Method and system for constructing man-machine hybrid enhanced smart power grid regulation and control meta-task

Abstract

The invention discloses a man-machine hybrid enhanced smart power grid regulation and control meta-task construction method and a system with man-machine hybrid enhanced smart power grid regulation and control meta-task construction, and the man-machine hybrid enhanced smart power grid regulation and control meta-task construction method proposes a meta-task model (TOEEI) and a construction principle (CRDM) suitable for a power grid regulation and control scene. The classification of the meta-tasks is changed into three classes from four classes, and man-machine meta-tasks are changed into human meta-tasks which can be replaced by AI from man-machine cooperation. Meta-task classification is more reasonable, the substitution relation of AI to manual work is defined, and focusing on the target of artificial intelligence method research is facilitated; meanwhile, an execution method of a meta-task in a power grid regulation and control scene is provided, a sub-task solidification principle is defined, and system implementation is simplified; a foundation is laid for practical application of man-machine hybrid intelligence in a power grid regulation and control scene, and the method is an important support for constructing an application system.

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic uncertainty measurement, and in particular to a fusion method and system for dynamic measurement uncertainty evaluation. Background Art

[0002] Based on traditional AI research, the research on human-machine hybrid enhanced intelligence focuses on technical challenges such as multi-machine collaboration, human-in-the-loop machine learning, and human-machine hybrid evolution optimization. In order to solve these problems and apply them to actual power grid control tasks, it is necessary to establish a basic model for human-machine hybrid enhanced intelligent grid control.

[0003] According to the basic theory of hybrid augmented intelligence and existing achievements, the basic model of human-machine hybrid augmented intelligent grid control should include three aspects: environment, behavior, and task. Among them, the environment model is used to describe the operation scenario of hybrid augmented intelligence; the behavior model is used to describe the relevant human behavior; the task model built on the basis of the environment and behavior models mainly models the grid control process and the division of labor between man and machine, which is the key to realizing hybrid augmented intelligence.

[0004] The existing technologies for human-machine hybrid intelligent meta-tasks mainly include the following points:

[0005] The task structure architecture consists of three levels, namely, tasks, subtasks and metatasks, forming a three-level multi-scale task architecture. Tasks generally refer to tasks with large scales and independent control purposes, such as safety and stability verification tasks; subtasks refer to smaller-scale task links contained in tasks, such as flow calculation subtasks and stability calculation subtasks; metatasks are discrete, indivisible basic tasks, and are the basis for completing the tasks at the previous level.

[0006] Metatasks can be divided into human intelligence metatasks, machine intelligence metatasks, machine automation metatasks and human-machine cooperation metatasks. Machine automation metatasks refer to metatasks that use computers to complete data operations such as power flow calculations; human intelligence metatasks include metatasks that are completed by humans, such as task goal specification and decision confirmation; machine intelligence metatasks refer to metatasks that can be completed by machine agents alone, such as mode matching; human-machine cooperation metatasks refer to metatasks that can be completed through the cooperation of human intelligence and machine intelligence, such as generator output adjustment, power grid topology adjustment, load adjustment and other metatasks.

[0007] Based on the multi-scale task relationship library that describes the relationship between tasks at different levels, a task attribute library is constructed to define the specific functions and calling methods of different meta-tasks in detail. The attributes of a single meta-task are stored in the database in the form of a dictionary.

[0008] The defects of the prior art are mainly reflected in the following aspects:

[0009] The paper does not combine the actual situation of power grid regulation to give a reasonable definition of meta-tasks and reasonable principles for their construction; in the classification of meta-task types, there is a deviation in the understanding of machine intelligence and human-machine cooperation. In essence, machine intelligence meta-tasks are a substitute for meta-tasks that can be completed by humans. Therefore, machine intelligence meta-tasks and human-machine cooperation meta-tasks should be a type of meta-task; the execution principles and process of meta-tasks are not given in detail. Summary of the invention

[0010] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a method for constructing meta-tasks of human-machine hybrid enhanced smart grid control, and proposes a meta-task model (TOEEI) and construction principle (CRDM) suitable for grid control scenarios, which lays a foundation for practical application; the classification of meta-tasks is simplified from four categories to three categories, among which human-machine meta-tasks are changed from human-machine collaboration to human meta-tasks that can be replaced by AI. The classification of meta-tasks is more reasonable, and the replacement relationship of AI to manual work is clarified, which is conducive to focusing on the goal of artificial intelligence method research; the execution method of meta-tasks in grid control scenarios is proposed, and the principle of sub-task solidification is clarified, which is conducive to simplifying system implementation.

[0011] The present invention also proposes a system having the above-mentioned method for constructing human-machine hybrid enhanced smart grid control meta-tasks.

[0012] According to the first aspect of the present invention, the method for constructing a human-machine hybrid enhanced smart grid control meta-task is characterized by comprising:

[0013] Based on the task execution subject, the types of meta-tasks are divided into human meta-tasks, machine meta-tasks and human-machine meta-tasks;

[0014] Based on the current level of AI technology, the principles that need to be followed in constructing the human-machine meta-task;

[0015] Based on the types of meta-tasks and the principles that human-machine meta-tasks need to follow, a meta-task model is constructed, wherein the meta-tasks can be divided into five tuples: Type, Objective, Executor, Environment, and IO;

[0016] During the execution of the meta-task, the model parameters are modified according to the requirements so that it can achieve the task objectives;

[0017] Based on the requirements during the execution of the meta-task, an environment model and a behavior model are defined so that the environment model can obtain the parameter data description required for the execution of the meta-task, and different types of meta-tasks can be supported by the same environment model;

[0018] Among them, the execution of the meta-task also needs to follow the principles of AI availability, AI trustworthiness, AI active interaction, and sub-task solidification.

