Optimizing the scheduling of electricity production by an electricity generation park

Abstract

The invention relates to a technique for optimizing the electrical power production schedule of a power generation facility comprising a plurality of generating units. A multi-agent system (23), generated from a formal representation (10) of said generating facility, constructed from an ontology (11) and a knowledge graph (12), simulates, based on a received production setpoint (16), the electrical power output of said generating units and generates, for a reference period, an optimized electrical power production schedule, comprising a production setpoint (18) adjusted according to the simulated electrical power output of said generating units and taking into account operating states and representative values ​​of the operating parameters specific to each unit and updated (17) in said formal representation. Abstract figure: Figure 1

Description

Title of the invention: Optimization of an electrical energy production schedule by an electrical generation park technical field

[0001] This disclosure falls within the field of energy systems management and optimization, and more particularly within the field of optimizing the production of electrical energy by a production fleet comprising several power plants, such as hydroelectric power plants, thermal power plants or nuclear power plants. Previous technique

[0002] In recent years, the energy sector has undergone a significant transformation, characterized by the decentralization of production and a substantial migration of electricity production methods towards renewable energy sources, involving a plurality of actors.

[0003] Moreover, power plants are increasingly vulnerable to climatic hazards, which change frequently. For example, when it is too hot, some nuclear power plants are forced to suspend their production, which may require the grid operator to reprogram the production demand in other power plants in the same valley in order to be able to meet the needs (e.g. an increase in the production and water storage capacity in hydroelectric plants located upstream or downstream of the nuclear power plant).

[0004] This shift towards a complex and decentralized energy system poses increasing challenges in terms of managing the balance between supply and demand on electrical networks, as well as in designing suitable management and optimization algorithms for companies in the energy sector.

[0005] The current model for forecasting and scheduling electricity production is based on a mathematical approach that assumes homogeneity among the different generating plants. However, it no longer adequately reflects the current reality of the electricity sector, as described above. Indeed, this overall mathematical model relies on the existence of several production models, each associated with a type of generating unit, for example, a model for nuclear power plants, a model for thermal power plants, and a model for hydroelectric power plants. However, these models do not take into account the local constraints affecting each power plant (for example, a meteorological constraint or a constraint structural technique of the power plant), and also do not take into account regional constraints, linked for example to interactions between power plants resulting from their geographical location.

[0006] To overcome this drawback, the Applicant initially attempted to adjust existing algorithms for generating electricity generation schedules for power plants by incorporating, into the mathematical equations underlying their operation, fictitious values ​​intended to represent the specific characteristics of each power plant. It should be noted that a generation schedule contains the active power cycle of an electricity generation source for a given day J+1, and therefore contains the production time slots and the desired power levels during each of these slots for a generation source, for that day J+1. However, despite these attempts at numerical adjustment, the approach remains general and fails to comprehensively address the local and heterogeneous constraints specific to each power plant.This often results in a mismatch between call schedules and actual production capacity in a park, requiring frequent adjustments throughout the day.

[0007] Furthermore, to date, these call-programming algorithms are designed for a specific type of power plant. In a power generation fleet comprising power plants of different types, coordination between these various production units therefore requires the fleet manager to consult several separate call-programming systems.

[0008] There is therefore a need for a technique for optimizing the programming of electrical energy production within a multi-sector production park (hydroelectric, thermal, nuclear, etc.), which is integrated within a single system, and which allows taking into account the production constraints specific to each power plant in the park.

[0009] There is also a need for such a technique which is scalable, and which can adapt dynamically to changes in the behavior of the production fleet in the event of evolution. Summary

[0010] This disclosure improves the situation.

[0011] A method is proposed for optimizing the scheduling of electrical energy production by an electrical production park comprising a plurality of production units, including: a. a receipt of an electrical energy production program, including a setpoint for electrical power production for the production units for a reference period; b. a simulation, based on the production instructions, of a production of electrical power from production units, implemented by a multi-agent system modeling the fleet, i. the multi-agent system being generated from a formal representation of the production park, constructed from an ontology and a knowledge graph comprising a set of nodes each representing an element of one of the production units, and a set of edges each representing a functional relationship between two nodes that they connect, ii. the formal representation being updated: 1. according to a determined time step, from representative values ​​of operating parameters of the nodes, received from sensors placed in the park; 2. on the fly, in the event of a change affecting at least one operating state of one of the nodes; c. a generation, for the reference period, of an optimized electrical power production program, including a production setpoint adjusted according to the simulated electrical power production for the production units and taking into account the operating states and representative values ​​of the operating parameters updated in the formal representation.

