Decision support system for small hydropower systems

tem for small hydropower Received 15 Jun 2021 systems with the implementation of more advanced rescheduling, control Accepted 16th Aug 2021 and forcasting in small hydroelectric system Therefore, a mathematical Available online 19th Dec 2021 model is developed. Particularly, this model uses real-time information of dams. The main objective is to maximize economic value over the time Keywords: decision support systems, horizon by producing electricity when it is most valuable. An approach of simulated annealing algorithm is used to solve this model. evolutionary algorithms, simulated annealing, Copyright © 2021 Hanoi University of Mining and Geology. All rights reserved. hydropower 193 HỘI NGHỊ KHOA HỌC TOÀN QUỐC VỀ CƠ KHÍ – ĐIỆN – TỰ ĐỘNG HÓA (MEAE2021) 1. Introduction Since the early 1970s, decision support systems Modeling (DSS) technology and applications have evolved significantly. Many technological and Implementation organizational developments have made an impact Problem Model Resolutionn on this evolution. Initially, DSSs possessed limited database, modeling and user interface functionality, but technological innovations enabled the development of more powerful DSS Decision Solution functionality (J. P. Shim, 2002). DSSs are, in fact, n computer technology solutions that can be used to support complex decision making and problem Interpretation solving. Decision making is the study of how decisions are actually taken, and how they can be Figure 1. DSS decision-making process better, or more successfully taken (B. Roy, 1993). In a DSS decision-making process (Figure 1), etc.) and/or soft computing systems (evolutionary once the problem is recognized, it is defined in algorithms, fuzzy logic, etc.). Moreover, in the terms that facilitate the creation actors and of the architectures of DSSs, the complexity reduction concerned entities, the definition of the decision tools should not curb the combinatorial horizon, of the parameters and the constraints, and capabilities of the system (I. A. Meystel, 2001). For also the criteria formalization. The resolution stage instance, when dealing with a DSS, such as on imposes a choice of an exact or a heuristic production scheduling systems (PSS), the algorithmic approach. A set of decision proposals is modeling approaches and the resolution tools are then established through the interpretation based on the study and the analysis of concrete stage and presented to the concerned actors. cases coming from real problems. Hence, we The final implementation stage consists in consider the combined task which includes applying the operational decisions, supervising “satisfaction needs cooperation needs their impacts, taking corrective actions, and computational complexity reduction,” as the major validating the decisions. Carlsson and Turban in (C. capability of such a DSS. Carlsson, 2002) state that modern support In order to validate the choice of an agent-based systems research is focused on the theory and approach for the real time management of small application of intelligent systems, and soft hydropower systems, it is necessary to grasp the computing in management. This includes characteristics of such an approach. For this processes of problem solving, planning, and reason, we start by defining agents as conceptual decision making. The context for this research entities that perceive and act in a proactive, or ranges from strategic management, business reactive manner within an environment where process re-engineering, effective collaboration, other agents exist and interact with each other improved user-computer interfaces, and mobile based on shared knowledge of communication and and electronic commerce to production, representation. A multi-agent system (MAS) can marketing, and financial management. The then be defined as a loosely coupled network of methodologies that are used may be analysis or problem solvers interacting to solve problems that system-oriented, action research or case-based, or are beyond their individual capabilities or they may be experimentally or empirically focused. knowledge. MASs constitute a powerful tool for An emerging common denominator for many field handling open, complex, and distributed systems studies, favored in DSS, is the design and use of since they offer modularity and abstraction. intelligent (expert systems, multi-agent systems, Accordingly, an agent-based approach seems the 194 HỘI NGHỊ KHOA HỌC TOÀN QUỐC VỀ CƠ KHÍ – ĐIỆN – TỰ ĐỘNG HÓA (MEAE2021) most appropriate for studying the real-time autonomous and have their own reasoning mode. control of