[0019] The method for constructing a human-machine hybrid enhanced smart grid control meta-task according to an embodiment of the present invention has at least the following beneficial effects:

[0020] This application proposes a meta-task model (TOEEI) and construction principle (CRDM) suitable for power grid control scenarios, and changes the classification of meta-tasks from four categories to three categories, among which human-machine meta-tasks are changed from human-machine collaboration to human meta-tasks that can be replaced by AI. The classification of meta-tasks is more reasonable, and the replacement relationship of AI to manual work is clarified, which is conducive to focusing on the goal of artificial intelligence method research; at the same time, the execution method of meta-tasks in power grid control scenarios is proposed, and the principle of sub-task solidification is clarified, which is conducive to simplifying system implementation; it lays the foundation for the actual application of human-machine hybrid intelligence in power grid control scenarios, and is an important support for building application systems.

[0021] According to some embodiments of the present invention, the human meta-task is a meta-task that can only be completed by humans; the machine meta-task is a meta-task that is completely completed by machines, such as power flow calculation and solution; the human-machine meta-task is a meta-task that is completed by humans in the current large power grid control process, but can be completed by machines instead of humans in human-machine hybrid enhanced intelligent control.

[0022] According to some embodiments of the present invention, the principles that the human-machine meta-task needs to follow specifically include:

[0023] AI capability principle: human-machine meta-task is the most complex task unit that AI can complete independently. Therefore, AI capability determines the "upper bound" of meta-task division, that is, it is impossible to construct more complex human-machine meta-tasks that exceed AI capability.

[0024] Reuse principle: Human-machine meta-tasks should be reusable in task model construction and actual application. That is, the granularity of meta-tasks should be fine enough to be reused in other tasks. The reuse principle determines the "lower bound" of meta-task division. On the premise of satisfying reuse, finer-grained human-machine meta-tasks are not needed.

[0025] Model definability principle: the determination of human-machine meta-tasks should have clear model definitions, such as input, output, objective function, execution method, constraints, etc., so that they can be completed by AI;

[0026] According to the principle of management recognition, the introduction of human-machine hybrid augmented intelligence does not change the existing grid control workflow. The definition of meta-tasks should be based on the existing workflow and cannot violate the requirements of the management system.

[0027] According to some embodiments of the present invention, the human meta-task is a meta-task that must be completed by a human, including the following situations:

[0028] Start a task, triggering one or a series of meta-tasks;

[0029] Judge and make decisions based on output data from automation systems, simulation programs, and AI.

[0030] According to some embodiments of the present invention, the machine meta-task is implemented by a program and needs to follow the software construction "algorithm + data" model, including the following situations:

[0031] The mathematical algorithms that machines rely on are available and have better results and higher efficiency than human processing and AI methods;

[0032] The data involved in machine meta-tasks should be processable by mathematical algorithms and are usually structured data.

[0033] According to some embodiments of the present invention, the human-machine meta-task includes two categories: data analysis and measure formulation. The data analysis focuses on discovering associations and extracting knowledge from data, and usually involves AI algorithms related to knowledge discovery; the measure formulation focuses on deriving countermeasures, and is usually related to technologies such as reinforcement learning and expert systems; the human-machine meta-task includes the following situations:

[0034] The objective function solved by AI and its constraints are derived from the human-machine meta-task model;

[0035] The execution of the human-machine meta-task in power grid regulation involves human-machine knowledge interaction;

[0036] Human-machine meta-tasks whose objective functions can be solved mathematically can correspond to automation procedures.

[0037] According to some embodiments of the present invention, the environmental model contains a unified and standardized definition of meta-task input and output data, including a structured general knowledge base, an AI model library, data files, and a database; the environmental model can obtain the parameter data description required for the execution of the meta-task, and different types of meta-tasks can be supported by the same environmental model.

[0038] According to some embodiments of the present invention, the relationship between the meta-task and the behavior model includes:

[0039] Human behavior is within the framework of human meta-tasks and human-machine meta-tasks performed by humans. In theory, mathematical models related to human behavior in meta-task execution can be derived from these two meta-task models. The form can be similar to the problem description solved by the aforementioned AI model, including the objective function, operating parameters and their range of variation. Human behavior can be understood as solving the objective function within the scope of the mathematical model description, which helps to quantify and understand human behavior;

[0040] The behavior of the machine is within the framework of the machine's meta-task and the human-machine meta-task executed by the machine. Similar to human behavior, the mathematical model related to the machine's behavior in the meta-task execution can also be derived from these two meta-task models, including the objective function, variable parameters, etc. The behavior of the machine can also be understood as solving the objective function within the scope of the mathematical model description, which is also helpful for the quantitative evaluation and understanding of the machine's behavior.

[0041] For human-machine meta-tasks, since they can be performed by machines or humans, and the behavioral models of humans and machines are related to the same meta-task model, the behaviors of the two are comparable, which also lays the foundation for interaction.

[0042] According to some embodiments of the present invention, the execution process of the meta-task includes two modes, namely:

[0043] Human-computer interaction based on human thinking is usually an operation where humans initiate the execution of meta-tasks and give instructions to machines;

[0044] Based on human-computer interaction, people's thinking is guided, which corresponds to people's judgment of the machine's conclusions and the subsequent analysis process; if the task is not completed, human-computer interaction will usually be carried out again.