[0012] According to another aspect, a system is proposed for optimizing the scheduling of electrical energy production by an electrical power generation park comprising a plurality of production units, including: a. a communication module configured for: i. receive an electrical power production program, including a setpoint for electrical power production for the production units for a reference period; ii. send to the production units an adjusted production instruction from an optimized electrical energy production program for the reference period; b. a semantic database configured to store a formal representation of the park, built from an ontology and a knowledge graph comprising a set of nodes each representing an element of one of the production units, and a set of edges each representing a functional relationship between two nodes that they connect; c. a set of sensors configured to provide, according to a determined time step, representative values ​​of operating parameters of the nodes, intended to feed the formal representation; d. a human-machine interface configured to allow the formal representation to be updated in the event of a change affecting at least one operating state of one of the nodes; e. a multi-agent system modeling the park, generated from the formal representation, and configured to simulate, from the production setpoint, an electrical power production by the production units, the multi-agent system comprising a central programming entity configured to generate, for the reference period, the optimized electrical power production program, including the production setpoint adjusted according to the simulated electrical power production for the production units and taking into account the operating states and representative values ​​of the operating parameters updated in the formal representation.

[0013] According to another aspect, a computer program is proposed comprising instructions for implementing all or part of a process as defined herein when this program is executed by a processor. According to another aspect, a non-transient, computer-readable recording medium is proposed on which such a program is recorded.

[0014] The features described in the following paragraphs may optionally be implemented independently of each other or in combination with each other:

[0015] Such an optimization process also includes sending, to said production units, the adjusted production setpoint of the optimized electrical energy production program for the reference period.

[0016] Such an optimization process also includes a rendering on a human-machine interface of the simulated electrical power production and / or the generated optimized electrical energy production program.

[0017] The park comprises at least two production units belonging to the group comprising: - hydroelectric power plants; - thermal power plants; - nuclear power plants; - photovoltaic energy sources; - wind energy sources; - tidal energy sources; - hydraulic pumping stations; - the batteries.

[0018] The nodes of the knowledge graph represent elements belonging to the group including: - water reservoirs; - pumping stations; - production facilities; - production generators (i.e., turbines or alternators); - production auxiliaries (i.e., internal production sources that provide initial power to start the production process).

[0019] The operating parameters belong to the group comprising: - a volume of water stored; - a range of required water flow rates at the inlet; - an output water flow rate; - electrical power produced; - a range of inlet water temperature; - a required electrical power; - a maximum power range available; - a minimum duration of the upper plateau; - a minimum duration of the lower plateau; - a type of operating point; - a system state (i.e., a positioning of the power plant in the various control mechanisms for the electricity transmission network manager, such as frequency, power or reserve control); - a type of fuel (for example, fuel oil, nuclear, combined cycle gas (CCG), coal, dual-fuel gas, ...); - a required gas power.

[0020] The operating states belong to the group comprising: - an active operating state; - a state of failure; - a state of standstill; - a state of electricity consumption.

[0021] The electrical power production setpoint takes into account a forecast of electrical consumption. Brief description of the drawings

[0022] Other features, details and advantages will become apparent upon reading the detailed description below, and upon analysis of the accompanying drawings, on which: Fig. 1

[0023] [Fig. 1] presents a synoptic diagram of an optimization system for the programming of electricity production of a production park according to a mode of rea- lisation. Fig. 2

[0024] [Fig.2] shows in synoptic form the steps of construction of a formal representation of the production park and generation of a multi-agent simulation system included in the optimization system of [Fig.1] according to an embodiment. Fig. 3

[0025] [Fig.3] details the steps of automatic generation of an SMA code of the [Fig.2] from three programmatic objects according to one embodiment. Fig. 4

[0026] [Fig.4] presents an example of graphic reproduction of an electrical production park of a hydraulic valley on the interface of the system of [Fig.1] according to one embodiment. Fig. 5

[0027] [Fig.5] illustrates the formal representation of the production park of [Fig.4] and the mapping of different classes of the ontology into Java classes according to an embodiment mode. Fig. 6

[0028] [Fig.6] shows an example of the behavior of an agent of a multi-agent simulation system according to one embodiment. Fig. 7

[0029] [Fig.7] illustrates interactions between agents of a multi-agent simulation system of the park of [Fig.4] according to an embodiment. Description of the implementation methods

[0030] The general principle of the invention is based on the design of a production scheduling optimization system for a power generation fleet, founded on the combined use of a multi-agent system and knowledge graphs representing the operation of the power plants in the fleet, and on the use of such a system to optimize the electricity production scheduling of the fleet. Advantageously, such a system makes it possible to simulate the operation of the power generation fleet, taking into account in real time the production constraints specific to each power plant, regardless of its type (hydroelectric, nuclear, thermal, photovoltaic, etc.), and thus to optimize electricity production schedules in an integrated, multi-sector approach.

