A method and system for constructing a power grid probability visualization model for machine learning

By building a grid probability visualization model, the difficulties in preprocessing and feature representation of grid data are solved, efficient and intelligent analysis of grid data is realized, and more effective fault diagnosis and optimized configuration are supported.

CN112116305BActive Publication Date: 2024-11-01CHINA ELECTRIC POWER RESEARCH INSTITUTE CO LTD +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202010555908.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-06-17
Publication Date
2024-11-01
Estimated Expiration
2040-06-17

AI Technical Summary

Technical Problem

It is difficult for the prior art to effectively use grid data for deep learning and intelligent analysis, especially in the aspects of fault diagnosis, fault recovery and grid optimization configuration, data preprocessing and feature representation are difficult.

Method used

A method for building a grid probability visual model is proposed. By determining the scale, depth, granularity and level of the grid information probability map, energy flow information is extracted from the physical grid, path information is designed, node and edge information is constructed, and grid probability is constructed based on these information. Visualize the model and correct the model parameters through machine learning methods.

Benefits of technology

It realizes efficient processing and intelligent analysis of power grid data, simplifies data complexity, extracts dynamic and static characteristics of the power grid, supports richer physical significance and flexible learning purposes, and improves the ability to diagnose faults and optimize configurations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112116305B_ABST
    Figure CN112116305B_ABST
Patent Text Reader

Abstract

The present invention provides a method and system for constructing a power grid probability visualization model for machine learning, including: determining the scale, depth, granularity and level of a power grid information probability graph according to the purpose of machine learning; extracting power grid energy flow information from the physical power grid to construct node information based on the scale, depth, granularity and level; designing and extracting power grid energy flow environment and path information from the path information in the physical network relationship to construct edge information; and constructing a power grid probability visualization model based on the node information and edge information. The present invention establishes a power grid information probability graph based on power grid energy flow information and path information, which can accelerate intelligent data processing and machine learning, spy on the supporting efficiency of data and machine learning, and verify the purpose of data and machine learning law discovery.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of power grid feature representation definition and extraction, and specifically relates to a power grid probability visualization model construction method and system for machine learning. Background Art

[0002] The physical meaning of the power grid has always been clear, and the analysis and calculation of the power system is based on the mathematical model of the physical power grid. The composition and structure of the power grid are complex, and its static and dynamic data characteristics are huge. The power system analysis and calculation based on simulation models and various algorithms have large dimensions of calculation data characteristics, and the physical meaning and network relationships they contain are also very complex.

[0003] The monitoring, calculation, simulation and analysis applications and corresponding systems of power grid planning and operation management have generated a large amount of business-related data in long-term industry services. From different perspectives such as RTU, monitoring and scheduling, stability analysis and simulation calculation, a considerable amount of power grid information is generated and mastered, such as power grid configuration and operation data, various monitoring and calculation data, etc. These data have not yet been deeply developed and researched.

[0004] The application of computer science and artificial intelligence in the field hopes to make more effective use of large amounts of these power grid data, learn power grid operation management methods with methods different from traditional simulation analysis and calculation, obtain fault diagnosis and early warning, and intelligent fault recovery solutions, learn to extract rules and knowledge such as power grid optimization configuration and planning, and establish power grid planning and operation plan evaluation methods.

[0005] The practice of machine learning algorithms in the field and the effective application of the intelligent characteristics of machine learning in the field are key to obtaining high-quality data representation and data set establishment methods with field characteristics and features. However, power grid data preprocessing is quite difficult, such as data collection and extraction, as well as data validity and quality assurance. Summary of the invention

[0006] In order to overcome the above-mentioned deficiencies of the prior art, the present invention proposes a method for constructing a power grid probability visualization model for machine learning, the improvement of which is that it includes:

[0007] Determine the scale, depth, granularity and level of the grid information probability map based on machine learning purposes;

[0008] Based on scale, depth, granularity and level, grid energy flow information is extracted from the physical grid to construct node information;

[0009] From the path information in the physical network relationship, the design extracts the grid energy flow environment and path information and constructs the edge information;

[0010] A power grid probabilistic visualization model is constructed based on node information and edge information.