the static economic dispatch problem As shown in Figure 3, this agent model contains a and the dynamic economic dispatch problem knowledge base, a base of strategies and modules within small hydropower systems. In fact, dispatch of communication, as well as, reasoning and problems typically consider the minimum and control. Hence, in order to define its behavior and maximum output constraints of each available unit be able to exchange information with the others as well as their engineering characteristics, such as and the environment, an agent has the following: head, release and efficiency characteristics and − Knowledge base that contains all the therefore, require a set of interacting distributed information and data concerning the agent entities. That is, the MADSS has to optimize the itself and the others; different regulation criteria since it can have a − Base of strategies used by the agent in its more global view on the dams than the regulator. reasoning; communication module that is The present MADSS for small hydropower systems responsible for the messages exchange; consists of the following two modules: − Reasoning module that defines and The supervision module, responsible for the complements the methods which allow the supervision of the creation of databases on the agent to make decisions concerning the task dams and also for continuous updating of the to perform. This module uses the data given geological, survey, and technological data; by the knowledge and strategies modules. The regulation module, responsible for the − A control module that ensures the cohesion in disturbance analysis and the generation of the the agent by the management of the internal appropriate rescheduling measures. It is tasks. It activates the internal modules and composed of the agents INCIDENT, ZONEPERT, undertakes an updating of the knowledge and ZONEREG. The agents of the two modules according to the evolution of the agent and the communicate with each other in order to environment. cooperate in the real-time treatment of the different incidents (Figure 2). The regulation module has a hierarchical organization with horizontal and vertical communication. Figure 2. MADSS modules and agents Figure 3. Agent model The roles of the agents will be explained in Sections 2-3-4-5, then some conclusions are finally 3. Supervision Module shown in Section 6, 7. This module controls the theoretical schedules 2. Agent Model by a time space representation of the network. It Agents constitute the basic entities of a MAS. operates in normal and disturbed conditions. It They have a specific model that allows them to be includes the agents of type Geological, Survey and Technological. 195 HỘI NGHỊ KHOA HỌC TOÀN QUỐC VỀ CƠ KHÍ – ĐIỆN – TỰ ĐỘNG HÓA (MEAE2021) Agent Geological: These agents intend for of the information related to the space-time zone the storage of primary geological information and affected by the disturbance. In order to control the integration of the data. The primary geological evolution of the disturbances, it is necessary to information includes core-sample data on bed define first the space-time limits of the search intersections and intersections of weather space. That is, a space-time horizon has to be forecast, obtained from geological rifts and identified by defining the set of hydropower units trenches. affected by the disturbance and the rescheduling − Agent Survey: These agents represent measures, according to the real state of the dams. information from performance monitoring Moreover, since the disturbance evolves according module. to time and space, the considered horizon has to be − Agent Technological: These agents intend for adapted to the real changing conditions of the the storage of information on the technical dams. It has then to be a dynamic space-time potential of the dams, the parameters of all the horizon or window. The schedules that are technological systems used in the dams or situated beyond this horizon should be equal to the considered as options at the design stage. theoretical ones. Consequently, the starting and 4. Regulation Module ending points of hydropower units in systems have to be respected. It cooperates therefore, with a This module contains the agents INCIDENT, society of agents Geological and Technological, ZONEPERT, and ZONEREG. It operates in called ZonePert representing the horizon. disturbed conditions. It is responsible for the Moreover, ZONEPERT generates, at a first identification, analysis, and resolution of the strategic level, some regulation decisions through incidents. This rescheduling process needs several a rule-based approach that