Academia.eduAcademia.edu

Figure 11 – uploaded by Sasanka Katreddi

See full PDFdownloadDownload figure
Table 5.4: Cross-validation results for Super-Learner Model slope, the better the model performs. SVM and k-NN models’ regression lines substantially deviate from the true performance lines making the models poor options. Each of the RF and XG boost

Table 5 4: Cross-validation results for Super-Learner Model slope, the better the model performs. SVM and k-NN models’ regression lines substantially deviate from the true performance lines making the models poor options. Each of the RF and XG boost

Related Figures (71)

Figure 1.1: Sources of United States Greenhouse Gas Emissions in 2020 [1 vas emissions result in global warming adversely impacting human health, the environment, and
Figure 1.1: Sources of United States Greenhouse Gas Emissions in 2020 [1 vas emissions result in global warming adversely impacting human health, the environment, and
compliant with standards to reduce fuel consumption and minimize exhaust pollution. Not only the higher emissions but the higher energy consumption also increases the total cost
compliant with standards to reduce fuel consumption and minimize exhaust pollution. Not only the higher emissions but the higher energy consumption also increases the total cost
Figure 2.1: Machine Learning and Deep Learning as part of Artificial Intelligence [3] interest in AI and the latest advances in computer technologies led to the rapid growth of Machine
Figure 2.1: Machine Learning and Deep Learning as part of Artificial Intelligence [3] interest in AI and the latest advances in computer technologies led to the rapid growth of Machine
haracteristics and complexity of the data [5]. The typical workflow for machine learning is showr The subsets of machine learning such as supervised learning, unsupervised learning, semi-
haracteristics and complexity of the data [5]. The typical workflow for machine learning is showr The subsets of machine learning such as supervised learning, unsupervised learning, semi-
In this proposal, several ML techniques are applied to one-dimensional tabular data collected even though it has seen tremendous success on images, videos, and text datasets [24]. With the for fast diagnosis than doctors. It is being possible to predict the chances of diseases ahead [20].
In this proposal, several ML techniques are applied to one-dimensional tabular data collected even though it has seen tremendous success on images, videos, and text datasets [24]. With the for fast diagnosis than doctors. It is being possible to predict the chances of diseases ahead [20].
Table 4.1: Previous studies related to Fuel Consumption
Table 4.1: Previous studies related to Fuel Consumption
Figure 4.1: Workflow for fuel consumption modeling gas temperature, CO2, NOx, GPS altitude, GPS longitude, GPS latitude, etc. were collected for
Figure 4.1: Workflow for fuel consumption modeling gas temperature, CO2, NOx, GPS altitude, GPS longitude, GPS latitude, etc. were collected for
Table 4.2: A subset of features selected for fuel consumption modeling accuracy. Since the data is collected from multiple sensors, noise or unwanted values can cause a
Table 4.2: A subset of features selected for fuel consumption modeling accuracy. Since the data is collected from multiple sensors, noise or unwanted values can cause a
Figure 4.4: Feature Importance using Recursive Feature Elimination
Figure 4.4: Feature Importance using Recursive Feature Elimination
Figure 4.5: Feature Importance using Ridge Regression
Figure 4.5: Feature Importance using Ridge Regression
respective layers, and an output layer with a single neuron as shown in Figure 4.6. The available dataset of vehicle 1 is divided into train and test sets with 70% to train the network and 30% to test the generalization of the network. The trained model weights are then
respective layers, and an output layer with a single neuron as shown in Figure 4.6. The available dataset of vehicle 1 is divided into train and test sets with 70% to train the network and 30% to test the generalization of the network. The trained model weights are then
Figure 4.7: Neural Network Performance plots. Mean Absolute Error (Left Panel) and Mean Square Error (Right Panel) plots during ANN training for training and validation data.
Figure 4.7: Neural Network Performance plots. Mean Absolute Error (Left Panel) and Mean Square Error (Right Panel) plots during ANN training for training and validation data.