[0011] Preferably, the path information includes:

[0012] The connection relationship and association relationship of energy flow information values ​​of observation points equipped with energy flow information sensors in the power grid in the power grid and the association relationship that is free from the constraints of the power grid environment.

[0013] Preferably, the energy flow information includes: power flow, voltage, current and frequency.

[0014] Preferably, the nodes in the power grid information probability graph include: source points, midpoints, end points and indefinite points;

[0015] Among them, the element points include power generation points, power supply points and power sources;

[0016] The end points include power consumption points and loads;

[0017] The midpoint includes an observation point set up for observing power grid information;

[0018] An indefinite point includes a composite point consisting of at least two of the origin, the middle point and the end point.

[0019] Preferably, the edges in the power grid information probability graph correspond to elements, connections and road networks in the power grid, the edges have directions, and the edges include resistance probabilities;

[0020] The resistance probability represents the influence weight of impedance and loss on the energy flow information flowing through the edge where the resistance probability is located.

[0021] Preferably, the value of the node is obtained by the resistance probability of the edge reaching the node and the value of the starting point of the edge, or is directly extracted from the power grid observation point corresponding to the node when constructing the power grid information probability model.

[0022] Preferably, the scale of the power grid information probability map is set according to the scale of the power grid;

[0023] The depth of the power grid information probability map is set according to research requirements, and the depth includes a primary network map and a secondary network map;

[0024] The granularity of the power grid information probability graph is set according to the learning and research purpose, and the granularity includes the number of selected nodes;

[0025] The levels of the power grid information probability map are set according to the power grid information, and the levels include levels divided according to the energy flow information and levels divided according to the scale of the power grid;

[0026] The secondary network diagram represents the electrical paths between nodes in the primary network diagram.

[0027] Preferably, the construction of the primary network model includes:

[0028] Determine the structure and node values ​​of the primary network model according to the energy flow information and the power grid topology, and set the value of the edge resistance probability;

[0029] The structure, node values ​​and edge resistance probability values ​​of the primary network model are modified by machine learning methods.

[0030] Preferably, the construction of the secondary network model includes:

[0031] Determine the structure of the secondary network model and the value of the edge resistance probability according to the path constituent elements between the power grid observation points;

[0032] The secondary network model structure and the edge blocking probability are modified by machine learning methods.

[0033] Preferably, after constructing the power grid probability visualization model based on the node information and the edge information, the method further includes:

[0034] The values ​​of nodes and the weights of edges in the power grid information probability graph are corrected using a machine learning method.

[0035] Based on the same inventive concept, the present invention also provides a power grid probability visualization model construction system for machine learning, the improvement of which is that it includes: an initialization module, a node information module, an edge information module and a graphics module;

[0036] The initialization module is used to determine the scale, depth, granularity and level of the power grid information probability map according to the machine learning purpose;

[0037] The node information module is used to extract grid energy flow information from the physical grid to construct node information based on scale, depth, granularity and hierarchy;

[0038] The edge information module is used to design and extract the power grid energy flow environment and path information from the path information in the physical network relationship to construct the edge information;

[0039] The graphic module is used to construct a power grid probability visualization model based on node information and edge information.

[0040] Preferably, the power grid probabilistic visualization model construction system for machine learning further includes a correction module;

[0041] The correction module is used to correct the values ​​of nodes and edge weights in the power grid information probability graph using a machine learning method.

[0042] Compared with the closest prior art, the present invention has the following beneficial effects:

[0043] The present invention provides a method and system for constructing a power grid probability visualization model for machine learning, including: determining the scale, depth, granularity and level of a power grid information probability graph according to the purpose of machine learning; extracting power grid energy flow information from the physical power grid to construct node information based on the scale, depth, granularity and level; designing and extracting power grid energy flow environment and path information from the path information in the physical network relationship to construct edge information; and constructing a power grid probability visualization model based on the node information and edge information. The present invention establishes a power grid information probability graph based on power grid energy flow information and path information, which can accelerate intelligent data processing and machine learning, spy on the supporting efficiency of data and machine learning, and verify the purpose of data and machine learning law discovery. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A schematic diagram of a flow chart of a method for constructing a power grid probability visualization model for machine learning provided by the present invention;