[0031] The following is considered an electricity production park, comprising a set of power plants of different types or of the same type. It should be noted that a power plant is an industrial site designed for the production of electricity, and which supplies... Electricity is supplied via the electrical grid to consumers, both residential and industrial, located far from the power plant. Electricity production is achieved by converting primary energy sources into electrical energy. These primary energy sources can be mechanical (wind, river, or tidal), chemical (combustion of fossil fuels or biomass), nuclear, or solar. Thus, a generation fleet may include, for example, one or more hydroelectric power plants, one or more thermal power plants, or one or more nuclear power plants or photovoltaic power plants. Throughout this document, such electricity generation sites will be referred to interchangeably as power plants, power plants, or generating units.

[0032] First, with reference to [Fig. 1], we describe the general block diagram of such a system according to one embodiment. Such a system comprises a database in which a formal representation, denoted KR and referenced 10, of the electricity generation fleet under consideration is kept up to date. As will be seen in more detail later with reference to Figures 2 and 3, this formal representation KR 10 is constructed from a referenced ontology 11 and a referenced knowledge graph 12. It is updated regularly: - on the one hand, from operating parameter values ​​which are provided by a set of 15 sensors (electric meters, flow meters, thermometers, anemometers...) placed in the production park; - on the other hand, from referenced updates 17 relating to the operating states of the elements contributing to the production of electricity within the park (state of failure, state of shutdown, state of activity, etc.); these updates can be carried out by the professionals managing the production park, through a referenced HMI 24.

[0033] Such a system also includes a multi-agent simulator (MAS) referenced 23, which is based on the formal representation KR 10, and which allows the simulation of electricity production within the power plant, based on different input scenarios. One of these scenarios can be a referenced setpoint 16 for electricity production, imposed on the different production units of the power plant, for a reference period (for example, the next 24 hours), and received by a receiver module Rx referenced 13. Based on this setpoint 16, the multi-agent simulation system (MAS) 23 simulates electricity production within the power plant, taking into account all the values ​​of the operating parameters and the operating states stored in the semantic database containing the formal representation KR 10.It thus acts as a digital twin of the production facility, allowing for real-time or near-real-time consideration of all local and regional operational constraints that may affect production units.

[0034] Based on the results of this simulation, the SMA 23 multi-agent system can adjust the electricity production schedule according to these various constraints so that it more accurately reflects the actual production capacity of the different generating units in the plant. It thus generates an optimized production setpoint 18, which can be displayed via the HMI 24 and also transmitted by a referenced Tx transmission module 14 to the different generating units. The generating units then know the electrical power levels to be produced, according to a predetermined time step of 5, 10, or 30 minutes, for example, during the next reference period, e.g., the next 24 hours.

[0035] We now present in more detail, in relation to figures 2 and 3, the preliminary steps El and E2 of construction of a multi-agent system, MAS, from knowledge representations, KR, associated with the electricity production fleet.

[0036] As illustrated in [Fig.2], the business knowledge specific to each power plant, as well as the values ​​associated with the elements which contribute to the operation of these power plants, can be systematically expressed through the use of a knowledge representation 10, combining an ontology 11 and a knowledge graph 12 (designated here by the acronym KR for the English “Knowledge Representation”, in French “représentation de connaissances”).

[0037] Ontology 11 acts as a data schema containing concepts and semantic relations to model a body of knowledge in a specific domain, facilitating its understanding by machines. In other words, an ontology defines concepts (principles, ideas, object categories, potentially abstract notions) and relations. It generally includes a hierarchical organization of the relevant concepts and the relations that exist between these concepts, as well as rules and axioms that constrain them. It offers a representation of general properties of what exists in a formalism that supports rational processing. It is the result of an exhaustive and rigorous formulation of the conceptualization of a domain, in this case, that of an electricity generation fleet.

[0038] Knowledge graph 12, for its part, represents a semantic network of the knowledge domain stored in a database in graph form, where the nodes represent the data and the edges describe the relationships between these data. In the context of this document, the nodes depict the different elements involved in electricity production at a power plant (identified by a unique code). For example, a hydroelectric power plant is associated with several elements involved in its electricity production, such as a water reservoir, a pumping station, and a generating unit. The links between these nodes symbolize the relationships between these elements and between the various entities within a generating unit at a specific scale.