Figure 4.8: Scatter plot for the predicted data and the actual measured data
Figure 4.8: Scatter plot for the predicted data and the actual measured data
Table 4.3: Comparison of performance on the test dataset accurately predict fuel consumption with few errors. The points far away from the regression line
Table 4.3: Comparison of performance on the test dataset accurately predict fuel consumption with few errors. The points far away from the regression line
indicate that the neural network model is generalized well for unseen data. Figure 4.9: Comparison of predicted total fuel consumed per trip using different machine learning algorithms 0.0021L for the training data. The R? value of 0.7806 on the test data and 0.7762 on the train data
indicate that the neural network model is generalized well for unseen data. Figure 4.9: Comparison of predicted total fuel consumed per trip using different machine learning algorithms 0.0021L for the training data. The R? value of 0.7806 on the test data and 0.7762 on the train data
Table 4.4: Comparison of performance on the test dataset Table 4.5: Performance on the test dataset using the fine-tuned model
Table 4.4: Comparison of performance on the test dataset Table 4.5: Performance on the test dataset using the fine-tuned model
The cumulative fuel consumption with distance is plotted against the actual data to determine
The cumulative fuel consumption with distance is plotted against the actual data to determine
Figure 4.10: Comparison of actual measured values and predicted values for Cummins X15 450 test vehicle (Left Panel). Total fuel consumed vs Distance (Right Panel)
Figure 4.10: Comparison of actual measured values and predicted values for Cummins X15 450 test vehicle (Left Panel). Total fuel consumed vs Distance (Right Panel)
Figure 4.11: Comparison of actual measured values and predicted values for Navistar A26 A400MT test vehicle (Left Panel). Total fuel consumed vs Distance (Right Panel)
Figure 4.11: Comparison of actual measured values and predicted values for Navistar A26 A400MT test vehicle (Left Panel). Total fuel consumed vs Distance (Right Panel)
Figure 4.12: Comparison of actual measured values and predicted values for Paccar MX-13 test vehicle (Left Panel). Total fuel consumed vs Distance (Right Panel)
Figure 4.12: Comparison of actual measured values and predicted values for Paccar MX-13 test vehicle (Left Panel). Total fuel consumed vs Distance (Right Panel)
Figure 4.13: Neural Network Performance Plots for fine-tuning. Mean Absolute Error (Left Panel) and Mean Square Error (Right Panel) on training and validation data
Figure 4.13: Neural Network Performance Plots for fine-tuning. Mean Absolute Error (Left Panel) and Mean Square Error (Right Panel) on training and validation data
Figure 4.14: Comparison of actual measured values and predicted values for Cummins ZX15 450 test vehicle (Left Panel). Total fuel consumed vs Distance (Right Panel)
Figure 4.14: Comparison of actual measured values and predicted values for Cummins ZX15 450 test vehicle (Left Panel). Total fuel consumed vs Distance (Right Panel)
certain vocations and regions. Due to the ongoing data collection process and lack of availability
certain vocations and regions. Due to the ongoing data collection process and lack of availability
Table 5.1: Features in the raw data collected
Table 5.1: Features in the raw data collected
input features in the dataset from 9 to 13 with cost-per-mile as the target variable. However, the effect of these derived features on the maintenance cost is unknown.
input features in the dataset from 9 to 13 with cost-per-mile as the target variable. However, the effect of these derived features on the maintenance cost is unknown.
tive fuels is very low in the heavy-duty vehicle sector [157] resulting in a lack of data. To prevent
tive fuels is very low in the heavy-duty vehicle sector [157] resulting in a lack of data. To prevent
Figure 5.5: Random Forest Model developed for Maintenance Cost Data its major drawback is that the number of neighbors and the performance is highly dependent on
Figure 5.5: Random Forest Model developed for Maintenance Cost Data its major drawback is that the number of neighbors and the performance is highly dependent on
Figure 5.6: XGBoost Model developed for Maintenance Cost Data the loss using gradients. XGB performs well with larger datasets, and its training is fast due to
Figure 5.6: XGBoost Model developed for Maintenance Cost Data the loss using gradients. XGB performs well with larger datasets, and its training is fast due to
Figure 5.7: Neural Network Model developed for Maintenance Cost Data. Super Learner Model
Figure 5.7: Neural Network Model developed for Maintenance Cost Data. Super Learner Model
Figure 5.8: Super Learner Model [9]
Figure 5.8: Super Learner Model [9]
Figure 5.9: Distribution of cost per mile ($/mile) vs fuel type for train data
Figure 5.9: Distribution of cost per mile ($/mile) vs fuel type for train data
Table 5.2: Comparison of cross-validation scores for all ML models using original features included in the analysis. Diesel vehicles include an after-treatment system that consists of multiple
Table 5.2: Comparison of cross-validation scores for all ML models using original features included in the analysis. Diesel vehicles include an after-treatment system that consists of multiple
Table 5.3: Comparison of cross-validation scores for all ML models using combined features
Table 5.3: Comparison of cross-validation scores for all ML models using combined features
Figure 5.10: Comparison of Cross Validation Performance for ML models. Mean Absolute Error (Left Panel), Root Mean Square Error (Center Panel), Coefficient of Determination (Right Panel) using machine learning models.