[0045] Figure 2 A probability graphical illustration of power grid information provided by the present invention;

[0046] Figure 3 An example of a power grid information probability map provided by the present invention;

[0047] Figure 4 A graphical illustration of a power grid information probability model provided by the present invention;

[0048] Figure 5 An example of a probability model diagram of power grid information provided by the present invention;

[0049] Figure 6 The 30-node electrical diagram of the present invention;

[0050] Figure 7 It is a 30-node electrical segmentation diagram involved in the present invention;

[0051] Figure 8 It is a 30-node power grid information probability map involved in the present invention;

[0052] Fig. 9 A 30-node power grid information probability model diagram involved in the present invention;

[0053] Fig.10 A schematic diagram of the basic structure of a power grid probability visualization model building system for machine learning provided by the present invention;

[0054] Fig.11 A detailed structural diagram of a power grid probability visualization model construction system for machine learning provided by the present invention. DETAILED DESCRIPTION

[0055] The present invention provides a method and system for constructing a power grid probability visualization model for machine learning. The method and system are used for obtaining, verifying, correcting and improving an analysis method for power grid probability speculation based on machine learning, such as power grid operation management methods and rules.

[0056] This method of representing power grid data can extract relevant power grid data features according to the purpose of learning and simplify the complexity of learning data. This method can also represent the dynamic and static features of complex power grid data in layers and extract them by regions, simplifying data representation while increasing the implication and flexibility of the physical meaning of the power grid to support the purpose of learning. That is, based on the power grid data model and characteristic data, a method for establishing, extracting, storing and using learning data of power grid information is designed.

[0057] The significance of the method of the present invention is to seek a method for representing power grid characteristic data, so as to contain the rich meaning of the power grid and interpret more data meanings and implicit dynamic and static laws of the power grid through intelligent data processing and machine learning.

[0058] The specific implementation modes of the present invention are further described in detail below with reference to the accompanying drawings.

[0059] A flow chart of a method for constructing a power grid probability visualization model for machine learning provided by the present invention is as follows: Figure 1 As shown, including:

[0060] Step 1: Determine the scale, depth, granularity and level of the grid information probability map based on the machine learning purpose;

[0061] Step 2: Based on scale, depth, granularity and level, extract grid energy flow information from the physical grid to construct node information;

[0062] Step 3: From the path information in the physical network relationship, design and extract the grid energy flow environment and path information, and construct edge information;

[0063] Step 4: Construct a power grid probability visualization model based on node information and edge information.

[0064] Specifically, the power grid information in the present invention, such as the flow, is a kind of data rich in dynamic meaning for the observation of power grid characteristics. And the flow, as the most basic energy flow data for power grid operation, can be used as basic data for further learning. The distribution of power grid information in the network environment is affected by the power grid environment between the power grid information observation points. The definition and description of the relationship between the distribution of power grid information and the factors affecting the distribution is the power grid information probability map; after learning, the improvement of the power grid information probability map becomes a power grid information probability model that can be directly predicted and inferred and contains richer knowledge.

[0065] The method of the present invention is not limited to the representation of power grid flow, etc. The observed or derived quantity of power grid information can be the dynamic or static characteristics of the power grid, or the observed or unobservable quantity generated by operation. In this way, the learning of power grid operation knowledge rules supported by data can be extended to different fields and scopes such as power grid analysis and control.

[0066] The method of the present invention is described by taking power grid flow calculation information as example power grid information.

[0067] (I) Basic principles

[0068] The basic idea of ​​the present invention is to design and construct a data representation model and construction method that supports probabilistic reasoning of power grid information based on the physical and simulation computing network of the power grid, expert knowledge, and machine learning algorithms. It is a power grid data representation definition and representation learning method that supports machine learning.

[0069] The basic principle of the present invention is to use a probabilistic model diagram to effectively and simply represent the grid data of grid associations and characteristics. Grid data is very complex and the data scale is extremely large. It also has dynamic and static characteristics and is related to connection relationships and time. The digitization and informatization of the power system have represented the grid data in a relatively fixed form. When monitoring the dynamic operation of the power grid, it is mainly through equipment monitoring and physical calculation of the power system. Almost all power system analysis is based on grid flow data.