[0039] A first step, referenced El, consists of collecting and formalizing all the knowledge of the professionals managing and operating the production facility, using a uniform formal representation method, KR. This step El thus enables a formal representation of the production facility, which professionals can modify at any time to adapt it to changes in the facility (modification of a production installation, addition of a new plant, shutdown of equipment due to breakdown or regulatory reasons, etc. – updates referenced 17 in [Fig. 1]). They can interact with ontology experts or by directly manipulating the knowledge graph, without needing to manipulate the code of the multi-agent system, which will be described below. Furthermore, the latter automatically adapts to changes in knowledge through queries performed on the knowledge graph.

[0040] The formal representation KR 10, obtained at the end of step El, can take the form of symbols that the system can store and manipulate (for example, logical languages ​​and operations, graph structures and operations). This formal representation KR 10 is both understandable by humans and manipulable by systems, by applying manipulation rules defined on the symbols of these representations and whose interpretation simulates, for example, reasoning.

[0041] The formal representation KR 10 is then used as input to the referenced step E2 of automatic multi-agent system (MAS) generation. A multi-agent system, as described, for example, by C. Donzelli et al. in “Onto2MAS: An ontology-based framework for automatic multi-agent system generation”, 2016 12th International Conference on Signal-Image Technology & Internet-Based Systems (SITIS), IEEE, 2016, pp. 381-388, can be defined as a complex system integrating a collection of autonomous agents with their own objectives and actions, capable of interacting, collaborating, and exchanging knowledge through communication channels. Thus, using a multi-agent system, one can numerically model the individual characteristics of each element of the power generation system and observe how these elements interact to influence the overall operation of the generation fleet.

[0042] The formal representation KR 10 feeds a Java code generation tool, referenced 20. This tool allows, on the one hand, the generation, from the ontology 11, of a set of Java classes 21 corresponding to the different agents and objects of the MAS system to be created, and on the other hand, the assignment of behaviors and actions, referenced 22, to the different agents. This tool 20 incorporates a conversion model between the ontology and the Java code, and thus allows the generation of the Java source code associated with all the agents and their behaviors, to form a multi-agent simulation platform, referenced 23. This platform 23 allows the user to visualize, thanks to a referenced HMI 24, the simulation results according to the input scenarios (for example, different production instructions, or different variants of the call program).

[0043] Such a simulation platform 23 can thus be adapted to various types of power plants. By improving the accuracy of generating the power demand, it not only makes it possible to better reflect the actual operational capacity of the plants, but also to verify the feasibility of demand schedules for each plant, taking into account their local and global constraints. Furthermore, it makes it possible to anticipate and examine potential disruption scenarios, including failures of generating units or events related to natural disasters. Such a multi-agent system (MAS), based on a knowledge representation (KR), makes it possible to better guarantee the stability, reliability, and efficiency of the electrical grid, avoiding the time losses associated with power adjustments and the associated financial losses.

[0044] Various attempts have been made in the prior art to combine multi-agent systems (MAS) and ontologies. However, none of these attempts, to the Applicant's knowledge, relied on the use of a knowledge graph that could be updated in real time. The construction scheme of the MAS system illustrated in [Fig. 2] relies on the combined use of ontologies and knowledge graphs, together called KR 10, as input or resource elements of the MAS, rather than as final outputs. Thus, the use of KR makes it possible to define the internal workings of the system's agents, which differs from some prior art approaches that focus more on the general structure of an MAS.

[0045] The automatic generation of an SMA code illustrated by [Fig.2] can be detailed in two steps using three programmatic objects called "Java classes", as illustrated by [Fig.3]: - a launcher of the algorithm, called Main, referenced 25; - an agent construction object, called ConstructorAgent, referenced 26; - a class creation object, called ClassBuilder, referenced 27.

[0046] The creation of classes from ontologies is directed by the programmatic object called "ConstructorAgent" referenced in 26. This step combines the loading of ontologies into memory, referenced in 261, and the generation of functional classes, referenced in 262, through the utility class called "ClassBuilder", referenced in 27. The ontology files, extracted from the KR 10 resource repository, undergo a transformation into Java code. Notably, the main ontology is separated into a distinct model, while the ontologies to which it references are encapsulated in a second model, thus preparing the ground for a judicious selection of the relevant classes in the construction of the final MAS.

[0047] The conversion rule is based on the use of a "ConstructorAgent" 26 which relies on a "ClassBuilder" 27. This programmatic object plays an essential role in the sequential and methodical translation of ontological concepts into executable Java code.

[0048] Hierarchical links between classes are established by exploiting the "SubClass Of" properties, while respecting Java's inheritance constraints.

[0049] Ontological data type properties (owkDatatypeProperty) are converted into variables according to the XSD to Java conversion rules, while object properties (owkObjectProperty) are converted into variables using the class defined by the "range" property (rdfs:range) as the type. Cardinality constraints, in particular the "minimum 1" constraint, guide the algorithm by signaling the representation of properties as lists (ArrayList).

