device (120) configured to:
receive input signals (102, 104) comprising operational parameters (102) from an internal network (104) of said vehicle (100), characterized in that,
process said operational parameters (102) using an estimation model (116), said estimation model (116) is based on a Long Short-Term Memory (LSTM) network, and
estimate said CO2 emission from said vehicle (100) as an output from said estimation model (116).

2. The computing device (120) as claimed in claim 1 further configured to,
compare said estimated CO2 with a threshold CO2 level (118) stored in a memory element (106), and
trigger an action (114) if said estimated CO2 exceeds said threshold CO2 level (118).

3. The computing device (120) as claimed in claim 1, wherein said operational parameters (102) is selected from engine load, wheel speed, accelerator pedal position and engine speed, and wherein each of said operational parameters (102) are measured using respective sensors in said vehicle (100) or derived using measured parameters.

4. The computing device (120) as claimed in claim 1, wherein an instantaneous fuel usage is estimated by said estimation model (116) before the estimation of said CO2 emission.

5. The computing device (120) as claimed in claim 1 is at least one of an internal device and an external device, wherein said internal device is an Electronic Control Unit (ECU) (110) in said vehicle (100), and said external device comprise a cloud (108) and a communication device (112) in communication with said ECU (110) of said vehicle (100).

6. A method for estimating carbon dioxide (CO2) emission from vehicles (100), said method comprising the steps of:
receiving input signals comprising operational parameters (102) from an internal network (104) of said vehicle (100), characterized by,
processing operational parameters (102) using an estimation model (116), said estimation model (116) is based on a Long Short-Term Memory (LSTM) network, and
estimating said CO2 emission from said vehicle (100) as an output from said estimation model (116).

7. The method as claimed in claim 6 further comprises
comparing said estimated CO2 with a threshold CO2 level (118), and
triggering an action (114) if said estimated CO2 exceeds said threshold CO2 level (118).

8. The method as claimed in claim 6, wherein said operational parameters (102) is selected from engine load, wheel speed, accelerator pedal position and engine speed, and wherein each of said operational parameters (102) are measured using respective sensors in said vehicle (100) or derived using said measured parameters.

9. The method as claimed in claim 6, wherein an instantaneous fuel usage is estimated by said estimation model (116) before estimating said CO2 emission.

10. The method as claimed in claim 6 is performed by a computing device (120), wherein said computing device (120) is at least one of an internal device and an external device, wherein said internal device is an Electronic Control Unit (ECU) (110) in said vehicle (100), and said external device comprise a cloud (108) and a communication device (112) in communication with said ECU (110) of said vehicle (100).