[0070] The present invention mainly includes a data representation method for constructing a probability diagram of the distribution of power grid dynamic characteristics such as power flow or voltage and current using prior knowledge of the power grid, and a construction and building method for a probability diagram of power grid information based on a probability inference method. This representation and building method ensures that the power grid information represents physical meaning.

[0071] To build a probabilistic model of power grid information, data and prior knowledge can be based on the current power system operation control, simulation analysis and calculation, or application system. These data are highly effective as learning data represented by power grid data.

[0072] The power grid information includes dynamic and static information. Generally speaking, the static characteristic data of the power grid mainly defines the influence relationship of the power grid information probability model, while the dynamic characteristic data of the power grid mainly defines the observation points in the power grid information probability model.

[0073] The grid information probability diagram can be designed and defined in a hierarchical manner.

[0074] For example, as a first-level power grid information probability diagram, the dynamic attributes of power grid information such as current are generally only related to the dynamic attributes of the operation of the observation point, such as the network node information of the traditional power grid calculation; the influencing factors of the observation point are related to the distance of the observation point, that is, the complexity of the power grid path between the observation points, which is a probability / weight representation of the impact on the observation point.

[0075] As a secondary power grid information probability diagram, it is the electrical path between the observation nodes in the primary diagram, that is, the elements and associations of the subnet composed of various power electronic components, as well as the redefined observation data and the probability / weight representation of the impact.

[0076] In the process of generating the representation of the power grid, it can be divided into the whole network-grid, which is to define the power grid information arbitrarily locally and globally. The semantics in the grid or power grid information probability map, the network composition can be the environment for energy flow and the pathways and networks for energy flow.

[0077] Essentially, these two levels of networks have the same construction, representation, and properties. They differ only in data construction and extraction methods. The first level network extracts data from observation points as nodes of power grid information such as power flow probability diagrams; the power path is designed as the weight value of power grid information distribution, which can be obtained through learning or after intelligent data processing; the second level network extracts weights from the network combination configuration design, and its internal information such as current and voltage can be obtained through learning or intelligent processing.

[0078] In addition, considering the scale of feature quantities, the network can be divided again at the same level, which is a secondary screening of the observed quantities and reconstruction of the probability model network.

[0079] That is, by abstracting the probability graph of the characteristic data of the power grid and then using the properties of the probability graph to process and learn the data, it is easier and more efficient to obtain research, analysis and operation guidance methods and discover laws for the power grid.

[0080] 1. (1) Grid characteristics extraction method

[0081] The structure and composition of the power grid is that different physical components are connected according to different rules, showing characteristics related to the connection method, rules and physical properties of the physical components. The dynamic characteristics of the power grid are more complex, which are manifested in the physical path arrival method of different monitoring points and the complex information types that need to be monitored. Including the static and dynamic properties of general physical components, the monitoring information impact of the network physical connection method, etc.

[0082] Based on the understanding of traditional power system calculation analysis and simulation methods, the basic dynamic information of the power grid is the power grid flow, voltage, current, frequency, etc. This is related to the setting of monitoring points for calculation analysis or monitoring system. This information is time-series and related to the power grid network structure.

[0083] In the method of the present invention, it is defined as a grid information probability diagram, and the feature extraction of grid data can be defined as: P, V, I, F, that is, flow, voltage, current, frequency, as a data representation of energy flow of grid information probability; and these characteristic data directly depend on the representation of the smoothness of the arrival path, which can generally be considered as the numerical influence representation of the comprehensive effect of impedance, etc. In the method of the present invention, it is defined as the resistance probability Pr, and the resistance probability Pr is directly related to the weight influence of the model. The grid structure is reflected in the correlation relationship and influence definition of energy flow between observation points.

[0084] By defining a probability graph of power grid information, we can extract power grid energy flow information, such as power flow, voltage, current, frequency, etc., from the physical power grid, and construct a power grid probability graph, that is, the node information of the power grid probability visualization model; from the path information in the physical network relationship, we can design and extract the power grid energy flow environment and path information, that is, the edge information of the power grid probability graph.

