1,721,110 research outputs found
MPC approaches for modulating air-to-water heat pumps in radiant-floor buildings
No full textA modulating heat pump and water tank result in a nonlinear model due to the load dependency of the heat pump performance, and variable water flows. Nonlinear model predictive control is an effective way to deal with many physical constraints and nonlinear formulations. Alternatively, linear time-varying MPC can be used, based on successive linearizations around a reference trajectory. The goal of the paper is to analyze the advantages and disadvantages of those MPC techniques for temperature control in radiant-floor buildings. The results show that nonlinear on-line optimization is real-time feasible for the application considered here, as the slow dynamics allows for a fairly long sampling time. Alternatively, the linear time-varying MPC approach shows a significantly better performance compared to the Standard MPC scheme if a feasible reference trajectory is provided. Nonlinear MPC can save up to 6% energy and improve the comfort by 4% with respect to Standard MPC for the given application, while its difference with LTV-MPC is negligible. Moreover, a robustness analysis has been conducted, showing the impact of the heat pump efficiency on the control performance
Model-Based Predictive Control Under Uncertainty
Full text linkThis thesis focuses on model-based predictive control under uncertainty, with a specific emphasis on systems affected by uncertainty characterized by worst-case geometric bounds and probability distributions. The research topic is motivated by the necessity for computationally efficient controllers tailored for uncertain dynamical systems, where safety of the controlled systems needs to be guaranteed to an acceptable level. A collection of novel algorithms addressing robust and stochastic model predictive control (MPC) problems has been reported.
In contrast to nominal MPC, which is widely employed in many control scenarios, robust and stochastic MPC have encountered fewer real-world applications primarily due to their numerical and theoretical challenges. However, uncertainty is a critical factor in many control scenarios. In MPC community, uncertainty is primarily addressed through two approaches: robust MPC and stochastic MPC. Robust MPC guarantees that all possible future state and control trajectories adhere to constraints while minimizing the “worstcase” cost or some generalized costs. In contrast, stochastic MPC aims to minimize an expectation cost or some generalized costs while incorporating chance (probabilistic) constraints. This approach allows the controlled system to deviate from constraints to an acceptable extent. The primary focus of this thesis is dedicated to the design of these two classes of approaches. Although strictly speaking, robust and stochastic MPC belong to distinct categories, robust MPC tends to yield conservative control policies since it disregards the stochastic information about the bounded uncertainty under consideration, if such stochastic information is available. Additionally, due to the utilization of chance constraints, stochastic MPC is capable of handling stochastic uncertainty with unbounded support. On the other hand, addressing the expectation cost and chance constraints, along with analyzing the control-theoretic properties of the controlled systems, present greater challenges in stochastic MPC compared to robust MPC. Similar to nominal MPC, the key properties to examine when analyzing these MPC approaches include the stability, optimality, and constraint satisfaction of the closed-loop controlled systems.
For scenarios where uncertainty is bounded and no further stochastic information is available, this thesis introduces a robust MPC controller using tubes, which is able to exponentially stabilize the controlled system. For scenarios where uncertainty is characterized by stochastic descriptions. Within this context, this thesis introduces two stochastic MPC controllers designed to stabilize the system while adhering to stage-wise (pointwise-in-time) chance constraints, commonly referred to as joint chance constraints in the stochastic MPC community. Subsequently, we investigate mission-wide (dynamic-joint) chance constraints over the state trajectory, which are, in generally, more meaningful in defining safety for engineering systems. In this regard, we offer a characterization of the exact solution to the mission-wide chance-constrained optimal control problems, and subsequently, we approximate these exact solutions using a stochastic MPC approach with shrinking time horizons.IEEE papers in this thesis © IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works
Reinforcement Learning-based Control and State Estimation using Model Predictive Control and Moving Horizon Estimation
Full text linkA new Reinforcement Learning (RL) algorithm based on Model Predictive Control (MPC) has been recently proposed in which the optimal state (-action) value function and the optimal policy can be captured by a parameterized MPC scheme even if the system model underlying the MPC scheme cannot capture the real system perfectly. However, the main idea above was investigated upon the Markov Decision Process (MDP), where a full observation of the states of the real system is needed. Moreover, the idea of using the MPC-based RL can be investigated for other types of MPC schemes such as robust MPC and Linear Parameter Varying- MPC (LPV-MPC).