[0050] The ontology conversion function is invoked recursively for each object property when the class of the variable type is defined in external ontologies.

[0051] Finally, the algorithm analyzes all the classes and then uses an Apache FreeMarker®-type engine (a template engine that automatically generates text files from an associative array data model) to efficiently translate the data into Java code, based on specific templates defined by the configuration file. The behaviors, initially coded manually, are integrated into the generated code via additional templates. After the source files are written, concurrent compilation prevents potential circular dependency problems.The Java Compiler receives a "path" module with essential dependencies, allowing MAS agents to take shape and be passed to an MAS modeling platform such as JADE (Java Agent DEvelopment Framework - an open-source, Java-based multi-agent systems development framework), while Java's Reflection API (process of analyzing and modifying all the characteristics of a class) instantiates objects with the compiled classes and enumerations.

[0052] It should be noted that an enumeration, or "enum", is a special data type in which a variable can only take a limited number of values. An enumeration is, in fact, a class. The values ​​of an enumeration are the only possible instances of that class.

[0053] The algorithm's rhythm is maintained by the programmatic object "ConstructorAgent" referenced 26, which oversees the final phase 263 of agent instantiation. This step relies on specific SparQL queries, offering the developer the flexibility to integrate constraints adapted to their context. For example, as we will see in more detail later with a specific implementation example within the context of a valley's production park, res Trictions are applied to select only the elements and entities of the production park belonging to this single valley.

[0054] Initially, a query is launched to obtain the complete list of classes present in the knowledge graph. Then, for each identified class, a new query is formulated to retrieve all related properties, including their cardinality where applicable. When properties have a cardinality, a specialized function is called to significantly reduce the number of queries, thus ensuring optimal performance. Each property is then stored in a Java variable of type Map, where the variable name serves as the key and its value type as the key. If the property corresponds to a class, a third SparQL query is deployed to obtain all the elements necessary for instantiating the agents. The instantiation phase 263 ends by launching an agent of the appropriate class and passing it the Map of variables as an argument.This last step is used at agent startup to initialize its variables using the Reflection API. This pragmatic approach ensures efficient agent initialization, while taking into account the context-specific dependencies of the MAS.

[0055] A specific example of the application of such a multi-agent system in the context of a hydroelectric power generation valley is now described in relation to Figures 4 to 7. The primary objectives of the electricity generation scheduling optimization system are to take into account the local constraints of each power plant in the valley 40 in order to produce the electrical power that best corresponds to the daily electricity consumption demand in that valley.

[0056] In this context, a valley represents all the power plants located on the same watercourse, with their respective upstream and downstream positions. Figure 4 illustrates, as an example, valley 40 of the Ain, which includes seven power plants, referenced 411 to 417, symbolized by two concentric circles.

[0057] A modeled power plant corresponds to a physical production site. It can encompass various types of elements involved in its electricity production. In the example of [Fig. 4], three elements are modeled: (a) a water reservoir (symbolized by a triangle, shaded to varying degrees depending on the volume of water contained in the reservoir, referenced 421 to 426), (b) a pumping station (symbolized by a disk, referenced 431 and 432), and (c) a production facility (i.e., a hydroelectric plant, symbolized by two concentric circles, referenced 411 to 417). Each element is associated with a unique identifier. For example, in [Fig. 4], the Vouglans hydroelectric power station is modeled by three agents: VOUGLR representing a water reservoir 421, VOUGLP representing a pumping station 431, and VOUGLH representing the power generation plant 411. Thus, an agent can be defined as the modeling of a Equipment within the power generation park is connected to other equipment within the park by a physical, functional, and / or causal link. For example, the level of electrical power that can be delivered from the VOUGLH plant is directly dependent on the water level in the VOUGLR reservoir. In [Fig. 4], the direction of water flow in the valley is indicated by arrows.

[0058] The business knowledge on the operation of a hydroelectric power plant has been conceptually modeled with an ontology, illustrated in [Fig.5], and the individuals of each concept are stored in the form of the knowledge graph.

[0059] These knowledge representation elements are used as resource generators for the multi-agent simulation system of [Fig. 1]. For example, a class "Production Facility", referenced 51, is defined as a subcategory of the class "Physical Object", referenced 50, which is a general category encompassing the common characteristics between production units and water storage systems (e.g., lakes, reservoirs), called agents "Reservoir", referenced 52.

[0060] The local constraints are specifically modeled at the level of the Production Facility 51, which can be, for example, a hydroelectric production facility 511 or a nuclear power plant 512. With regard to the "Reservoir" agents 52, the local constraints are primarily related to their water volume. They are defined, for example, according to the following operating parameters: - minimum water volume 520; - maximum water volume 521; - water volume 522; - rating 523.