[0045] According to the second aspect of the present invention, the human-machine hybrid enhanced smart grid control meta-task construction system has the following features, including:

[0046] A category classification module is used to classify the types of meta-tasks based on the task execution subject, including human meta-tasks, machine meta-tasks and human-machine meta-tasks;

[0047] A principle formulation module, used to construct the principles that the human-machine meta-task needs to follow based on the current AI technology level;

[0048] A model building module, used to build a meta-task model based on the types of the meta-tasks and the principles that the human-machine meta-tasks need to follow, wherein the meta-tasks can be divided into five tuples: Type, Objective, Executor, Environment, and IO;

[0049] The parameter adjustment module is used to modify the parameters of the model according to the requirements during the execution of the meta-task so that it can achieve the task objectives;

[0050] A model interaction module is used to define an environment model and a behavior model based on the requirements during the execution of the meta-task, so that the environment model can obtain the parameter data description required for the execution of the meta-task, and different types of meta-tasks can be supported by the same environment model;

[0051] Among them, the execution of the meta-task also needs to follow the principles of AI availability, AI trustworthiness, AI active interaction, and sub-task solidification.

[0052] According to some embodiments of the present invention, according to some embodiments of the present invention, the human meta-task is a meta-task that can only be completed by humans; the machine meta-task is a meta-task that is completely completed by machines, such as power flow calculation and solution; the human-machine meta-task is a meta-task that is completed by humans in the current large power grid control process, but can be completed by machines instead of humans in human-machine hybrid enhanced intelligent control.

[0053] According to some embodiments of the present invention, the principles that the human-machine meta-task needs to follow specifically include:

[0054] AI capability principle: human-machine meta-task is the most complex task unit that AI can complete independently. Therefore, AI capability determines the "upper bound" of meta-task division, that is, it is impossible to construct more complex human-machine meta-tasks that exceed AI capability.

[0055] Reuse principle: Human-machine meta-tasks should be reusable in task model construction and actual application. That is, the granularity of meta-tasks should be fine enough to be reused in other tasks. The reuse principle determines the "lower bound" of meta-task division. On the premise of satisfying reuse, finer-grained human-machine meta-tasks are not needed.

[0056] Model definability principle: the determination of human-machine meta-tasks should have clear model definitions, such as input, output, objective function, execution method, constraints, etc., so that they can be completed by AI;

[0057] According to the principle of management recognition, the introduction of human-machine hybrid augmented intelligence does not change the existing grid control workflow. The definition of meta-tasks should be based on the existing workflow and cannot violate the requirements of the management system.

[0058] According to some embodiments of the present invention, the human meta-task is a meta-task that must be completed by a human, including the following situations:

[0059] Start a task, triggering one or a series of meta-tasks;

[0060] Judge and make decisions based on output data from automation systems, simulation programs, and AI.

[0061] According to some embodiments of the present invention, the machine meta-task is implemented by a program and needs to follow the software construction "algorithm + data" model, including the following situations:

[0062] The mathematical algorithms that machines rely on are available and have better results and higher efficiency than human processing and AI methods;

[0063] The data involved in machine meta-tasks should be processable by mathematical algorithms and are usually structured data.

[0064] According to some embodiments of the present invention, the human-machine meta-task includes two categories: data analysis and measure formulation. The data analysis focuses on discovering associations and extracting knowledge from data, and usually involves AI algorithms related to knowledge discovery; the measure formulation focuses on deriving countermeasures, and is usually related to technologies such as reinforcement learning and expert systems; the human-machine meta-task includes the following situations:

[0065] The objective function solved by AI and its constraints are derived from the human-machine meta-task model;

[0066] The execution of the human-machine meta-task in power grid regulation involves human-machine knowledge interaction;

[0067] Human-machine meta-tasks whose objective functions can be solved mathematically can correspond to automation procedures.

[0068] According to some embodiments of the present invention, the environmental model contains a unified and standardized definition of meta-task input and output data, including a structured general knowledge base, an AI model library, data files, and a database; the environmental model can obtain the parameter data description required for the execution of the meta-task, and different types of meta-tasks can be supported by the same environmental model.

[0069] According to some embodiments of the present invention, the relationship between the meta-task and the behavior model includes:

[0070] Human behavior is within the framework of human meta-tasks and human-machine meta-tasks performed by humans. In theory, mathematical models related to human behavior in meta-task execution can be derived from these two meta-task models. The form can be similar to the problem description solved by the aforementioned AI model, including the objective function, operating parameters and their range of variation. Human behavior can be understood as solving the objective function within the scope of the mathematical model description, which helps to quantify and understand human behavior;

[0071] The behavior of the machine is within the framework of the machine's meta-task and the human-machine meta-task executed by the machine. Similar to human behavior, the mathematical model related to the machine's behavior in the meta-task execution can also be derived from these two meta-task models, including the objective function, variable parameters, etc. The behavior of the machine can also be understood as solving the objective function within the scope of the mathematical model description, which is also helpful for the quantitative evaluation and understanding of the machine's behavior.

[0072] For human-machine meta-tasks, since they can be performed by machines or humans, and the behavioral models of humans and machines are related to the same meta-task model, the behaviors of the two are comparable, which also lays the foundation for interaction.