To investigate the above mentioned ideas and develop new frameworks in the context of MPC-based reinforcement learning, in the first part of this thesis, we investigate the use of the MPC-based RL framework in the context of Partially Observable Markov Decision Process (POMDP). We next show that the core idea of modifying the MPC scheme by RL can also be used for modifying a Moving Horizon Estimation (MHE) scheme so that the MHE performance is improved even if the system model underlying the MHE scheme is imperfect. Moreover, we propose an MHE/MPC-based RL in the context of LPV systems. In the second part of the thesis, we investigate the use of the MPC-based RL for an approximate Robust Nonlinear MPC (RNMPC). We then use a second-order Q-learning algorithm to adjust a set of parameters attached to this approximate RNMPC scheme aiming to achieve the best closed-loop performance.
In the context of POMDP, we propose an observer-based framework for solving POMDPs, where the real system is partially observable. We first propose to use a Moving Horizon Estimation-Model Predictive Control (MHE-MPC) scheme in order to provide a policy for the POMDP problem, where the states of the real system are not fully measurable and necessarily known. We propose to parameterize both the MPC and MHE formulations, where certain adjustable parameters are regarded for tuning the policy. In this work, for the sake of tackling the unmodeled and partially observable dynamics, we leverage the RL to tune the parameters of MPC and MHE schemes jointly, with the closed-loop performance of the policy as a goal rather than model fitting or the MHE performance.
To deal with the model-based state estimation problems with imperfect models, we next present a reinforcement learning-based observer/controller using MHE and MPC schemes, where the model used in the MHE-MPC scheme cannot accurately capture the dynamics of the real system. We show how an MHE cost modification can improve the performance of the MHE scheme such that a true state estimation is delivered even if the underlying MHE model is imperfect. A compatible Deterministic Policy Gradient (DPG) algorithm is then proposed to directly tune the parameters of both the estimator (MHE) and controller (MPC) aiming to achieve the best closed-loop performance.
The LPV models use a linear structure to capture time-varying and nonlinear dynamics of complex systems. These models then facilitate the formulation of computationally efficient design algorithms for observers and controllers synthesis of nonlinear systems. In the LPV framework, we propose an MHE/MPC-based RL method for the polytopic LPV systems with inexact scheduling parameters (as exogenous signals with inexact bounds), where the Linear Time Invariant (LTI) models (vertices) captured by combinations of the scheduling parameters becomes wrong. We first propose to adopt an MHE scheme to simultaneously estimate the convex combination vector and unmeasured states based on the observations and model matching error. To tackle the wrong LTI models used in both the MPC and MHE schemes, we then exploit a Policy Gradient (PG) to learn both the estimator (MHE) and controller (MPC) so that the best closed-loop performance is achieved.
In the context of robust MPC, we present an RL-based Robust Nonlinear Model Predictive Control (RL-RNMPC) framework for controlling nonlinear dynamical systems in the presence of disturbances and uncertainties. An approximate RNMPC of low computational complexity is used in which the state trajectory uncertainty is modelled via ellipsoids. Reinforcement Learning is then used in order to handle the ellipsoidal approximation and improve the closed-loop performance of the scheme by adjusting the MPC parameters generating the ellipsoids
Design and Implementation of the Dual-Hormone Artificial Pancreas in Animal Studies – A Model Predictive Control Approach with Intraperitoneal Hormone Injections
Full text linkIn this thesis, the exploration of new ways to help people with Type 1 Diabetes better control their condition is described. In this research, our primary focus was on the feasibility and benefits of utilizing the intraperitoneal (IP) route for insulin and glucagon injections in Type 1 DiabetesMellitus (T1DM), chosen due to its significantly faster absorption and more rapid effects on glucose levels compared to the subcutaneous (SC) route. The core contribution of this research lies in the development of a fully automated dual hormone AP system and testing in various animal experiments.
In the first major part of this work, a new model is introduced, designed with a minimal number of parameters and states, exclusively intended for control applications within dual-hormone AP systems. Demonstrating remarkable prediction accuracy in over 30 animal experiments, this model represents a significant advancement that has the potential to facilitate future advancements in diabetes management.