Figure 5.10: Comparison of Cross Validation Performance for ML models. Mean Absolute Error (Left Panel), Root Mean Square Error (Center Panel), Coefficient of Determination (Right Panel) using machine learning models.
Figure 5.11: Distribution of cost per mile ($/mile) vs fuel type for test data combined feature input. In this case, the error for most of the models is lower than in the previou ase and the coefficient of determination has increased.
Figure 5.11: Distribution of cost per mile ($/mile) vs fuel type for test data combined feature input. In this case, the error for most of the models is lower than in the previou ase and the coefficient of determination has increased.
Figure 5.12: Comparison of ML model performance on test data. Mean Absolute Error (Left Panel), Root Mean Square Error (Center Panel), and Coefficient of Determination (Right Panel) using machine learning models. networks have gained success on data containing images, videos, text, audio, etc. their performance
Figure 5.12: Comparison of ML model performance on test data. Mean Absolute Error (Left Panel), Root Mean Square Error (Center Panel), and Coefficient of Determination (Right Panel) using machine learning models. networks have gained success on data containing images, videos, text, audio, etc. their performance
Figure 5.14: Scatter plots for actual and predicted cost per mile based on fuel type Diesel for test data using different machine learning models
Figure 5.14: Scatter plots for actual and predicted cost per mile based on fuel type Diesel for test data using different machine learning models
Figure 5.16: Cross Validation Performance using Super-Learner Model. Mean Absolute Error (Left Panel), Root Mean Square Error (Center Panel), Coefficient of Determination (Right Panel)
Figure 5.16: Cross Validation Performance using Super-Learner Model. Mean Absolute Error (Left Panel), Root Mean Square Error (Center Panel), Coefficient of Determination (Right Panel)
Figure 5.17: Model Evaluation on Test Data using Super-Learner Model. Mean Absolute Error (Left Panel), Root M ean Square | Error (Center Panel), Coefficient of Determination (Right Panel)
Figure 5.17: Model Evaluation on Test Data using Super-Learner Model. Mean Absolute Error (Left Panel), Root M ean Square | Error (Center Panel), Coefficient of Determination (Right Panel)
Figure 5.18: Scatter plot for actual and predicted cost per mile based on fuel type Natural Gas (Left Panel) and Diesel (Right Panel) for test data using super-learner ensemble model
Figure 5.18: Scatter plot for actual and predicted cost per mile based on fuel type Natural Gas (Left Panel) and Diesel (Right Panel) for test data using super-learner ensemble model
‘igure 5.19: Average Cost per mile over Years for Natural Gas (Left Panel) and Diesel (Right -anel) Test Vehicles using Super-Learner Ensemble Model Comparison of Natural Gas and Diesel Average Cost Per Mile
‘igure 5.19: Average Cost per mile over Years for Natural Gas (Left Panel) and Diesel (Right -anel) Test Vehicles using Super-Learner Ensemble Model Comparison of Natural Gas and Diesel Average Cost Per Mile
Figure 5.20: Comparison of Predicted Average Cost Per Mile over Vehicle Age for Natural Gas and Diesel Delivery Trucks using the Super-Learner Model
Figure 5.20: Comparison of Predicted Average Cost Per Mile over Vehicle Age for Natural Gas and Diesel Delivery Trucks using the Super-Learner Model
Table 6.1: Features and Description
Table 6.1: Features and Description
Figure 6.1: The distribution of data based on activity and fuel type plots for IQR, a careful inspection of data points outside the whiskers is performed to determine
Figure 6.1: The distribution of data based on activity and fuel type plots for IQR, a careful inspection of data points outside the whiskers is performed to determine
The convergence of the MERF algorithm is monitored using generalized log-likelihood (GLL)
The convergence of the MERF algorithm is monitored using generalized log-likelihood (GLL)
Table 6.2: The number of vehicles available per cluster The MERF assumes the random effects term to be correct for estimating the forest function
Table 6.2: The number of vehicles available per cluster The MERF assumes the random effects term to be correct for estimating the forest function
Figure 6.4: Generalized Log-Likelihood (Left Panel) and Mean Square Error (Right Panel)
Figure 6.4: Generalized Log-Likelihood (Left Panel) and Mean Square Error (Right Panel)
Figure 6.5: Distribution of values of b; after training From the plot, the model converges by the 50th iteration.
