arXiv subjects

Yifei Guo

Publications and source records attributed to Yifei Guo.

At least 19 recordsLinked to original sources

arXiv2026

FloVerse: Floor Plan-Guided Multi-Modal Navigation

Floor plans encapsulate compact spatial priors, enabling agents to navigate unseen scenes more efficiently. While prior work has explored floor plan-guided navigation, it has focused mainly on PointNav and a limited set of environments. To bridge this gap, we introduce FloVerse, a new task for floor plan-guided embodied navigation that unifies PointNav, ObjectNav, and ImageNav. To support FloVerse, we assemble FloVerse-1.6K, a large-scale dataset of 1.6K scenes from HM3D and Gibson 4+, paired with corresponding floor plans, comprising 240K expert trajectories and 12M RGBD frames. We further propose ThreeDiff, a two-stage imitation learning policy comprising a planner, a diffusion-based multimodal goal-reasoning module trained via masked-modality modeling, and a refiner, a depth-based trajectory-refinement module for safe execution. Extensive experiments demonstrate that (1) floor-plan priors improve navigation performance across all goal modalities, and (2) ThreeDiff implicitly captures spatial information from floor plans. These results underscore the effectiveness of spatial priors and validate our proposed unified approach for floor plan-guided embodied navigation.

cs.RO↗

arXiv2026

The Combined Problem of Online Task Assignment and Lifelong Path Finding in Logistics Warehouses: Rule-Based Systems Matter

We study the combined problem of online task assignment and lifelong path finding, which is crucial for the logistics industries. However, most literature either (1) focuses on lifelong path finding assuming a given task assigner, or (2) studies the offline version of this problem where tasks are known in advance. We argue that, to maximize the system throughput, the online version that integrates these two components should be tackled directly. To this end, we introduce a formal framework of the combined problem and its solution concept. Then, we design a rule-based lifelong planner under a practical robot model that works well even in environments with severe local congestion. Upon that, we automate the search for the task assigner with respect to the underlying path planner. Simulation experiments conducted in warehouse scenarios at Meituan, one of the largest shopping platforms in China, demonstrate that (a)in terms of time efficiency, our system requires only 83.77% of the execution time needed for the currently deployed system at Meituan, outperforming other SOTA algorithms by 8.09%; (b)in terms of economic efficiency, ours can achieve the same throughput with only 60% of the agents currently in use. The code and demos are available at https://github.com/Fernadoo/Online-TAPF.

cs.MA↗

arXiv2025

To See a World in a Spark of Neuron: Disentangling Multi-task Interference for Training-free Model Merging

Fine-tuning pre-trained models on targeted datasets enhances task-specific performance but often comes at the expense of generalization. Model merging techniques, which integrate multiple fine-tuned models into a single multi-task model through task arithmetic, offer a promising solution. However, task interference remains a fundamental challenge, leading to performance degradation and suboptimal merged models. Existing approaches largely overlooked the fundamental roles of neurons, their connectivity, and activation, resulting in a merging process and a merged model that does not consider how neurons relay and process information. In this work, we present the first study that relies on neuronal mechanisms for model merging. Specifically, we decomposed task-specific representations into two complementary neuronal subspaces that regulate input sensitivity and task adaptability. Leveraging this decomposition, we introduced NeuroMerging, a novel merging framework developed to mitigate task interference within neuronal subspaces, enabling training-free model fusion across diverse tasks. Through extensive experiments, we demonstrated that NeuroMerging achieved superior performance compared to existing methods on multi-task benchmarks across both natural language and vision domains. Our findings highlighted the importance of aligning neuronal mechanisms in model merging, offering new insights into mitigating task interference and improving knowledge fusion. Our project is available at https://ZzzitaoFang.github.io/projects/NeuroMerging/.

