The eco-toll estimation problem quantifies the expected environmental cost (e.g., energy consumption, exhaust emissions) for a vehicle to travel along a path. This problem is important for societal applications such as eco-routing, which aims to find paths with the lowest exhaust emissions or energy need. The challenges of this problem are threefold: (1) the dependence of a vehicle's eco-toll on its physical parameters; (2) the lack of access to data with eco-toll information; and (3) the influence of contextual information (i.e. the connections of adjacent segments in the path) on the eco-toll of road segments. Prior work on eco-toll estimation has mostly relied on pure data-driven approaches and has high estimation errors given the limited training data. To address these limitations, we propose a novel Eco-toll estimation Physics-informed Neural Network framework (Eco-PiNN) using three novel ideas, namely, (1) a physics-informed decoder that integrates the physical laws governing vehicle dynamics into the network, (2) an attention-based contextual information encoder, and (3) a physics-informed regularization to reduce overfitting. Experiments on real-world heavy-duty truck data show that the proposed method can greatly improve the accuracy of eco-toll estimation compared with state-of-the-art methods. *The full version of the paper can be accessed at https://arxiv.org/abs/2301.05739

The goal of spatially explainable artificial intelligence (AI) classification approach is to build a classifier to distinguish two classes (e.g., responder, non-responder) based on the their spatial arrangements (e.g., spatial interactions between different point categories) given multi-category point data from two classes. This problem is important for generating hypotheses towards discovering new immunotherapies for cancer treatment as well as for other applications in biomedical research and microbial ecology. This problem is challenging due to an exponential number of category subsets which may vary in the strength of their spatial interactions. Most prior efforts on using human selected spatial association measures may not be sufficient for capturing the relevant spatial interactions (e.g., surrounded by) which may be of biological significance. In addition, the related deep neural networks are limited to category pairs and do not explore larger subsets of point categories. To overcome these limitations, we propose a Spatial-interaction Aware Multi-Category deep neural Network (SAMCNet) architecture and contribute novel local reference frame characterization and point pair prioritization layers for spatially explainable classification. Experimental results on multiple cancer datasets (e.g., MxIF) show that the proposed architecture provides higher prediction accuracy over baseline methods. A real-world case study demonstrates that the proposed work discovers patterns that are missed by the existing methods and has the potential to inspire new scientific discoveries.

Given a collection of N geo-located point samples of k types, we aim to detect spatial mixture patterns of interest, which are sub-regions of the study area that have significantly high or low mixture of points of different types. Spatial mixture patterns have important applications in many societal domains, including resilience of smart cities and communities, biodiversity, equity, business intelligence, etc. The problem is challenging because ranking and selection of candidate patterns can be highly susceptible to the effect of natural randomness, and real-world data often consists of various mixture patterns. In related work, the multi-nomial scan statistic does not support identification of high or low mixture due to its "directionless" nature and high sensitivity to the composition of mixture patterns in data. While species richness indices in biodiversity research allow specification of directions, the measures are very sensitive to spatial randomness effects. To bridge the gap, we first propose a spatial mixture index to provide robust ranking among candidate patterns. Then, we present a dual-level Monte-Carlo estimation method with a baseline algorithm for spatial mixture pattern detection. Finally, we propose both an exact algorithm and a distribution-inspired sequence-reduction heuristic to accelerate the baseline approach. Experiment results with both synthetic and real-world data show that the proposed approaches can detect mixture patterns with high accuracy, and the acceleration methods can greatly reduce computational cost while maintaining high solution quality.

In recent years, deep reinforcement learning (DRL) algorithms have been widely studied and utilized in the area of Intelligent Transportation Systems (ITS). DRL agents are mostly trained with transition pairs and interaction trajectories generated from simulation, and they can achieve satisfying or near optimal performances under familiar input states. However, for relative rare visited or even unvisited regions in the state space, there is no guarantee that the agent could perform well. Unfortunately, novel conditions are inevitable in real-world problems and there is always a gap between the real data and simulated data. Therefore, to implement DRL algorithms in real-world transportation systems, we should not only train the agent learn a policy that maps states to actions, but also the model uncertainty associated with each action. In this study, we adapt the method of Bayesian ensemble to train a group of agents with imposed diversity for an energy management system of a delivery vehicle. The agents in the ensemble agree well on familiar states but show diverse results on unfamiliar or novel states. This uncertainty estimation facilitates the implementation of interpretable postprocessing modules which can ensure robust and safe operations under high uncertainty conditions.

Rule-based (RB) energy management strategies are widely used in hybrid-electric vehicles because they are easy to implement and can be used without prior knowledge about future trips. In the literature, parameters used in RB methods are tuned and designed using known driving cycles. Although promising results have been demonstrated, it is difficult to apply such cycle-specific methods on real trips of last-mile delivery vehicles that have significant trip-to-trip differences in distance and energy intensity. In this paper, a reinforcement learning method and a RB strategy is used to improve the fuel economy of an in-use extended range electric vehicle (EREV) used in a last-mile package delivery application. An intelligent agent is trained on historical trips of a single delivery vehicle to tune a parameter in the engine-generator control logic during the trip using real-time information. The method is demonstrated on actual historical delivery trips in a simulation environment. An average of 19.5% in fuel efficiency improvement in miles per gallon gasoline equivalent is achieved on 44 test trips with a distance range of 31 miles to 54 miles not used for training, demonstrating promise to generalize the method. The presented framework is extendable to other RB methods and EREV applications like transit buses and commuter vehicles where similar trips are frequently repeated day-to-day.