Subsequently, an estimator based on the Moving Horizon Estimation (MHE) method is designed, incorporating embedded prior knowledge to effectively estimate non-measurable states of the model, as well as meals and exercises. The experimental evaluation showcases the high accuracy of the estimator, further validating its potential as a valuable tool in diabetes care future.
The work proceeds with the development of an MPC-based controller, adeptly incorporating practical considerations. Extensively tested in both in vivo and in silico experiments, the controller demonstrates high performance, surpassing existing Hybrid Closed-Loop AP systems in the market. Importantly, the proposed controller does not necessitate the meals and exercise announcements, enhancing its user-friendliness and autonomy compared to the commercial devices which all require meal announcements.
Beyond the primary research target, this study delves into various other areas within diabetes management. The investigation includes testing a two-layer PID controller scheme, developing a method to compensate for CGM sensor time lag, exploring sensor fusion techniques to enhance glucose measurements, and studying experimental design strategies to increase model parameter identification accuracy.
The findings of this research contribute to the advancement of diabetes research, which in turn may result in advances in diabetes care. The proposed model, estimator, and controller collectively offer a comprehensive and efficient solution for achieving reliable glycemic control in T1DM patients with IP injections. Ultimately, this work represents a vital step forward in personalized care and opens new avenues for future research and technological innovations
Optimal Coordination of Automated Vehicles at Intersections: Theory and Experiments
Full text linkWith the introduction of Cooperative Automated Vehicles, traffic lights can be replaced by coordination algorithms. In this paper, we present a bi-level, model predictive controller for coordination of automated vehicles at intersection. The bi- level controller consists of a coordination level, where intersection occupancy timeslots are allocated, and vehicle-level controllers, where the control commands for the vehicles are computed. We establish persistent feasibility and stability of the bi-level controller under some mild assumptions, and derive conditions under which closed-loop collision avoidance can be ensured with bounded position uncertainty. We thereafter detail an implemen- tation of the coordination controller on a three-vehicle test bed, where the intersection-level optimization problem is solved using a distributed Sequential Quadratic Programming (SQP) method. We present and discuss results from an extensive experimental campaign where the proposed controller was validated. The experimental results indicate the practical applicability of the proposed controller, and validates that safety can be ensured for large positioning uncertainties
Machine Learning Based Digital Twins for Temperature and Power Dynamics of a Household
Full text linkUtfordringer med kraftproduksjonen i Europa og den stadig økende elektrifiseringen av samfunnet har resultert i økte kostnader ved strømforbruk de siste par årene. Dette har stor innvirkning på økonomien til husholdninger i Norge. Oppvarming utgjør en stor del av strømforbruket i boliger, og bruk av smart-styring av oppvarming kan redusere strømforbruket drastisk - spesielt når det er høy belastning på strømnettet, og strømprisene er høye. Modellbasert prediktiv kontroll (MPC) er bevist å være en god kontrollmetode for inneklima. Metoden utnytter svingende spotpriser til å redusere kostnadene ved oppvarming samtidig som en ønsket temperatur opprettholdes. MPC-metoder krever gode system-modeller til å simulere temperatur og strømforbruk for varmeapparater. Å lage disse modellene krever ofte omfattende simuleringer og datainnsamling, og er dermed en ressurskrevende metode.