[0073] According to some embodiments of the present invention, the execution process of the meta-task includes two modes, namely:

[0074] Human-computer interaction based on human thinking is usually an operation where humans initiate the execution of meta-tasks and give instructions to machines;

[0075] Based on human-computer interaction, people's thinking is guided, which corresponds to people's judgment of the machine's conclusions and the subsequent analysis process; if the task is not completed, human-computer interaction will usually be carried out again.

[0076] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0078] Figure 1 A schematic diagram of the relationship between tasks, subtasks, metatasks and agents in the method for constructing metatasks of human-machine hybrid enhanced smart grid control in an embodiment of the present invention;

[0079] Figure 2 A schematic diagram of the relationship between meta-tasks and behavior models in the method for constructing meta-tasks of human-machine hybrid enhanced smart grid control in an embodiment of the present invention;

[0080] Figure 3 Schematic diagram of the basic execution process of the meta-task in the method for constructing the meta-task of human-machine hybrid enhanced smart grid control in an embodiment of the present invention

[0081] Figure 4 for Figure 1 The structural block diagram of the system built by human-machine hybrid enhanced smart grid control meta-task is shown. DETAILED DESCRIPTION

[0082] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0083] In the description of the present invention, it should be understood that descriptions involving orientations, such as up, down, front, back, left, right, etc., and orientations or positional relationships indicated are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.

[0084] In the description of the present invention, "several" means one or more, "multiple" means more than two, "greater than", "less than", "exceeding", etc. are understood not to include the base number, and "above", "below", "within", etc. are understood to include the base number. If there is a description of "first" and "second", it is only for the purpose of distinguishing technical features and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features or implicitly indicating the sequence relationship of the indicated technical features.

[0085] In the description of the present invention, unless otherwise clearly defined, words such as "set", "installed", "connected", etc. should be understood in a broad sense, and those skilled in the art can reasonably determine the specific meanings of the above words in the present invention in combination with the specific content of the technical solution.

[0086] Embodiment 1

[0087] The embodiment of the present invention provides a method for constructing a meta-task of human-machine hybrid enhanced intelligent power grid regulation, which specifically includes the following structures.

[0088] I. Principles for establishing meta-tasks

[0089] Generally, in the problem of human-machine hybrid enhanced intelligent power grid regulation, the execution subjects of tasks involve humans and machines. Humans mainly refer to the personnel participating in the analysis of power grid operation modes and regulation operations; machines include both existing dispatching automation systems and power grid simulation calculations (referred to as "automation systems" for short), and also include AI for reasoning and analysis. In hybrid enhanced intelligence, the roles, task assignments of humans and machines, and the interaction between the two are all based on meta-tasks.

[0090] Generally, humans have strong reasoning abilities. The introduction of AI is mainly used to replace some of the analysis and reasoning work of humans, and the introduction of AI should not change the existing power grid regulation work process. Therefore, the present invention classifies meta-tasks into the following three types: (1) Human meta-tasks: Meta-tasks that can only be completed by humans, such as final result confirmation and final decision-making; (2) Machine meta-tasks: Meta-tasks that are completely completed by machines, such as power flow calculation and solution, etc.; (3) Human-machine meta-tasks: Meta-tasks that are completed by humans in the current large power grid regulation process, but can be replaced by machines (including AI or automation programs) in human-machine hybrid enhanced intelligent regulation. Among them, the first type of meta-tasks mainly refers to the decision-making link in the current regulation work process and the human recognition link after the execution of the AI model, etc.; the second type of meta-tasks mainly includes dispatching automation systems, simulation calculation systems, and automation links constructed to cooperate with AI; the third type of meta-tasks mainly involves the interaction, collaboration, reasoning, and optimization evolution between humans and AI, which is the focus of the present invention.

[0091] Based on the current AI technology level, this patent proposes the CRDM principle for establishing human-machine meta-tasks:

[0092] 1) AI capacity principle. Human-machine meta-tasks are the most complex task units that AI can complete independently. Therefore, AI capacity determines the "upper bound" of meta-task division, that is, more complex human-machine meta-tasks that exceed AI capacity cannot be constructed;

[0093] 2) Reusability. Human-machine meta-tasks should be reusable in task model construction and actual application, that is, the granularity of meta-tasks is fine enough to be reused in other tasks. The reusability principle determines the "lower bound" of meta-task division. On the premise of satisfying reusability, finer-grained human-machine meta-tasks are not needed.

[0094] 3) Model definability principle. The determination of human-machine meta-tasks should have clear model definitions, such as input, output, objective function, execution method, constraints, etc., so that they can be completed by AI;

[0095] 4) Management recognition principle (management). The introduction of human-machine hybrid augmented intelligence does not change the existing grid control workflow. The definition of meta-tasks should be based on the existing workflow and cannot violate the requirements of the management system.

[0096] For human meta-tasks, since they need to cooperate with human-machine meta-tasks, the division scale should match that of human-machine meta-tasks. For machine meta-tasks, they need to follow principles similar to 1)-3) of the CRDM principles for human-machine meta-tasks, that is, they should not exceed the capabilities of the automation system and simulation calculations, be reusable, and have clear meta-task objective functions, inputs, outputs, and other conditions.

[0097] The determination of meta-tasks is related to the task objectives, and can be decomposed downward according to the major categories and main processes of power grid control tasks to form sub-tasks with finer granularity, and finally form meta-tasks with the finest granularity.

[0098] 2. Model and Characteristics of Meta-Tasks

[0099] 1. Basic model of meta-task.