cs.LG↗

arXiv2025

Neural Parameter Search for Slimmer Fine-Tuned Models and Better Transfer

Foundation models and their checkpoints have significantly advanced deep learning, boosting performance across various applications. However, fine-tuned models often struggle outside their specific domains and exhibit considerable redundancy. Recent studies suggest that combining a pruned fine-tuned model with the original pre-trained model can mitigate forgetting, reduce interference when merging model parameters across tasks, and improve compression efficiency. In this context, developing an effective pruning strategy for fine-tuned models is crucial. Leveraging the advantages of the task vector mechanism, we preprocess fine-tuned models by calculating the differences between them and the original model. Recognizing that different task vector subspaces contribute variably to model performance, we introduce a novel method called Neural Parameter Search (NPS-Pruning) for slimming down fine-tuned models. This method enhances pruning efficiency by searching through neural parameters of task vectors within low-rank subspaces. Our method has three key applications: enhancing knowledge transfer through pairwise model interpolation, facilitating effective knowledge fusion via model merging, and enabling the deployment of compressed models that retain near-original performance while significantly reducing storage costs. Extensive experiments across vision, NLP, and multi-modal benchmarks demonstrate the effectiveness and robustness of our approach, resulting in substantial performance gains. The code is publicly available at: https://github.com/duguodong7/NPS-Pruning.

cs.LG↗

arXiv2024

Parameter Competition Balancing for Model Merging

While fine-tuning pretrained models has become common practice, these models often underperform outside their specific domains. Recently developed model merging techniques enable the direct integration of multiple models, each fine-tuned for distinct tasks, into a single model. This strategy promotes multitasking capabilities without requiring retraining on the original datasets. However, existing methods fall short in addressing potential conflicts and complex correlations between tasks, especially in parameter-level adjustments, posing a challenge in effectively balancing parameter competition across various tasks. This paper introduces an innovative technique named PCB-Merging (Parameter Competition Balancing), a lightweight and training-free technique that adjusts the coefficients of each parameter for effective model merging. PCB-Merging employs intra-balancing to gauge parameter significance within individual tasks and inter-balancing to assess parameter similarities across different tasks. Parameters with low importance scores are dropped, and the remaining ones are rescaled to form the final merged model. We assessed our approach in diverse merging scenarios, including cross-task, cross-domain, and cross-training configurations, as well as out-of-domain generalization. The experimental results reveal that our approach achieves substantial performance enhancements across multiple modalities, domains, model sizes, number of tasks, fine-tuning forms, and large language models, outperforming existing model merging methods. The code is publicly available at: \url{https://github.com/duguodong7/pcb-merging}.

cs.CV↗

arXiv2024

Modeling Fault Recovery and Transient Stability of Grid-Forming Converters Equipped With Current Reference Limitation

When grid-forming (GFM) inverter-based resources (IBRs) face severe grid disturbances (e.g., short-circuit faults), the current limitation mechanism may be triggered. Consequently, the GFM IBRs enter the current-saturation mode, inducing nonlinear dynamical behaviors and posing great challenges to the post-disturbance transient angle stability. This paper presents a systematic study to reveal the fault recovery behaviors of a GFM IBR and identify the risk of instability. A closed-form expression for the necessary condition that a GFM IBR returns from the current-saturation mode to the normal operation mode is presented. Based on these analyses, it is inferred that the angle of the magnitude-saturated current significantly affects the post-fault recovery and transient stability; with different angle selection, the system may follow multiple post-fault trajectories depending on those conditions: 1) Convergence to a normal stable equilibrium point (SEP), 2) convergence to a saturated stable equilibrium point (satSEP), or 3) divergence (instability). In this paper, the circumstances under which a GFM IBR cannot escape from the current-saturation mode are thoroughly investigated. The theoretical analyses are verified by dynamic simulations.

eess.SY↗

arXiv2024

Knowledge Fusion By Evolving Weights of Language Models

Fine-tuning pre-trained language models, particularly large language models, demands extensive computing resources and can result in varying performance outcomes across different domains and datasets. This paper examines the approach of integrating multiple models from diverse training scenarios into a unified model. This unified model excels across various data domains and exhibits the ability to generalize well on out-of-domain data. We propose a knowledge fusion method named Evolver, inspired by evolutionary algorithms, which does not need further training or additional training data. Specifically, our method involves aggregating the weights of different language models into a population and subsequently generating offspring models through mutation and crossover operations. These offspring models are then evaluated against their parents, allowing for the preservation of those models that show enhanced performance on development datasets. Importantly, our model evolving strategy can be seamlessly integrated with existing model merging frameworks, offering a versatile tool for model enhancement. Experimental results on mainstream language models (i.e., encoder-only, decoder-only, encoder-decoder) reveal that Evolver outperforms previous state-of-the-art models by large margins. The code is publicly available at {https://github.com/duguodong7/model-evolution}.

cs.CL↗

arXiv2024

CADE: Cosine Annealing Differential Evolution for Spiking Neural Network

Spiking neural networks (SNNs) have gained prominence for their potential in neuromorphic computing and energy-efficient artificial intelligence, yet optimizing them remains a formidable challenge for gradient-based methods due to their discrete, spike-based computation. This paper attempts to tackle the challenges by introducing Cosine Annealing Differential Evolution (CADE), designed to modulate the mutation factor (F) and crossover rate (CR) of differential evolution (DE) for the SNN model, i.e., Spiking Element Wise (SEW) ResNet. Extensive empirical evaluations were conducted to analyze CADE. CADE showed a balance in exploring and exploiting the search space, resulting in accelerated convergence and improved accuracy compared to existing gradient-based and DE-based methods. Moreover, an initialization method based on a transfer learning setting was developed, pretraining on a source dataset (i.e., CIFAR-10) and fine-tuning the target dataset (i.e., CIFAR-100), to improve population diversity. It was found to further enhance CADE for SNN. Remarkably, CADE elevates the performance of the highest accuracy SEW model by an additional 0.52 percentage points, underscoring its effectiveness in fine-tuning and enhancing SNNs. These findings emphasize the pivotal role of a scheduler for F and CR adjustment, especially for DE-based SNN. Source Code on Github: https://github.com/Tank-Jiang/CADE4SNN.

cs.NE↗

arXiv2023

A Model Predictive Approach for Enhancing Transient Stability of Grid-Forming Converters

A model predictive control (MPC) method for enhancing post-fault transient stability of a grid-forming (GFM) inverter based resources (IBRs) is developed in this paper. This proposed controller is activated as soon as the converter enters into the post-fault current-saturation mode. It aims at mitigating the instability arising from insufficient deceleration due to current saturation and thus improving the transient stability of a GFM-IBR. The MPC approach optimises the post-fault trajectory of GFM IBRs by introducing appropriate corrective phase angle jumps and active power references where the post-fault dynamics of GFM IBRs are addressed. These two signals provide controllability over GFM IBR's post-fault trajectory. This paper addresses the mitigation of oscillations between current-saturation mode and normal mode by forced saturation if conditions for remaining in the normal mode do not hold. The performance of the proposal is tested via dynamic simulations under various grid conditions and compared with other existing strategies. The results demonstrate significant improvement in transient stability.

eess.SY↗

arXiv2022

Model-Free Optimal Control of Inverter for Dynamic Voltage Support

Inverter-based resources (IBRs) are required to provide dynamic voltage support (DVS) during voltage dips to enhance the low-voltage ride-through capability. In this paper, we develop a model-free control method to achieve the optimal DVS (ODVS) without relying on the knowledge of grid parameters. Delving into the optimum trajectory of the ODVS problem, it is found that either the current constraint and the maximum active power constraint of IBRs are binding or one of the constraints is binding. This inspires us to search for the optimum in a closed-loop way by a perturb-and-observe (P&O)-based optimum seeking (OS) controller with either the power factor angle or the reactive current being the manipulated (perturbed) variable. The system is guaranteed to converge asymptotically to the optimum provided the stepsize sequence is diminishing and non-summable. The proposed model-free optimal control is finally implemented within a single-stage photovoltaic (PV) system, where dynamic simulations demonstrate the optimal and fast DVS performance

eess.SY↗

arXiv2022

Online Voltage Control for Unbalanced Distribution Networks Using Projected Newton Method

This paper proposes an online voltage control strategy of distributed energy resources (DERs), based on the projected Newton method (PNM), for unbalanced distribution networks. The optimal Volt/VAr control (VVC) problem is formulated as an optimization program, with the goal of maintaining the voltage profile across the network by coordinating the VAr outputs of DERs. To overcome the slow convergence rate of conventional gradient-based methods, a PNM-based solution algorithm is developed to solve this VVC problem. It utilizes a non-diagonal symmetric positive definite matrix, developed from the Hessian matrix of the objective, to scale the gradient, and thus a fast convergence performance can be expected in this Newton-like algorithm. Moreover, taking advantage of the instantaneous feedback of voltage measurements, the online implementation of the PNM-based VVC is further designed to deal with fast system variations. In this online PNM-based VVC scheme, each bus agent communicates the instantaneous voltage measurements to the central agent, and the central agent communicates the VAr output commands of DERs back to each bus agent. The fast convergence performance of PNM results in a stronger capability to track the system variations in real time. Finally, numerical case studies are performed to validate the effectiveness, superiority, and scalability of the proposed method.

eess.SY↗

arXiv2021

Analyzing Photovoltaic's Impact on Conservation Voltage Reduction in Distribution Networks

Conservation voltage reduction (CVR) has been widely implemented in distribution networks and helped utilities effectively reduce energy and peak load. However, the increasing penetration level of solar photovoltaic (PV) has affected voltage profiles and the performance of CVR. It remains an outstanding question how CVR and solar PV interact with each other. Understanding this interaction is important for utilities in implementing CVR and assessing its performance. This paper studies the impact of solar PV on CVR in a real distribution system in the Midwest U.S. using comprehensive simulations. We have considered various PV allocations and penetration levels, as well as different inverter control modes according to IEEE Std 1547-2018. Three metrics are used to quantify the impact of solar PV on CVR: voltages at the substation, voltage distribution across the network, and energy consumption reduction due to CVR. The results show that the allocations of solar PV have the most significant effect on the CVR performance, where a dispersed allocation of solar PV will help flatten voltage profile and achieve deeper voltage reductions at the substation, less energy consumption and line losses.

eess.SY↗

arXiv2021

On the Optimality of Voltage Unbalance Attenuation by Inverters

In this paper, we investigate the control of inverter-based resources (IBRs) for optimal voltage unbalance attenuation (OVUA). This problem is formulated as an optimization program under a tailored dq-frame, which minimizes the negative-sequence voltage at the point of common coupling (PCC) subject to the current, active power, synchronization stability, and feasibility constraints. The program is inherently nonconvex and intractable. So, to guarantee the optimality, a rigorous optimality analysis is performed by leveraging analytical optimization. The analysis is divided into two cases: full mitigation of VU and partial attenuation of VU. For the former case, we analytically solve the original program since the resultant VU is immediately deducible. For the latter one, directly solving the problem becomes very hard. Thus, we reformulate the program into an equivalent but more tractable form under certain conditions, by which the analytical optimum can be finally derived. It is found that the optimum trajectory has three stages (O1-O3), depending on two critical boundary conditions (C1 and C2). Finally, we implement the optimum with a photovoltaic (PV)-storage system by developing an OVUA controller. The proposed approach is demonstrated by dynamic simulations under different VU conditions and is compared with several existing practices.

eess.SY↗

arXiv2021

An Online Feedback-Based Linearized Power Flow Model for Unbalanced Distribution Networks

The non-linearity and non-convexity of power flow models and the phase coupling challenge the analysis and optimization of unbalanced distribution networks. To tackle the challenges, this paper proposes an online feedback-based linearized power flow model for unbalanced distribution networks with both wye-connected and delta-connected loads. The online feedback-based linearized model is grounded on the first-order Taylor expansion of the branch flow model, and updates the model parameters via online feedback by leveraging the instantaneous measured voltages and load consumption at the previous time step. Its closed-loop nature can asymptotically mitigate the model mismatch, thus lending itself to a good performance. In addition, exploiting the connection structure of unbalanced radial distribution networks, we provide a unified matrix-vector compact form of the proposed linearized power flow model. Finally, the numerical tests on the IEEE 123-bus test system validate the effectiveness and superiority of the proposed model. A simple optimal power flow case is also provided to illustrate the application of the online feedback-based linearized model.