I denne masteroppgaven brukes rene maskinlæringsmodeller for å modellere dynamikken til innendørs temperatur og varmepumpe-strømforbruk. Modelleringen er utført ved hjelp av data samlet inn fra en husstand i Trondheim som kjører en MPC-basert kontrollalgoritme for varmepumpestyring. Lineære direkte- og lineære iterative one-step-ahead-prediksjonsmodeller lages for å modellere temperaturdynamikken. De sammenlignes med den dyplærings-baserte modellen, Temporal Fusion Transformer (TFT). Alle modellene trenes til å gi stokastiske prediksjoner. Den lineære direkte modellen utkonkurrerer den itererative og viser lignende ytelse som TFT-en når den trenes på store datasett. TFT-en overgår de lineære modellene når lite data er tilgjengelig for trening. Den lineære direkte modellen er lovende og klarer å lære temperaturdynamikker i huset som reflekterer de sanne dynamikkene. Lineære og dyplærings-baserte prediksjonsmodeller trenes til å forutsi varmepumpens strømforbruk stokastisk. Den dyplærings-baserte modellen består av dype nevrale nett for hver varmepumpe. De lineære modellene utkonkurrerer dyplæringsmodellene på små datamengder, og deres relative ytelse på større data er ganske lik. Alle foreslåtte modeller for prediksjon av strømforbruk sliter med å lære den sanne dynamikken til varmepumpenes strømforbruk.Challenges with power production in Europe and the ever-increasing electrification of society have resulted in an increased cost of power consumption in the last couple of years. This has a significant impact on the household finances of the everyday consumer in Norway. Heating is a major part of residential power consumption, and utilizing smart control of heating could reduce power consumption drastically - especially during peak power spot prices. Model Predictive Control (MPC) is proven to be a good control scheme for indoor climate. The method exploits fluctuating spot prices to reduce power consumption bills whilst maintaining desired indoor temperatures. MPC schemes require good system models to simulate the indoor temperature and power consumption of heating appliances. Modeling these dynamics often require extensive simulations and data gathering.
In this Master's Thesis, pure machine learning methods are used as modeling tools for temperature and heat pump power consumption dynamics. The modeling is performed using data collected from a household in Trondheim that runs an MPC scheme for heat pump control. Direct linear and iterative one-step-ahead linear forecasting models are created to model the temperature dynamics. They are compared to the deep learning-based Temporal Fusion Transformer (TFT) model. All models are trained to give stochastic forecasts. The direct linear model outperforms the iterated one and shows similar performance as the TFT when trained on large data sets. The TFT outperforms the linear models when little data is available for training. The direct linear model shows great promise and manages to learn the temperature dynamics of the household quite well. Linear and deep learning-based prediction models are trained to give stochastic predictions for heat pump power consumption. The deep learning-based model consists of deep neural networks for each indoor heat pump unit. The linear models outperform the deep learning on little data, and their relative performance on longer data is quite similar. All proposed power consumption prediction models struggle to learn the true dynamics of the power consumption of heat pumps
Going Beyond Counting First Authors in Author Co-citation Analysis
Full text linkThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
CNN-based Remaining Useful Life Prediction of Lithium Ion Batteries Using Fast-Rate Incremental Capacity
Full text linkLithium-ione-batterier spiller en sentral rolle i det grønne skiftet gjennom bruk i energilagring og elektrifisering av transportsektoren. Det er midlertidig utfordringer knyttet til forringelse av lithium-ione-batterier, noe som medfører redusert ytelse og levetid. Dette er en høyst kompleks prosess, noe som gjør maskinlæringsmodeller spesielt egnet for å estimere helsetilstand og levetid til batterier. Nøyaktig estimering og modellering av batteriets helsetilstand er avgjørende for sikker og effektiv drift, samtidig som det åpner for økt innsikt i underliggende årsaker til batteriforringelse.
Dette arbeidet presenterer et enkelt 1D konvolusjonelt nevralt nettverk (CNN) for prediksjon av gjenværende levetid, både i antall ladesykler og normalisert mot total levetid, basert på informasjon fra inkrementell kapasitet (IC) under hurtigladning. Gjennom analyse av beslutningsprosessen til CNN-et undersøkes hvilke spenningsområder og inngangssignaler som modellen relaterer til redusert gjenværende levetid.
CNN-et oppnår en RMSE og MAD på henholdsvis 171 og 130 sykler. Ved prediksjon av normalisert gjenværende levetid, er RMSE og MAD på henholdsvis 0,075 og 0,057. Til tross for en enkel arkitektur og begrenset inngangssignal, demonstrerer modellen evnen til å identifisere relevante sammenhenger mellom IC kurver og levetid. Videre undersøkelse tyder på at CNN-et assosierer redusert levetid med IC-reduksjon i spenningsregionen forbundet med interkalering av anoden. Mønstrene gjenkjent av modellen kan dermed tilskrives redusert tilgjengelig kapasitet i dette spenningsvinduet. I tillegg ser også signal ved lavere spenninger ut til indikere redusert RUL.