Figure 6.5: Distribution of values of b; after training From the plot, the model converges by the 50th iteration.
Table 6.3: Evaluation Metrics on Test Data Clusters achieved a coefficient of determination (R?) of 98.96% and a mean square error (MSE) of 0.008%
Table 6.3: Evaluation Metrics on Test Data Clusters achieved a coefficient of determination (R?) of 98.96% and a mean square error (MSE) of 0.008%
Table 6.4: School Bus Test Data Summary 6.3.1 School Bus The summary of test data for a diesel and natural gas school bus operated in California and
Table 6.4: School Bus Test Data Summary 6.3.1 School Bus The summary of test data for a diesel and natural gas school bus operated in California and
Table 6.5: Delivery Test Data Summary niles with one major engine & transmission at a mileage of around 68,000. Engine & transmissior
Table 6.5: Delivery Test Data Summary niles with one major engine & transmission at a mileage of around 68,000. Engine & transmissior
Figure 6.6: Average predicted cost per mile in school bus
Figure 6.6: Average predicted cost per mile in school bus
Figure 6.7: Average predicted cost per mile in delivery truck 360k miles incurring 10 times higher cost than the maximum of other corrective maintenance costs than doubled every 100,000 miles. The periodic maintenance for the chassis had constant costs
Figure 6.7: Average predicted cost per mile in delivery truck 360k miles incurring 10 times higher cost than the maximum of other corrective maintenance costs than doubled every 100,000 miles. The periodic maintenance for the chassis had constant costs
Table 6.6: Vocational Test Data Summary Comparison of Average Cost Per Mile for a Diesel, a Natural Gas and a Propane Vocational Truck
Table 6.6: Vocational Test Data Summary Comparison of Average Cost Per Mile for a Diesel, a Natural Gas and a Propane Vocational Truck
Figure 6.8: Average predicted cost per mile in vocational truck 6.3.4 Refuse Truck The summary of test data for a diesel refuse truck operated in Ohio and a natural gas refuse
Figure 6.8: Average predicted cost per mile in vocational truck 6.3.4 Refuse Truck The summary of test data for a diesel refuse truck operated in Ohio and a natural gas refuse
Table 6.7: Refuse Test Data Summary
Table 6.7: Refuse Test Data Summary
Figure 6.9: Average predicted cost per mile in refuse truck
Figure 6.9: Average predicted cost per mile in refuse truck
Table 6.8: Goods Movement Truck Test Data Summary & transmission, and fuel system. The number of corrective maintenance on this vehicle is almost
Table 6.8: Goods Movement Truck Test Data Summary & transmission, and fuel system. The number of corrective maintenance on this vehicle is almost
Figure 6.10: Average predicted cost per mile in goods movement truck equal to the number of periodic and preventative maintenance together.
Figure 6.10: Average predicted cost per mile in goods movement truck equal to the number of periodic and preventative maintenance together.
Table 6.9: Transit Bus Test Data Summary 6.3.6 Transit Bus Since there is no data for transit buses using diesel and propane fuels, the transit bus data is
Table 6.9: Transit Bus Test Data Summary 6.3.6 Transit Bus Since there is no data for transit buses using diesel and propane fuels, the transit bus data is
Figure 6.11: Average predicted cost per mile in transit bus and vocational trucks which have stop-and-go activity operated at lower speeds have lower average
Figure 6.11: Average predicted cost per mile in transit bus and vocational trucks which have stop-and-go activity operated at lower speeds have lower average
Related topics:
Artificial IntelligenceMachine LearningApplications of Artificial Intelligence in ManufacturingAutomotive EngineeringCompressed Natural GasFuel efficiencyAlternate FuelsTotal Cost of OwnershipNOx EmissionHeavy Duty Vehicles
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved