Search

Fossil fuels represent the highest share of the world's energy sources for thermal energy generation (IEA Report, 2019), of which a fraction is used in District Heating Networks (DHN). Therefore, decarbonising these district heating sectors would diminish \chemform{CO_2} emissions. This research tackles this issue by considering energy efficiency improvement, which can be effectively accomplished by optimising DHNs. Though there is an increase in the utilisation of renewable sources in district heating thermal energy generation, optimisation is still paramount to minimise energy consumption and cost, and extend the life span of the generational unit components. Several optimisation approaches can be applied to District Heating Systems. However, an efficient optimisation technique is Dynamic Real-time Optimisation (DRTO) due to the need to consider real-time control and operation of DHNs, for design or operation improvement purposes. An initial stage of the DRTO is the planning phase to obtain the optimal design parameters, optimal trajectories of the operating parameters and topology of the network. The planning phase cannot be efficiently realised without the foundational models of the DHN. This research is part of the RESEAUDATA project, which aims to improve heat network management through machine learning and dynamic optimisation techniques. The machine learning agent must be trained with Big Data from a process simulation to derive a black box model which fits real-time optimisation techniques due to its low computational time. This paper focuses on modelling the DHN to accomplish the optimal planning phase of DRTO and simulating the network to generate the Big Data to form and train the machine learning model. A transient model is developed from the principle of general mass and energy balance around the pipes and the unit components in the considered District Heating System, giving a Partial Differential Equation (PDE) after eliminating the algebraic equations. The space domain of the partial differential equation is discretised using the Finite Element Backward Difference Approach, and the time domain is retained in its differential form to give a system of Ordinary Differential Equations. A case study of ten (10) consumers in Pau, France, was considered. These consumers include a residence, Large office, Hospital, Hotel, Strip Mall, Restaurant, Super Market, Apartment, Secondary School and a Warehouse with varying heat demands over time. The supply temperature and mass flow rate of the heat transfer fluid at each consumer were simulated with time alongside the network temperature during the forward and return flow distribution. The simulation of the hydraulic parameters in the time domain was also considered. The general principle for building and training efficient machine learning models that could represent the knowledge-based model of a District Heating Network was explicitly explained.

National audience

ACL; International audience

ACL; National audience

International audience

International audience; Nous présentons un modèle de transfert pour la distillation catalytique utilisant les équations de Maxwell-Stefan et incluant la diffusion effective dans le catalyseur poreux. L’écriture et la résolution de ce modèle ont été adaptées à son inclusion dans un simulateur commercial. Il est donc robuste et souple. Nous étudions l’infl uence des caractéristiques du catalyseur, de la confi guration de la colonne, de la non-idéalité du mélange et de la qualité de l’échange liquide/vapeur. La conclusion principale de ces essais est l’existence d’un optimum de fonctionnement notamment pour le taux de refl ux et l’emplacement du catalyseur.

International audience; An industrial application of mathematical programming is presented. The objective is to propose a tool for computer aided design of compact plate fin heat exchangers. A proprietary sizing procedure, COLETH, is used to initialize the optimization problem and to evaluate the objective function and the constraints during the optimization problem resolution. The main features of COLETH are presented. The partial infeasible path strategy is discussed: at each optimization step, duty requirements are satisfied, but pressure drop constraints are not. Intermediate information can be used and the cpu time is quite low. A non linear programming optimization problem is formulated. Then the problem is solved using a reduced Hessien Successive Quadratic Programming algorithm, which can take advantage of the system sparcity. Other features of this algorithm are discussed. Three test problems are presented and the program abilities are illustrated.