[0100] Grid control has strict rules and regulations and clear standards and specifications. Each work task has a relatively clear process. This patent refers to work tasks and their processes as "tasks". Metatasks are the basic units that constitute tasks, and can be considered as the basic work links divided from the perspective of hybrid intelligence. From the perspective of application system construction, metatasks cannot be executed spontaneously, and subtasks need to be established on top of metatasks to organize functions and data.

[0101] Subtasks consist of a set of ordered meta-tasks. They are sub-workflows that can be completed by human-machine cooperation with clear phased results, can be used as system function models, and can support reuse. For example, flow adjustment convergence, section power adjustment, stability property judgment, etc. In theory, more levels of workflow division can be formed on top of subtasks, but since the scope of work of power grid regulation is limited, there will not be many sub-workflows. In addition, considering the complexity of application system construction and data exchange between subtasks and metatasks, an independent power grid regulation task should only correspond to a combination of 1 layer of subtasks.

[0102] For the same meta-task, there may be different AI models to choose from, corresponding to different intelligent agents. Of course, it is also possible to use intelligent agents to carry out meta-tasks performed by all machines (including AI), but from the perspective of efficiency, it may be more efficient to use program-fixed methods for machine meta-tasks.

[0103] Figure 1 The relationship between tasks, subtasks, metatasks, and agents is shown. Among them, subtask 1 is reused in task 1 (workflow 1) and task 2 (workflow 2); metatask 5 is a human-machine metatask executed by AI, which can be executed by an agent loading an AI model, or by two or more agents, and each agent can load the same AI model or different AI models. The former is usually used to achieve parallel data processing, and the latter is usually used to select the better one among the results of multiple AI models; subtasks can be composed of a series of metatasks, but usually not just human-machine metatasks executed by AI.

[0104] The present invention defines a meta-task as the following 5-tuple:

[0105] TOEEI={Type,Objective,Executor,Environment,IO}

[0106] in:

[0107] 1) Type: meta-task type, including three types: human meta-task (Type = Human), machine meta-task (Type = Machine) and human-machine meta-task (Type = Hybrid);

[0108] 2) Objective: The objective of the meta-task. The objectives of machine meta-tasks and human-machine meta-tasks need to be mathematically described to provide an optimized objective function for the construction of the corresponding algorithm; for human meta-tasks, this attribute is a descriptive field of human responsibilities and behaviors;

[0109] 3) Executor: The execution method of the meta-task. This attribute specifically refers to the program or AI model that needs to be called by the machine meta-task or the human-machine meta-task executed by AI, and also gives the default configuration parameters; since human meta-tasks or human-machine meta-tasks that need to be executed by humans require the cooperation of machine meta-tasks, that is, the program receives human input or gives output to humans, so in both cases this field gives the default input and default output of the program;

[0110] 4) Environment: The operating environment of the meta-task, including requirements for the software environment (such as the version of the simulation program, the version of the operating system, etc.) and the hardware environment;

[0111] 5) IO: The input and output interface of the meta-task, including the data structure, storage format, file location or database link and naming of the input and output data.

[0112] 2. Characteristics of meta-tasks.

[0113] It includes human meta-tasks, machine meta-tasks and human-machine meta-tasks. Among them:

[0114] 2.1 Human Meta-Task

[0115] According to the current grid control process and the characteristics of human-machine hybrid intelligence, the meta-tasks that must be completed by humans mainly include two situations:

[0116] 1) Start a task, triggering one or a series of meta-tasks;

[0117] 2) Judge and make decisions based on the output data of automation systems, simulation calculation programs, and AI.

[0118] Both situations involve "decision-making" power, which requires support from people's advanced thinking abilities and management responsibilities.

[0119] 2.2 Machine Meta-Task

[0120] Machine meta-tasks are implemented by programs and need to follow the software construction "algorithm + data" model, so the following requirements are required:

[0121] 1) The mathematical algorithms that machines rely on are available and have better results and higher efficiency than human processing and AI methods;

[0122] 2) The data involved in the machine meta-task should be processable by mathematical algorithms and is usually structured data.

[0123] Machine meta-tasks include the functions provided by existing scheduling automation systems and simulation computing programs, as well as hybrid intelligent human-computer interaction, human feedback information collection to support human-in-the-loop machine learning and hybrid intelligent evolution, and AI execution information recording.

[0124] 2.3 Human-machine meta-task

[0125] In power grid control applications, human-machine meta-tasks can be divided into two categories: data analysis and measure formulation. The former focuses on discovering associations and extracting knowledge from data, usually involving AI algorithms related to knowledge discovery; the latter focuses on deriving countermeasures, usually related to technologies such as reinforcement learning and expert systems. When humans and AI perform the same human-machine meta-task, it can be regarded as a process of solving the task objective function. The constraints, variable definitions, etc. should be similar, so they can be described by similar mathematical models.

[0126] Based on the above human-machine meta-task model, the following conditions are met:

[0127] The objective function solved by AI and its constraints are derived from the human-machine meta-task model;

[0128] The execution of the human-machine meta-task of power grid control involves human-machine knowledge interaction, but the human-machine meta-task itself is only responsible for the generation and reception of interactive data. The human-machine interaction process is completed through data visualization, reports, etc., and is completed by the machine meta-task that cooperates with the human-machine meta-task and is specifically responsible for the interaction;

[0129] Human-machine meta-tasks whose objective functions can be solved by mathematical methods can correspond to automated programs. If AI methods are used, the objective functions of the two are the same, and the only difference is the solution algorithm used. For example, the optimal power flow calculation can use the interior point method or the evolutionary algorithm. At this time, it is necessary to compare the cost and efficiency of the two to determine which algorithm to use.