The trajectory-aware lowest-cost path selection problem aims to find the lowest-cost path using trajectory data. Trajectory data is valuable since it carries information about travel cost along paths, and also reflects travelers' routing preference. Path-centric travel cost estimation models using trajectory data grows popular recently, which considers the auto-correlation of the energy consumption on different segments of a path. However, path-centric models are more computationally expensive than edge-centric models. The main challenge of this problem is that the travel cost of every candidate path explored during the process of searching for the lowest-cost path need to be estimated, resulting in high computational cost. The current path selection algorithms that use path-centric cost estimation models still follow the pattern of "path + edge" when exploring candidate paths, which may result in redundant computation. We introduce a trajectory-aware graph model in which each node is a maximal trajectory-aware path. Two nodes in the trajectory-aware graph are linked by an edge if their union forms a trajectory-union path. We then propose a path selection algorithm to find a path in the proposed trajectory-aware graph which corresponds to the lowest-cost path in the input spatial network. We prove theoretically the proposed algorithm is correct and complete. Moreover, we prove theoretically that the proposed path selection algorithm cost much less computational time than the algorithm used in the related work, and validate it through experiments using real-world trajectory data.

Class ambiguity refers to the phenomenon whereby similar features correspond to different classes at different locations. Given heterogeneous geographic data with class ambiguity, the spatial ensemble learning (SEL) problem aims to find a decomposition of the geographic area into disjoint zones such that class ambiguity is minimized and a local classifier can be learned in each zone. The problem is important for applications such as land cover mapping from heterogeneous earth observation data with spectral confusion. However, the problem is challenging due to its high computational cost. Related work in ensemble learning either assumes an identical sample distribution (e.g., bagging, boosting, random forest) or decomposes multi-modular input data in the feature vector space (e.g., mixture of experts, multimodal ensemble) and thus cannot effectively minimize class ambiguity. In contrast, we propose a spatial ensemble framework that explicitly partitions input data in geographic space. Our approach first preprocesses data into homogeneous spatial patches and uses a greedy heuristic to allocate pairs of patches with high class ambiguity into different zones. We further extend our spatial ensemble learning framework with spatial dependency between nearby zones based on the spatial autocorrelation effect. Both theoretical analysis and experimental evaluations on two real world wetland mapping datasets show the feasibility of the proposed approach.

Given a spatial road network, an origin, a destination, and trajectory data of vehicles on the network, the Energy-efficient Path Selection (EPS) problem aims to find the most energy-efficient path (i.e., with least energy consumption) between the origin and the destination. With world energy consumption growing rapidly, estimating and reducing the energy consumption of road transportation is becoming critical. The main challenge of this problem is to adopt energy consumption as the cost metric of paths, which is neglected by the related work in shortest path selection problem whose typical metrics are distance and time. Additionally, negative energy consumption caused by the use of regenerative braking on electrified vehicles prevents classical algorithms like Dijkstra's algorithm from functioning correctly. We introduce a Physics-guided Energy Consumption (PEC) model based on a low-order physics model, which estimates energy consumption as a function of the vehicle parameters (e.g., mass and powertrain system efficiency) and use the estimation in the proposed adaptive dynamic programming algorithm for path selection. Our PEC model treats energy consumption as a unique metric that is determined not only by the path and vehicle's motion along the path, but also on properties of the vehicle itself. Experiments show that the PEC model estimates are more similar to real trajectory data than the estimates represented by the mean or histogram of historical data. Also, the path found by the proposed method is more energy-efficient than both the currently used path and the fastest path found by a commercial routing package. As far as we know, this is the first paper to use a physics-guided method to estimate the vehicle energy consumption and perform path selection.

Class ambiguity refers to the phenomenon whereby samples with similar features belong to different classes at different locations. Given heterogeneous geographic data with class ambiguity, the spatial ensemble learning (SEL) problem aims to find a decomposition of the geographic area into disjoint zones such that class ambiguity is minimized and a local classifier can be learned in each zone. SEL problem is important for applications such as land cover mapping from heterogeneous earth observation data with spectral confusion. However, the problem is challenging due to its high computational cost (finding an optimal zone partition is NP-hard). Related work in ensemble learning either assumes an identical sample distribution (e.g., bagging, boosting, random forest) or decomposes multi-modular input data in the feature vector space (e.g., mixture of experts, multimodal ensemble), and thus cannot effectively minimize class ambiguity. In contrast, our spatial ensemble framework explicitly partitions input data in geographic space. Our approach first preprocesses data into homogeneous spatial patches and uses a greedy heuristic to allocate pairs of patches with high class ambiguity into different zones. Both theoretical analysis and experimental evaluations on two real world wetland mapping datasets show the feasibility of the proposed approach.