eess.SY↗

arXiv2021

Global Optimality of Inverter Dynamic Voltage Support

This paper investigates the dynamic voltage support (DVS) control of inverter-based resources (IBRs) under voltage sags to enhance the low-voltage ride-through performance. We first revisit the prevalent droop control from an optimization perspective to elaborate on why it usually suffers from suboptimality. Then, we formulate the DVS problem as an optimization program that maximizes the positive-sequence voltage magnitude at the point of common coupling (PCC) subject to the current, active power, and stability constraints. The program is inherently nonconvex due to the active power limits, of which the global optimality is not guaranteed by off-the-shelf solvers. In this context, we perform the optimality analysis to explore the global optimum analytically. It is found that the unique global optimum has three scenarios/stages (S1--S3), which depends on the specific relationship among grid voltage, grid strength, as well as physical limits of IBRs. The closed-form solutions in S1 and S3 are derived and the optimality conditions for S2 are provided, which guarantees the optimality and compatibility with the fast real-time control. We implement the optimum with a grid-connected photovoltaic (PV) power plant by integrating a DVS controller. Dynamic simulations are carried out under different scenarios to test our proposal and compare it with other existing methods. Additionally, the robustness of optimality against model errors is discussed and numerically demonstrated.

eess.SY↗

arXiv2021

Distributed Optimal Conservation Voltage Reduction in Integrated Primary-Secondary Distribution Systems

This paper proposes an asychronous distributed leader-follower control method to achieve conservation voltage reduction (CVR) in three-phase unbalanced distribution systems by optimally scheduling smart inverters of distributed energy resources (DERs). One feature of the proposed method is to consider integrated primary-secondary distribution networks and voltage dependent loads. To ease the computational complexity introduced by the large number of secondary networks, we partition a system into distributed leader-follower control zones based on the network connectivity. To address the non-convexity from the nonlinear power flow and load models, a feedback-based linear approximation using instantaneous power and voltage measurements is proposed. This enables the online implementation of the proposed method to achieve fast tracking of system variations led by DERs. Another feature of the proposed method is the asynchronous implementations of the leader-follower controllers, which makes it compatible with non-uniform update rates and robust against communication delays and failures. Numerical tests are performed on a real distribution feeder in Midwest U. S. to validate the effectiveness and robustness of the proposed method.

eess.SY↗

arXiv2021

Disaggregating Customer-level Behind-the-Meter PV Generation Using Smart Meter Data and Solar Exemplars

Customer-level rooftop photovoltaic (PV) has been widely integrated into distribution systems. In most cases, PVs are installed behind-the-meter (BTM), and only the net demand is recorded. Therefore, the native demand and PV generation are unknown to utilities. Separating native demand and solar generation from net demand is critical for improving grid-edge observability. In this paper, a novel approach is proposed for disaggregating customer-level BTM PV generation using low-resolution but widely available hourly smart meter data. The proposed approach exploits the strong correlation between monthly nocturnal and diurnal native demands and the high similarity among PV generation profiles. First, a joint probability density function (PDF) of monthly nocturnal and diurnal native demands is constructed for customers without PVs, using Gaussian mixture modeling (GMM). Deviation from the constructed PDF is utilized to probabilistically assess the monthly solar generation of customers with PVs. Then, to identify hourly BTM solar generation for these customers, their estimated monthly solar generation is decomposed into an hourly timescale; to do this, we have proposed a maximum likelihood estimation (MLE)-based technique that utilizes hourly typical solar exemplars. Leveraging the strong monthly native demand correlation and high PV generation similarity enhances our approach's robustness against the volatility of customers' hourly load and enables highly accurate disaggregation. The proposed approach has been verified using real native demand and PV generation data.

eess.SP↗

arXiv2021

Distribution Grid Modeling Using Smart Meter Data

The knowledge of distribution grid models, including topologies and line impedances, is essential to grid monitoring, control and protection. However, this information is often unavailable, incomplete or outdated. The increasing deployment of smart meters (SMs) provides a unique opportunity to address this issue. This paper proposes a two-stage data-driven framework for distribution grid modeling using SM data. In the first stage, we propose to identify the topology via reconstructing a weighted Laplacian matrix of distribution networks, which is mathematically proven to be robust against moderately heterogeneous R/X profiles. In the second stage, we develop nonlinear least absolute deviations (LAD) and least squares (LS) regression models to estimate line impedances of single branches based on a nonlinear inverse power flow, which is then embedded within a bottom-up sweep algorithm to achieve the identification across the network in a branch-wise manner. Because the estimation models are inherently non-convex programs and NP-hard, we specially address their tractable convex relaxations and verify the exactness. In addition, we design a conductor library to significantly narrow down the solution space. Numerical results on the modified IEEE 13-bus, 37-bus and 69-bus test feeders validate the effectiveness of the proposed methods.

eess.SP↗