[0130] 3. Issues Related to the Meta-Task Model

[0131] 1. Parameter modification during the meta-task execution process.

[0132] When executing a meta-task, the process of modifying parameters in order to achieve the task goal is not a meta-task, but only the execution process of the meta-task. For example, the process of a person or AI model adjusting a power flow section and making the section power reach certain requirements constitutes a meta-task. Modifying the capacitor parameters to complete this meta-task is the execution process of the meta-task. The definition of a meta-task can be converted into mathematical model elements such as objective functions and constraints, and the behavior of people and AI can be regarded as the process of adjusting the adjustable variables to approach the target within the parameter definition range. Therefore, the modification of parameters does not constitute a meta-task.

[0133] 2. The connection between the meta-task model, the environment model and the behavior model.

[0134] In terms of the environmental model, the definition of the environmental model is to support the execution of metatasks, so it should be formed by elements such as the input, output and environmental requirements of the metatask. Therefore, a unified description of the environmental model can be established for different control metatasks, avoiding the description of a large amount of scheduling information that is not related to the business. In terms of form, the relationship between the environmental model and the metatask model is similar to the relationship between the AC line physical parameter model and the AC line model of the power flow equation, that is, the parameter data description required for the execution of the metatask can be obtained from the environmental model, and different types of metatasks can be supported by the same environmental model. In this sense, the environmental model contains a unified and standardized definition of the input and output data of the metatask, including a structured general knowledge base, an AI model library, data files, databases, etc. The input of each metatask comes from the environment and the output goes to the environment.

[0135] In terms of behavioral models, in human-machine hybrid enhanced intelligent control, the meaningful behaviors of humans and machines are reflected in the meta-tasks, and the behaviors of humans and machines that are not related to the meta-tasks are not considered in the behavioral model. Figure 2 As shown, for meaningful behaviors, from the perspective of meta-tasks, there are:

[0136] 1) Human behavior is within the framework of human meta-tasks and human-machine meta-tasks performed by humans. In theory, mathematical models related to human behavior in meta-task execution can be derived from these two meta-task models. The form can be similar to the problem description solved by the aforementioned AI model, including the objective function, operating parameters and their range of variation. Human behavior can be understood as solving the objective function within the scope of the mathematical model description, which helps to quantify and understand human behavior;

[0137] 2) The behavior of the machine is within the framework of the machine's meta-task and the human-machine meta-task executed by the machine. Similar to human behavior, the mathematical model related to the machine's behavior in the meta-task execution can also be derived from these two meta-task models, including the objective function, variable parameters, etc. The behavior of the machine can also be understood as solving the objective function within the scope of the mathematical model description, which also helps to quantitatively evaluate and understand the machine's behavior;

[0138] 3) For human-machine meta-tasks, since they can be performed by machines or humans, and the behavioral models of humans and machines are related to the same meta-task model, the behaviors of the two are comparable, which also lays the foundation for interaction.

[0139] 3. Meta-task model and human-computer interaction.

[0140] From the perspective of meta-task execution, any human-computer interaction that is not related to meta-task execution is not a meaningful interaction. The data involved in meaningful human-computer interaction must be within the scope described by the objective function and related mathematical model solved by the person or machine derived from the meta-task model, and it must be helpful in solving the problem. It should be noted that when a person or machine is executing a meta-task, it may switch to another meta-task for some reason. For example, during the flow adjustment process, a person observes that the parameters are incomplete and temporarily switches to the corresponding meta-task to supplement the parameters. At this point, the interaction between people and machines may be unrelated to the original meta-task, but it is related to the new meta-task and is still meaningful.

[0141] Based on meta-tasks and their related mathematical models, human behavior can be transformed into a solution strategy for the objective function that is easy for AI to learn. Similarly, the output of AI can be transformed into a standardized data form that is easy for humans to understand, or further transformed into text or formula expressions. These two transformations will help realize human-computer knowledge interaction.

[0142] 4. Execution of Meta-Tasks

[0143] 1. Principles of Meta-task Execution

[0144] Each specific task can be considered as an instance of a class of tasks. Similarly, a group of meta-tasks corresponding to a task also constitutes an instance of a meta-task after being combined with data and work content. The execution process of a meta-task is the generation and processing process of a meta-task instance. The main principles that need to be followed in the execution of a meta-task are:

[0145] 1) AI is available. The automated systems and computing programs required for machines to perform meta-tasks usually need to pass testing. Similarly, the trained AI must also be tested and meet the access requirements before it can be used in the execution process. If the AI ​​corresponding to a human-machine meta-task fails to pass the test, the meta-task will be performed by humans;

[0146] 2) AI is trustworthy. In view of the possibility that AI may make mistakes during execution, an evaluation link for AI is established, including simulation and physical meaning interpretation. The process of verifying AI conclusions based on the simulation link belongs to the machine meta-task, which provides credibility guarantee for AI conclusions. The physical meaning interpretation of AI conclusions belongs to the human-machine meta-task, which provides support for the acceptance of AI conclusions by humans. Whether the evaluation link is passed, that is, whether the AI ​​results are adopted is decided by humans, which belongs to the human meta-task. If AI fails repeatedly, the corresponding human-machine meta-task will be taken over by humans, and the retraining and testing process of AI will be initiated;

[0147] 3) AI can actively interact. AI can evaluate its own capabilities and propose the need to interact with people or knowledge bases;

[0148] 4) Subtask solidification. Since the content, process and management requirements of power grid regulation are relatively clear and limited in number, the subtask process composed of meta-tasks is mainly solidified by program to simplify the structure of the hybrid intelligent system. On the basis of subtask solidification, the task process of various common tasks can be further solidified. In fact, due to the complex data associations between subtasks, it is sometimes difficult to flexibly build subtasks and tasks in a customized way, and it can only be used as an auxiliary function from the perspective of engineering application.