[0061] The values ​​of these different parameters are provided by the sensors present in or near the reservoir, and stored in the database of the system of [Fig.1]: they are then used, in operation, to simulate the behavior of the "reservoir" agent 52.

[0062] The "Physical Object" 50 is, for example, associated with operating parameters relating to an incoming water flow rate 501 and an outgoing water flow rate 502, as well as a possible input 503 from a pumping station or naturally via a nearby river or its tributaries. The values ​​of these parameters can be updated in the database of the formal representation KR 10 at a predetermined time interval, for example every five or ten minutes, from flow meter-type sensors located upstream and downstream of the physical object 50.

[0063] In the example of the production park in [Fig. 4], whose model focuses on For hydroelectric power plants, the following constraints are taken into account at the local level: - factory activation times, - the production power adjustment times (transition time to the upper operating point 514 and transition time to the lower operating point 515), - production capacities (“is available” 513), - the required water flow rates 517, - initial water volumes 522, - energy production required 516.

[0064] At the regional scale, more specifically in the context of Valley 40 [Fig. 4], the constraints revolve primarily around the relative geographical position of each power plant in relation to the others in the valley, whether upstream or downstream. These spatial relationships exert a direct influence on input and output restrictions, ultimately determining the operational capacity and the sequence of collaboration or independence of each power plant.

[0065] At the global level, the multi-agent system contains an agent which plays a specific role, called the "Central Programming Entity", referenced 53. This entity plays a crucial role at two levels: as input and output of the simulation.

[0066] In one embodiment, this central programming entity 53 defines the production time slots and the desired power levels for each of these time slots for each power plant on day J+1. This constitutes an essential constraint in the entire model, managing the collective demand shared by all the power plants in a specific valley. This scheduling program can serve as the input setpoint for the optimization system in [Fig. 1]. It defines, according to a predetermined time step of five minutes, for example, the expected electrical power output level of each generating installation. It can, in particular, be established based on a forecast of electricity consumption in the valley in [Fig. 4] during day J+1.

[0067] The Central Programming Entity 53 is also the agent that generates, as the final output of the optimization system in [Fig. 1], the optimized electricity production program, including a production setpoint for each associated generating plant, adjusted during the simulation to maximize the satisfaction of the specified overall demand. This adjustment takes into account the local constraints of each power plant. Thus, the initial call-up program of each central programming entity 53 is modified according to the power and water flow rate at the outlet of each associated power plant. This generates an optimized call-up program, adapted to the actual capacity of the electricity generating plants.

[0068] This call program can be transmitted to the different production units, via a wired or wireless communication network, for example a fiber optic data transmission network, or a 4G, 5G or higher type radio communication network.

[0069] It should be noted that one or more central programming entities 53 can be provided per valley, each central programming entity determining the production instructions for one or more turbines associated with it.

[0070] Figure 6 illustrates a model, within the SMA 23 system, of the behavior of a "Production Plant" agent 51. It all begins with the "wake-up" or reactivation state 60 of the "Production Plant" agent 51, where the agent starts electricity production 61, communicates information about its output to the upstream and downstream "physical objects" 50, and then waits during the operating point rise period. During the "production phase" 61, the agent regularly records electricity production and the output used. When there is a change 62 in the operating point on the call schedule, the agent transmits the output information to neighboring agents and then waits during the transition period after the operating point change. In the event of a plant shutdown 63, the agent informs the "physical objects" 50 of the new outputs and then waits for the minimum shutdown time.Finally, the agent remains inactive for certain periods 64, periodically reactivating to check whether it should resume activity.

[0071] The "reservoir" agents 52 have a simpler behavior. They update the water volume at each time step. They adjust the inlet and outlet flow rates based on received messages and use this data to update the reservoir's water volume. In addition, they make random adjustments to the water inputs to reflect natural fluctuations. Finally, they notify upstream agents if the water level exceeds a high threshold and downstream agents if it falls below a low threshold.

[0072] The "Central Programming Entity" agent 53 sends the call-off programs to the plants that need them, generally once a day, to facilitate the coordination and planning of power production. During the simulation, the call-off program is adjusted according to the actual power levels that each plant can achieve. The set of power levels for each hour then constitutes the new complete call-off program for that day.

[0073] It is important to note that the output of each agent is used to update the parameters of each node in the knowledge graph. Furthermore, for each calling program, the system records a log file containing all the information about the power program that can be called. Finally, the evolution of each agent during the simulation is also displayed visually via The platform's user interface 24 displays the plants' status and power output, while the reservoir shows the water volume at the currently simulated time step. In one embodiment, a time step is equivalent to 30 minutes.