describes the nature of simulations in order to forecast the impact of the the rescheduling measures adapted to the type of incidents and the regulation decisions on the dams. the incident. The MADSS regulation module has in fact, a c) Agent ZONEREG: This agent is created by hierarchical organization that can be considered as INCIDENT. It operates by an anytime evolutionary an expert community where each agent is regulation approach that takes into account the specialized for performing a particular task and the several rescheduling criteria and the solutions solutions are constructed through a mutual proposed by ZONEPERT. Through a comparison adjustment. between the situations before and after regulation, a) Agent Incident: An agent Survey, associated ZONEREG considers the regularity that have been to, creates, at, an agent INCIDENT when a previously stated. disturbance caused by appears. Being responsible This agent considers the present decision- for the considered disturbance, this agent first making problem as a dynamic economic dispatch identifies its characteristics (disturbed, stop, delay, problem which is a mathematical optimization cause, etc.). Then, it creates an agent ZONEPERT problem which can identify how to optimally for the analysis and the first-level regulation of the manage one or more hydropower units over a incident. According to the importance of the specified time horizon. The time horizon disturbance, INCIDENT can decide to create an considered might consist of a day (24-hours), a agent ZONEREG that will generate several possible week (168-hours) or some other period. In fact, it rescheduling solutions through a simulated is characterized with an important number of annealing (SA) approach. This agent has then a variables, a multi-objective, and nonlinear coordination role in the rescheduling process. objective function and discrete variables. These agents propose the relevant final 5. The Dynamic Economic Dispatch Problem rescheduling measures to the regulator. b) Agent Zonepert: This agent has a diagnosis The hydropower plant operator is faced with a role. It is responsible for the gathering and analysis challenging dynamic optimization problem. Given 196 HỘI NGHỊ KHOA HỌC TOÀN QUỐC VỀ CƠ KHÍ – ĐIỆN – TỰ ĐỘNG HÓA (MEAE2021) the amount of water available for release and the 6. Simulated annealing algorithms (SA) anticipated price of electricity over a particular As one of the widely used heuristic approaches time horizon (T), the plant operator must decide (including genetic algorithm and local search) to how much water to release for generation in each solve combinatorial problems, simulated period (t) in order to maximize the economic value annealing (SA) can produce a good though not of the electricity produced. Typically, the total necessarily global optimal solution within a amount of water available for release (Q) over the reasonable computing time. Simulated annealing is planning horizon is fixed and known. The vector of a Monte Carlo simulation-based search algorithm. prices (P) over the planning horizon (T) is The term “simulated annealing” is derived from a assumed or anticipated, based on prior experience process of heating and then cooling a substance and knowledge. slowly to finally arrive at the solid state. In this In general, the optimal dynamic dispatch simulation, a minimum of the cost function problem can be written in mathematical notation corresponds to this ground state of the substance. as shown in equations (1) through (4). T The whole search algorithm simply mimics the Maximize Pgt t q t (1) physical process as below. In the early stages of the 1 execution, the temperature is high, which results in Subject to: a higher probability for jumping to occur more T frequently. In this case, the frequent jumping, qQ (2)  t which occurs as a way of avoiding local minima, 1 q q  q  1... T (3) may produce a higher probability of a poor mintt max solution. In another way, simulated annealing g g  g  1... T(4) mintt max selects the next point randomly. If a lower cost Where: solution is found, it is selected. If a higher cost Pt: Price ($/MWh) at time (t) solution is found, it has a nonzero selection gt: generation (MW) at time (t) probability. The function that governs the qt: release at time (t) behaviour of the acceptance probability is called Q: total release the cooling schedule. As the execution time elapses, qmax: maximum release the temperature decreases, and the cooling qmin: minimum release schedule reduces the frequency of jumping. gmax: maximum generation level The simulation process terminates after a gmin: minimum generation level number of successive executions with no T: planning horizon improvements, and returns the best solution In practice, the operator attempts to maximize found. The following code provides an illustration economic value over the time horizon by of the SA algorithm in pseudo-code (Eglese, et al., producing electricity when it is most valuable. 