Samlet sett demonstrere arbeidet hvordan CNN-prosesserte IC kurver fra hurtiglading kan benyttes for levetidsestimering. Funnene bidrar til en bedre forståelse av beslutningsprosessen og gir innsikt i forringelsesprosessen.Lithium-ion batteries (LIBs) are essential in transitioning to clean and renewable energy, serving as energy storage and power sources in transportation. However, battery degradation poses a significant challenge to their durability and performance. Due to the inherent complexity of the degradation process, machine learning methods have emerged as a promising modeling approach for SOH estimation and lifetime prediction due to their ability to model complex processes solely from data. Accurate estimation of the state of health (SOH) is essential to ensuring safe and effective operation, potentially revealing new insights into the underlying degradation process.
In this work, we propose a shallow 1D convolutional neural network (CNN) architecture to leverage information in fast-rate incremental capacity (IC) curves for predicting the remaining useful life (RUL) of LIBs, both in units of cycles and normalized to the total lifetime. Additionally, we look into the decision-making process of the CNN by identifying voltage regions and input patterns that correlate with reduced RUL.
The CNN demonstrates a root mean squared error (RMSE) and a mean absolute deviation (MAD) of 171 cycles and 131 cycles, respectively, when predicting RUL. For normalized RUL, the predictive performance improves, achieving an RMSE and MAD of 0.075 and 0.057, respectively. Despite the shallow architecture and limited input features, the CNN demonstrates the ability to connect features in the IC curves to remaining life. Furthermore, The CNN identifies consistent patterns indicating cell degradation, particularly related to IC peak reductions in voltage regions associated with graphite staging. The patterns can be attributed to reduced accessible capacity in this voltage window. Furthermore, the appearance of peaks at lower voltages may also indicate decreased RUL.
Overall, this work demonstrates the effectiveness of CNN-processed fast-rate IC difference curves for lifetime predictions of LIBs. The findings contribute to a better understanding of the CNN's decision-making process and provide insights into degradation patterns in LIBs
IoT Software for Smart Houses
Full text linkDette prosjektet har som mål å lage en programvare for et Tingenes Internet (IoT) basert avansert oppvarmingsstyringssystem, med det formål å redusere energiforbruket og toppeffektbelastning i smarte bygninger.
Systemet er basert på å bruke Systemidentification (SYSID) til å generere en dynamisk temperaturmodell av huset, og bruke Modell Prediktiv Kontroll (MPC) til å kontrollere temperaturen. Selv om bruk av MPC for styring av energi i bygninger ikke er nytt, har tidligere forskning vært basert på bruk av dyrt utstyr og kompliserte oppsett som ikke er tilgjengelig for den gjennomsnittlige personen og hjemmet. Med fremveksten av IoT enheter eksisterer det nå billig og kommersielt tilgjengelig utstyr som er enkelt å sette opp for allerede eksisterende hjem. I dette prosjektet var det ønsket å undersøke levedyktigheten av å bruke slike enheter til et MPC-basert varmesystem.
Programvaren som presenteres i denne oppgaven bruker et flertrådet datainnsamlingssystem sammen med et webbasert Applikasjonsprogrammeringsgrensesnitt (API) for å betjene MPC-algoritmen med data. Resultatene viste at denne typen system og programvare er levedyktig og har potensial.
Dette arbeidet fungerer som grunnlag for videre forskning på dette området. Videre forskning bør fokusere på å forbedre programvaren, samt undersøke forskjellige temperaturmodeller og kontrollalgoritmer for bedre ytelse.This project aims to create a software for an Internet of Things (IoT) based advanced heating control system, for the purpose of reducing energy consumption and power peak load in smart buildings.
The system is based on using System Identification (SYSID) to generate a dynamic temperature model of the house, and using Model Predictive Control (MPC) to control the temperature. While using MPC for managing energy in buildings is not new, previous research has been based on using expensive equipment and complicated setups that are not readily available for the average person and home. With the emergence of IoT devices, there now exists cheap and commercially available equipment that is easy to set up for already existing homes. In this project we wanted to investigate the viability of using such devices for an MPC based heating system.
The software presented in this thesis uses a multithreaded data collection system together with a web-based Application Programming Interface (API) to service the MPC algorithm with data. The results showed that this kind of system and software is viable and has potential.
This work serves as laying the ground work for further research in this area. Further research should focus on improving the software and, as well as investigate different temperature models and control algorithms for better performance
- …