[0085] A probability graph of power grid information is defined, and the setting of observation quantities that are not directly physically connected is supported. For example, from the observation point of power generation setting, the observation point setting that is not directly to the power load is set, and a virtual association is set between the two. The network path is a virtual network path, but the impact relationship after the model is established is real.

[0086] Similarly, for the pathway network between observation points, its feature extraction and representation mainly lies in the extraction and definition of the components and connection methods of the power grid, but the information of the characteristic probability map of the pathway / network is different.

[0087] The probability diagram of power grid information of different observations and observation effects can be established in turn. The scope of the power grid is not limited, and the physical network data can be extracted based on the network relationship and calculation point information in the PSASP / BPA power flow calculation as a reference for data observation point extraction.

[0088] 1. (2) Data representation of power grid knowledge learning

[0089] Different from the traditional power grid calculation method, the machine learning algorithm uses statistics and probability to learn methods and rules for judging, inferring, and guiding many laws of the power grid. The essence of its learning method for acquiring domain knowledge is the intelligent statistical and probabilistic analysis of domain feature data.

[0090] Based on the composition and operation characteristics of the power grid, a Bayesian-like probability model of power grid data is constructed to support the utilization of its characteristics in machine learning algorithms.

[0091] Machine learning based on power grid information such as power flow probability diagram data is to obtain and discover more and smarter knowledge and rules contained in power grid information by inferring the flow characteristics of source flow data and the degree of influence of different network paths.

[0092] The grid information probability graph is easy to express grid information. The graphical representation of grid data can well carry the grid structure and attribute-related characteristics, and it is also easy to use the graph properties and probabilistic properties to support machine learning algorithms.

[0093] The power grid information probability visualization model designed by the method of the present invention provides good learning data for machine learning, and machine learning will improve the effective relationship represented by the data.

[0094] The power grid information probability graph can be represented as G(N, E), which consists of points and edges. Points are observation points in the power grid and have certain monitoring information; edges are the road network connecting points, which is composed of points and edges. The physical property of the edge is the weight that the observation point information can obtain / pass through the path.

[0095] Edge is a context composed of the association and relationship of information of points in a power grid. It can be as small as a line and a single component or as large as a subnet or regional network of the power grid. At the same time, it supports virtual setting relationships.

[0096] Point - is a kind of power grid semantic information definition, which can be any characteristic or combination of power grid information.

[0097] For example, the power flow probability graph is designed as a directed graph model, i.e., a quasi-Bayesian network. For example, the power flow distribution point variables are x1, x2, ... xn. In the graph, each flow variable is only related to one or several flow variables that it reaches by its path; its arrival path is used as an edge variable. Then, we can use graphical tools and mathematical methods to process and learn power grid data, and we can learn:

[0098] Improve the definition and construction of Bayesian-like networks for power grids, and represent the influence and reasoning relationships of spatially variable information including the flow over time and the relationship between connection locations;

[0099] Methods of reasoning and inference, such as exact inference or approximate inference, can be used and obtained;

[0100] Utilize the parameters of the probabilistic model graph and structured learning methods and reverse reasoning decision-making under uncertain conditions such as power grid planning or problem assumptions.

[0101] 1. (3) Grid information such as power flow probability diagram principle and method characteristics

[0102] Using probability and graphics as two tools, power grid information, including power grid information attributes and power grid structure, is represented and established. Its method features facilitate power grid knowledge representation and practical machine learning methods. On the other hand, the constructed power grid information, such as power flow probability graph model method, also has the following characteristics:

[0103] With the help of the deformation of the power grid connection diagram, the power grid information and energy flow distribution are represented, and the observed feature associations are intuitively visualized, which facilitates the effective expression of correlations and abstract main features.

[0104] The visualization of the probabilistic model graph of the deduction relationship is convenient for designing and deriving new models, discovering new characteristic rules, etc. For example, the graphical representation of the impact of power grid components and structures on the distribution of power grid information. The graphical model can be used to design new models and learn and solve more fine-grained power grid problems.