[0149] In addition, the monitoring and data collection of human and machine behaviors during the execution of meta-tasks should not affect the execution of meta-tasks. Feedback obtained from monitoring results, such as human intervention, should be separated into meta-tasks and should not be embedded in the monitored meta-tasks, that is, the singleness and integrity of meta-tasks should be maintained.

[0150] For scenarios where multiple task flows are intertwined, the task graph formed by subtasks and their contained metatasks can be optimized. However, this optimization is not a change or reconstruction of the workflow, but rather the reuse of the subtasks or metatask related information that can be performed by a solidified program. For example, Figure 1 In the example, if two tasks appear at the same time and have the same environment data, then subtask 1 may be reused. In this case, subtask 1 only needs to be executed once and the results can be shared without changing the structure of the two task processes.

[0151] 2. Meta-task execution process

[0152] For human meta-tasks, there are two modes for the execution of meta-tasks:

[0153] 1) Human thinking -> human-computer interaction, which is usually the operation of human initiating the execution of sub-tasks (meta-tasks), where humans issue instructions to machines;

[0154] 2) Human-computer interaction -> human thinking, which corresponds to the human's judgment of the machine's conclusion and the subsequent analysis process. If the task is not completed, it will usually connect to mode 1 to form human-computer feedback.

[0155] For machine meta-tasks, including human-machine meta-tasks performed by AI, the basic process of execution is as follows: Figure 3 .

[0156] in,

[0157] 1) The subtask program prepares the environment for the metatask execution according to the metatask model, including setting the path, requesting computing resources, etc.

[0158] 2) Environmental data includes the public database data required for meta-task execution, the output of other meta-tasks, AI models, knowledge base data, etc. These data are uniformly modeled and managed by the environmental model. During the execution of meta-tasks, data can be continuously read in and output, and data comes from the environment and returns to the environment;

[0159] 3) When running the model, the AI ​​model read from the model library can be loaded and executed in the form of an agent. Multiple agents can be established for the same meta-task at the same time, and a coordination strategy can be adopted to achieve the best results. Agents corresponding to different meta-tasks can interact through standard data structures described by the environment model, such as knowledge bases, interactive data files, etc.

[0160] Embodiment 2

[0161] Another embodiment of the present application provides a human-machine hybrid enhanced smart grid control meta-task construction system, such as Figure 4 As shown, the system 40 includes:

[0162] A category classification module 401 is used to classify the types of meta-tasks based on the task execution subject, including human meta-tasks, machine meta-tasks and human-machine meta-tasks;

[0163] A principle formulation module 402 is used to construct the principles that the human-machine meta-task needs to follow according to the current AI technology level;

[0164] A model building module 403 is used to build a meta-task model based on the type of the meta-task and the principles that the human-machine meta-task needs to follow, wherein the meta-task can be divided into five tuples: Type, Objective, Executor, Environment, and IO;

[0165] The parameter adjustment module 404 is used to modify the parameters of the model according to the requirements during the execution of the meta-task so that the model can achieve the task objectives;

[0166] Model interaction module 405, used to define environment model and behavior model based on the requirements in the meta-task execution process, so that the environment model can obtain parameter data description required for the meta-task execution, and different types of meta-tasks can be supported by the same environment model;

[0167] Among them, the execution of the meta-task also needs to follow the principles of AI availability, AI trustworthiness, AI active interaction, and sub-task solidification.

[0168] The specific implementation process is the same as the method for constructing the human-machine hybrid enhanced smart grid control meta-task described in Example 1, and will not be repeated here.

[0169] The device embodiments described above are only illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0170] It will be appreciated by those skilled in the art that all or some of the steps and systems in the disclosed method above may be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or may be implemented as hardware, or may be implemented as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or a non-transitory medium) and a communication medium (or a temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that may be used to store desired information and may be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0171] The above is a specific description of the preferred implementation of the present application, but the present application is not limited to the above-mentioned implementation mode. Technical personnel familiar with the field can also make various equivalent deformations or substitutions without violating the spirit of the present application. These equivalent deformations or substitutions are all included in the scope defined by the claims of the present application.

Claims

1. A method for constructing a human-machine hybrid enhanced smart grid control meta-task, characterized in that: include: Based on the task execution subject, the types of meta-tasks are divided into human meta-tasks, machine meta-tasks and human-machine meta-tasks; Based on the current level of AI technology, the principles that need to be followed in constructing the human-machine meta-task; Based on the types of meta-tasks and the principles that human-machine meta-tasks need to follow, a meta-task model is constructed, wherein the meta-tasks can be divided into five tuples: Type, Objective, Executor, Environment, and IO; During the execution of the meta-task, the model parameters are modified according to the requirements so that it can achieve the task objectives; Based on the requirements during the execution of the meta-task, an environment model and a behavior model are defined so that the environment model can obtain the parameter data description required for the execution of the meta-task, and different types of meta-tasks can be supported by the same environment model; Among them, the execution of the meta-task also needs to follow the principles of AI availability, AI trustworthiness, AI active interaction, and sub-task solidification.