[0074] The dynamic interaction and coordination between all these agents are illustrated in [Fig. 7]. Too low or too high a water volume in a reservoir 52 necessitates an adjustment in the electrical power output of a plant 51, and therefore an adjustment in the water flow rate from the reservoir 52. The central programming unit 53 controls and orchestrates all these adjustments. It is thus possible to recreate the complex interactions within the energy system, providing a realistic and detailed simulation of its operation.

[0075] In addition to the updates provided by the SMA 23 system agents during the simulation, the knowledge graph can also play a central role, allowing operational staff to report a change in a power plant component during the simulation by querying the graph. This capability is facilitated by the standardization of communication between the business concept and the system via the knowledge graph and its associated ontology. This enables power plant managers to report the loss of equipment due to failure or regulatory reasons (for example, the shutdown of a nuclear power plant due to excessively high water temperature or insufficient water level upstream of the plant).

[0076] The multi-agent system thus designed constitutes a digital twin of the electricity generation fleet, taking into account the specific operation of each multi-sector generation facility. Furthermore, it is perfectly synchronized with the knowledge graph, enabling it to adapt immediately to any changes in the generation fleet reflected in the knowledge graph. This convergence of the different sources of electricity generation within a unified system represents a remarkable innovation for the energy sector.

[0077] Although described in the context of a specific example of a power generation facility limited to a single hydroelectric valley, it is readily understood that it can be generalized to any type of energy system, and in particular to the European electricity system, in the market sense. This system is inherently dynamic, due in particular to changes in regulations. The proposed technique makes it possible to express this regulatory evolution through a knowledge graph, the data and their values ​​of which are directly provided by all the sensors present within the energy system, and in particular the various electricity meters located on the different power plants, the water flow sensors at the inlet and outlet of reservoirs, pumping stations and hydroelectric plants, the temperature sensors, the anemometers, etc. List of reference signs

[0078] - 10: formal KR representation - 11: ontology; - 12: knowledge graph; - 13: receiving module; - 14: transmission module; - 15: sensor; - 16: production instruction; - 17: updates; - 18: optimized production instruction; - 20: Java code generation tool; - 21: classes; - 22: set of behaviors and actions; - 23: multi-agent simulation system; - 24: man-machine interface; - El: step of generating a formal representation; - E2: step of generating a multi-agent system; - 25: "Main" programmatic object; - 26: programmatic object “ConstructorAgent”; - 27: ClassBuilder programmatic object; - 261: Loading ontologies; - 262: creation of classes; - 263: instantiation of agents; - 40: valley; - 411-417: hydraulic plants; - 421-426: reservoirs; - 431-432: pumping stations; - 50: physical object; - 51: production facility; - 52: reservoir; - 53: central programming entity; - 501-503: operating parameters; - 520-523: operating parameters; - 511: hydraulic plant; - 512: nuclear plant; - 513-517: operating parameters; - 60: wake-up; - 61: production; - 62: change of operating point; - 63: stop; - 64: inactivity phase. List of documents cited Patent documents

[0079] For the avoidance of doubt, the following patent documents are cited: - patcitl: CN110445173A (publication number); - patcit2: US2017262007Al (publication number); and - patcit3: CN113991641A (publication number). Non-patent literature

[0080] For the sake of clarity, the following non-patent elements are cited: - nplcitl: C. Donzelli, SA Kidanu, R. Chbeir, and Y. Cardinale, “Onto2MAS: An ontology-based framework for automatic multi-agent System generation”, in 2016 12th International Conference on Signal-Image Technology & Intemet-Based Systems (SITIS), IEEE, 2016, p. 381-388; - nplcit2: G. Poveda and R. Schumann, “An ontology-driven approach for modeling a multi-agent-based electricity market”, in Multiagent System Technologies: 14th German Conference, MATES 2016, Klagenfurt, Ôsterreich, September 27-30, 2016. Proceedings 14, Springer, 2016, p. 27-40; - nplcit3 : S. Marchenkov, “Automated Code Génération of Multi-Agent Interaction for constructing Semantic Services”, in Proc. 14th Int’l Conf. on Mobile Ubiquitous Computing, Systems, Services and Technologies (UBICOMM). IARIA XPS Press, 2020 ; - nplcit4 : Quynh-Nhu Numi Tran et Graham Low. « MOBMAS : A methodology for ontology based multi-agent Systems development ». In : Information and Software Technology, 50.7-8 (2008), p. 697-722; et - nplcit5:1. Stankov, Y. Yang, B. A. Langellier, J. Purtle, K. L. Nelson, and A. V. Diez Roux, “Dépréssion and alcohol misuse among older adults: Exploring me-chanisms and policy impacts using agent-based modelling”, Social psychiatry and psychiatrie epidemiology, vol. 54, pp. 1243-1253, 2019.