1990): While doing so, we cannot exceed the amount of Select an initial state iS water available for release over the time horizon Select an initial temperature T  0 (equation 2), must respect the minimum and Set temperature change counter t = 0 maximum release levels (equation 3), must Repeat respect the minimum and maximum generation Set repetition counter (number of iterations to levels (equation 4). be performed at each temperature) This problem falls into the class of Repeat mathematical problems known as constrained Generate state j optimization problems. Depending on the nature a neighbour of i of the generation and head relationships, the  =−f()() j f i problem may be highly nonlinear. Calculate If  = 0 then ij= 197 HỘI NGHỊ KHOA HỌC TOÀN QUỐC VỀ CƠ KHÍ – ĐIỆN – TỰ ĐỘNG HÓA (MEAE2021) Else if random (0, 1) < exp(− /T ) then ij= n+ =1 Until n= N() t t+ = 1 T = T (t) Until stopping criteria is true. As can be seen, the annealing schedule consists of: - the initial value of: T - a cooling functions - the number of iterations N(t) to be performed at each temperature - a stopping criterion to terminate the algorithm. Figure 4. Location of the sampling sites In SA, the algorithm attempts to avoid entrapment in a local optimum by sometimes The application of the solution given by our SA accepting a neighborhood movement, which is illustrated in Figure 5. increases the value of the objective function. The acceptance or rejection of an uphill move is determined by a sequence of random numbers, but with a controlled probability. The probability of accepting a move, which causes an increase  in f is called the acceptance function and is normally set to exp(− /T ) where T is a control parameter, which is analogous to temperate in a physical annealing. In this paper, for the model describe in section 5, the algorithm was coded in Visual C# 2017 and implemented on an Intel(R) Core (TM) i7-4790 Figure 5. Decision support system tool in Thac xang with 3.6GHz CPU. 6.1. Simulation results The results of using the SA algorithm for the Consider a small hydropower system at Lang problem (1) applied in Thac xang hydropower Son in Vietnam (Figure 4) have brought high economic efficiency compared - Factory name: Thac Petrol Hydroelectric to before use the application of decision support Plant tools Figure 6, Figure 7. - Location: Hung Viet Commune, Trang Dinh Dist., T. Lang Son - Name of the river: Bac Giang - Factory type: After the dam - Number of units: 02 - Capacity: 20MW - Useful reservoir capacity: 13.91 million m3 - Basin area: 2660 km2 Figure 6. Daily hyeto-hydrograph in Thac xang 198 HỘI NGHỊ KHOA HỌC TOÀN QUỐC VỀ CƠ KHÍ – ĐIỆN – TỰ ĐỘNG HÓA (MEAE2021) References 1. J. P. Shim, M. Warkentin, J. F. Courtney, D. J. Power, R. Sharda, and C. Carlsson, “Past, present, and future of decision support technology,” J. Dec. Support Syst., vol. 33, no. 2, pp. 111–126, June 2002. 2. B. Roy and D. Bouyssou, Aide Multicritère à la Décision: Méthodes et Cas: ECONOMICA, 1993. Figure 7. Cumulative curve in 3 years 3. C. Carlsson and E. Turban, “DSS: directions for 7. Conclusion the next decade,” J.Dec. Support Syst., vol. 33, In our paper, we present a decision support no. 2, pp. 105–110, June 2002. system for small hydropower systems that provide 4. I. A. Meystel, “The tools of intelligence: Are we hydro power equipment operator with smart enough to handle them?,” in Proc. information required to optimize dams European Workshop Intelligent Forecasting, performance in terms of power efficiency and DiagnosisControl, Santorini, Greece, June 24– effectiveness. 28, 2001, pp. 2–4. The efficiency of the DSS tool was tested using a 5. Claudio J.C. Blanco, Yves Secretan, André L. real numerical example. As perspective of this Amarante Mesquita , ‘Decision support research work need in the comparison and the system for micro-hydro power plants in the combination with other methods, develop other Amazon region under a sustainable control strategies. development perspective’ Energy for Sustainable Development • Volume XII No. 3, Acknowledgment pp.13-21, September 2008 The authors are grateful to the Thac xang 6. Eglese, R.W (1990). Simulated annealing: A company for financial support of the work. Tool for Operational Research, European Journal of Operational Research, Vol. 46, pp.271-281. 199