2. The method according to claim 1, characterized in that The human meta-task is a meta-task that can only be completed by humans; the machine meta-task is a meta-task that is completely completed by machines, such as power flow calculation and solution; the human-machine meta-task is a meta-task that is completed by humans in the current large power grid control process, but can be completed by machines instead of humans in human-machine hybrid enhanced intelligent control.

3. According to the method of claim 1, the principles that the human-machine meta-task needs to follow specifically include: AI capability principle: human-machine meta-task is the most complex task unit that AI can complete independently. Therefore, AI capability determines the "upper bound" of meta-task division, that is, it is impossible to construct more complex human-machine meta-tasks that exceed AI capability. Reuse principle: Human-machine meta-tasks should be reusable in task model construction and actual application. That is, the granularity of meta-tasks should be fine enough to be reused in other tasks. The reuse principle determines the "lower bound" of meta-task division. On the premise of satisfying reuse, finer-grained human-machine meta-tasks are not needed. Model definability principle: the determination of human-machine meta-tasks should have clear model definitions, such as input, output, objective function, execution method, constraints, etc., so that they can be completed by AI; According to the principle of management recognition, the introduction of human-machine hybrid augmented intelligence does not change the existing grid control workflow. The definition of meta-tasks should be based on the existing workflow and cannot violate the requirements of the management system.

4. The method according to claim 2, characterized in that: The human meta-task is a meta-task that must be completed by humans, including the following situations: Start a task, triggering one or a series of meta-tasks; Judge and make decisions based on output data from automation systems, simulation programs, and AI.

5. The method according to claim 2, characterized in that: The machine meta-task is implemented by a program and needs to follow the software construction "algorithm + data" model, including the following situations: The mathematical algorithms that machines rely on are available and have better results and higher efficiency than human processing and AI methods; The data involved in machine meta-tasks should be processable by mathematical algorithms and are usually structured data.

6. The method according to claim 2, characterized in that The human-machine meta-tasks include data analysis and measure formulation. The data analysis focuses on discovering associations and extracting knowledge from data, usually involving AI algorithms related to knowledge discovery; the measure formulation focuses on deriving countermeasures, usually related to technologies such as reinforcement learning and expert systems; the human-machine meta-tasks include the following situations: The objective function solved by AI and its constraints are derived from the human-machine meta-task model; The execution of the human-machine meta-task in power grid regulation involves human-machine knowledge interaction; Human-machine meta-tasks whose objective functions can be solved mathematically can correspond to automation procedures.

7. The method according to claim 1, characterized in that The environmental model contains a unified and standardized definition of meta-task input and output data, including a structured general knowledge base, an AI model library, data files, and a database; the environmental model can obtain the parameter data description required for meta-task execution, and different types of meta-tasks can be supported by the same environmental model.

8. The method according to claim 1, characterized in that: The relationship between the meta-task and the behavior model includes: Human behavior is within the framework of human meta-tasks and human-machine meta-tasks performed by humans. In theory, mathematical models related to human behavior in meta-task execution can be derived from these two meta-task models. The form can be similar to the problem description solved by the aforementioned AI model, including the objective function, operating parameters and their range of variation. Human behavior can be understood as solving the objective function within the scope of the mathematical model description, which helps to quantify and understand human behavior; The behavior of the machine is within the framework of the machine's meta-task and the human-machine meta-task executed by the machine. Similar to human behavior, the mathematical model related to the machine's behavior in the meta-task execution can also be derived from these two meta-task models, including the objective function, variable parameters, etc. The behavior of the machine can also be understood as solving the objective function within the scope of the mathematical model description, which is also helpful for the quantitative evaluation and understanding of the machine's behavior. For human-machine meta-tasks, since they can be performed by machines or humans, and the behavioral models of humans and machines are related to the same meta-task model, the behaviors of the two are comparable, which also lays the foundation for interaction.

9. The method according to claim 1, characterized in that: The execution process of the meta-task includes two modes: Human-computer interaction based on human thinking is usually an operation where humans initiate the execution of meta-tasks and give instructions to machines; Based on human-computer interaction, people's thinking is guided, which corresponds to people's judgment of the machine's conclusions and the subsequent analysis process; if the task is not completed, human-computer interaction will usually be carried out again.

10. A human-machine hybrid enhanced smart grid control meta-task construction system, Its characteristics include: A category classification module is used to classify the types of meta-tasks based on the task execution subject, including human meta-tasks, machine meta-tasks and human-machine meta-tasks; A principle formulation module, used to construct the principles that the human-machine meta-task needs to follow based on the current AI technology level; A model building module, used to build a meta-task model based on the types of the meta-tasks and the principles that the human-machine meta-tasks need to follow, wherein the meta-tasks can be divided into five tuples: Type, Objective, Executor, Environment, and IO; The parameter adjustment module is used to modify the parameters of the model according to the requirements during the execution of the meta-task so that it can achieve the task objectives; A model interaction module is used to define an environment model and a behavior model based on the requirements during the execution of the meta-task, so that the environment model can obtain the parameter data description required for the execution of the meta-task, and different types of meta-tasks can be supported by the same environment model; Among them, the execution of the meta-task also needs to follow the principles of AI availability, AI trustworthiness, AI active interaction, and sub-task solidification.

Images