[0105] The intuitive model internal characteristics of the association relationship, including the conditional dependence and independence of power grid information, can be easily obtained from the graphical representation, such as the implicit and abstract representation of the influence of power grid component relationships and the representation of power grid structure, i.e., topological factors.

[0106] The layering and zoning abstraction of complex power grid structures generates simple multi-layer or regional data models, degrading complex reasoning problems. For example, complex calculations in the power grid are related to the arrival path and timing of energy and its information flow; complex calculations on the probability of power flow impact are understood as information transmission on the graph and the display of degree and method.

[0107] From another perspective, the acquired observations contain the main power grid information, and more and more complete information can be obtained through data algorithm learning; the probability model effectively represents the power grid information and the constraints between them, and the probability theory is used in the power grid data processing process.

[0108] (II) Representation method of power grid probability diagram

[0109] II. (1) Basic Definition and Representation

[0110] These definitions are used for power grid information probability diagrams, which are an engineered and de-attributed definition of the representation of power grid components in probability diagrams, where the de-attribution is mainly in terms of dimension and precision data.

[0111] Energy flow / grid information flow: It is the dynamic information of the grid. Such as active and reactive power, current, voltage, frequency, and cycle, etc. The energy flow is expressed by normalization method. Normalization refers to the processing on data accuracy.

[0112] Resistance probability: The degree of smoothness of the path, which indicates the weight of the influence of its impedance, loss, etc. on the energy flow.

[0113] Power Flow Probability Diagram: The association table defines a storage representation.

[0114] Road network: The path from one observation point to another, which can be a connection composed of different component configurations.

[0115] Path (network) resistance: It is the static information of the power grid. The composite value of the impedance information of the path circuit where the power grid information flows can be obtained through simple circuit calculation.

[0116] These concepts and methods are used to process data according to the representation needs when extracting power grid information learning data.

[0117] II. (2) Grid Information Abstraction Methods and Rules

[0118] The power grid components are de-attributed, and the impact of component-related attributes on power grid information such as power flow is defined as the resistance weight of the resistance probability to the energy flow. Different components and different network connections are represented by the influence of different resistance probabilities on the distribution of power grid information in the network structure.

[0119] Reliance refers to the consistent processing of the knowledge affecting the power grid information relationship based on component attributes.

[0120] That is, the properties of all components are attributed to their influence on the power grid information; their values ​​and the relationship between the static and dynamic properties of the components and other components are determined through circuit calculation and probability deduction / machine learning.

[0121] That is, the power grid is normalized and represented. The information in the power grid corresponds to the nodes in the probability graph, and the paths and subnets are abstracted as the edges in the probability graph; the information type and network structure are classified and defined to define the node class and edge class in the probability graph; the principle is to simplify the point-graph relationship and streamline the power grid probability graph structure.

[0122] Abstract principles and available methods for power grid data representation:

[0123] Remove features to facilitate retention of data master information;

[0124] The representation method can contain the characteristics of the power grid structure;

[0125] Build a good data model for deep learning;

[0126] Normalization methods, such as random continuous and random discrete transformation of data, express engineering data as discrete integers that meet upper and lower limits according to the highest accuracy requirements, such as 0.0001, expressed as 1, and the transformation digital-to-analog ratio is 10000;

[0127] The power flow probability diagram of the entire network is the full data set.

[0128] Reference methods for reversion and normalization:

[0129]

[0130] II. (3) Definition method of grid information probability graph

[0131] The probability of power grid information is defined as the data structure of the power flow diagram:

[0132] Node: Flow information, the connection between points is established through edge paths (networks).

[0133] Edge: The weight of the path's impact on power flow distribution; the edge is directed, from the starting point to the end point.

[0134] nature:

[0135] The node information value is a mathematical calculation of the weights of all edges reaching the point and the information value of the starting point of the edge. The node information value is calculated by mathematical methods through the weights of all edges reaching the point and the information value of the starting point of the edge; or it can be directly extracted from the power grid observation point during the initial construction.

[0136] One edge can only reach one node; one information point can be the starting point of multiple edges; multiple edges can reach one information point.

[0137] The number of edges that a node of the origin type reaches is 0; the number of edges that a node of the end type starts at is 0.