Claims

Demands

1. A method for optimizing the scheduling of electrical power generation by an electrical power generation park comprising a plurality of generation units (411-417; 51), comprising: a. a receipt of a program (516) for the production of electrical energy, including a setpoint (16) for the production of electrical power for said production units for a reference period; b. a simulation, based on said production setpoint (16), of electrical power production by said production units (411-417; 51), implemented by a multi-agent system (23) modeling said fleet, i. said multi-agent system (23) being generated from a formal representation (10) of said production park, constructed from an ontology (11) and a knowledge graph (12) comprising a set of nodes each representing an element of one of said production units (411-417; 51), and a set of edges each representing a functional relationship between two nodes that they connect, ii. said formal representation (10) being updated:

1. according to a determined time step, from representative values ​​of operating parameters of said nodes, received from sensors (15) arranged in said park; 2. on the fly, in the event of a change (17) affecting at least one operating state of one of said nodes; c. a generation, for said reference period, of an optimized electrical power production program, including a production setpoint (18) adjusted according to the simulated electrical power production for said production units and taking into account said operating states and said representative values ​​of the operating parameters updated in said formal representation.

2. Optimization method according to claim 1, characterized in that it also includes sending, to said production units, said adjusted production setpoint (18) of the optimized electrical power production program for said reference period.

3. Optimization method according to any one of claims 1 or 2, characterized in that it also includes a rendering on a human-machine interface (24) of said simulated electrical power production and / or said generated optimized electrical power production program.

4. Optimization method according to any one of claims 1 to 3, characterized in that said park comprises at least two production units (411-417; 51) belonging to the group comprising: - hydroelectric plants; - thermal plants; - nuclear plants; - photovoltaic energy sources; - wind energy sources; - tidal energy sources; - hydraulic pumping stations; - batteries.

5. Optimization method according to any one of claims 1 to 4, characterized in that said nodes of said knowledge graph represent elements belonging to the group comprising: - water reservoirs (421-426; 52); - pumping stations (431-432); - production facilities (411-417; 51); - production generators; - production auxiliaries.

6. Optimization method according to any one of claims 1 to 5, characterized in that said operating parameters belong to the group comprising: - a volume of water stored; - a range of required inlet water flow rate; - an outlet water flow rate; - an electrical power produced; - a range of inlet water temperature; - a required electrical power; - a maximum available power range; - a minimum high plateau duration; - a minimum low plateau duration; - a type of operating point; - a system state; - a fuel type; - a required gas power.

7. Optimization method according to any one of claims 1 to 6, characterized in that said operating states belong to the group comprising: - an active operating state; - a failure state; - a stopped state; - a state of electrical consumption.

8. Optimization method according to any one of claims 1 to 7, characterized in that said electrical power production setpoint (16) takes into account a forecast of electrical consumption.

9. A computer program comprising instructions for carrying out the method according to any one of claims 1 to 8 when this program is executed by a processor.

10. A non-transient, computer-readable recording medium on which a program is recorded for the implementation of the method according to any one of claims 1 to 8 when this program is executed by a processor.

11. System for optimizing the electrical power generation schedule of an electrical power generation facility comprising a plurality of generating units (411-417; 51), comprising: a. a communication module (13-14) configured to: i. receive an electrical power generation schedule, including a setpoint (16) for electrical power generation for said generating units (411-417; 51) for a reference period; ii. send to said generating units (411-417; 51) an adjusted setpoint (18) for the generation of an electrical power generation schedule optimized for said reference period; b. a semantic database configured to store a formal representation (10) of said park, constructed from an ontology (11) and a knowledge graph (12) comprising a set of nodes each representing an element of one of said production units (411-417; 51), and a set of edges each representing a functional relationship between two nodes that they connect; c. a set of sensors (15) configured to provide, according to a determined time step, representative values ​​of operating parameters of said nodes, intended to feed said formal representation (10); d. a human-machine interface (24) configured to allow the updating (17) of said formal representation in the event of a change affecting at least one operating state of one of said nodes; e. a multi-agent system (23) modeling said park, generated from said formal representation (10), and configured to simulate, from said production setpoint (16), an electrical power production by said production units (411-417; 51), said multi-agent system (23) comprising a central programming entity (53) configured to generate, for said reference period, said optimized electrical power production program, comprising said production setpoint (18) adjusted according to the simulated electrical power production for said production units (411-417; 51) and taking into account said operating states and said representative values ​​of the operating parameters updated in said formal representation (10).

Images