[0138] Representation method:

[0139] Power grid flow probability graph Gp(Nn,Em): There are m edges, i.e., power grid paths E 1_2 , E 2_2 ,......,E i_j There are n information points N1, N2, ..., N n . where i,j≤n,E i_j is an edge from node i to node j; the naming of edges and nodes in the graph can be defined according to application purpose and edge and node types.

[0140] Data storage structure definition:

[0141] As shown in Table 1.

[0142] Table 1: Power grid information probability map data storage table

[0143] Node ID Node Type Arrival nodes / edge lists Starting edge / node list N1 0-yuan points <![CDATA[E 1_3 AND 1_5 ]]> N2 0-yuan points <![CDATA[E 2_4 AND 2_9 ]]> N3 1-Midpoint <![CDATA[N1 / E 1_3 ]]> <![CDATA[E 3_7 AND 3_4 ]]> N4 1-Midpoint <![CDATA[N3 N2 / E 3_4 AND 2_4 ]]> <![CDATA[E 4_8 ]]> N5 1-Midpoint <![CDATA[N1 / E 1_5 ]]> <![CDATA[E 5_9 ]]> N6 0-yuan points N7 6-End point <![CDATA[N3 / E 3_7 ]]> N8 6-End point <![CDATA[N4 / E 4_8 ]]> N9 6-End point <![CDATA[N2 N5 / E 2_9 AND 5_9 ]]>

[0144] Note: Node type: 0-element point, 6-end point, 1-midpoint, 2-compound type (indefinite point)

[0145] Visualization graphics method:

[0146] Visualize the graph for example Figure 2 As shown, the indefinite point (composite type) is a combination of multiple types of points, including both the midpoint and the end point, or the origin and the end point, or all three types;

[0147] There are two forms of edge labels: on-line and on-edge. The lower-level labels are the start and end point labels. D* can be used to represent a bidirectional edge.

[0148] Graphics and data storage correspondence method:

[0149] Graph-number: Establish mapping methods for single-class points, indefinite points, and single edges according to the point types in the graph.

[0150] Number-graph: Based on the data type, search for the starting edge and the arrival edge, determine the type of the point class in the graph and the relationship between the edges; you can also determine the mapping method from graph to number based on the purpose of data use.

[0151] See Table 3, where the arrival node / edge list and the starting edge / node list columns contain duplicate edges; in the actual construction algorithm, for the "arrival node / edge list", the edges whose starting node number is less than the current node number are exempted; and for the "starting edge / node list", the edges whose arrival node number is less than the current node number need to be exempted.

[0152] II. (4) Design of power grid information learning data

[0153] According to the probabilistic graphic design of power grid information, the learning data of power grid information, i.e. data samples, can be designed and extracted to support the representation of its graphic features. Its learning data includes: data set of observation point features and initial association relationship set, where the initial association relationship set is mainly for the observation relationship setting, which is generally preliminarily set according to the actual application system.

[0154] II. (5) Probabilistic model of power grid information

[0155] The grid information probability map and the grid information probability model are consistent under certain settings.

[0156] The construction of the power grid information probability graph and the improvement of the graph parameters through representation learning to obtain the power grid information probability graph is a representation method about the power grid knowledge rules that can be used for inference, judgment and other applications.

[0157] The data information of the model is shown in Table 2 below:

[0158] Table 2: Power grid information probability model data storage table

[0159]

[0160] Graphical method of the model:

[0161] Basic symbol diagrams such as Figure 4 shown.

[0162] Consistent with the graph representation, the nodes are labeled with information types, and the edges are labeled with weights. D* can be used to represent bidirectional edges with two weight labels.

[0163] Example picture Figure 5 shown.

[0164] Graphics and data storage correspondence method:

[0165] Model-Number: Same as Figure-Model. The content of the data column is a list of values, and the two correspond.

[0166] Table 3 is the data table of the IEEE-30 node power grid information probability diagram, and Table 4 is the data table of the IEEE-30 node power grid power flow probability model.

[0167] Table 3: IEEE-30 node power grid information probability diagram data table

[0168]

[0169]

[0170] Note: Node type: 0-element point, 6-end point, 1-midpoint, 2-compound point

[0171] Table 4: IEEE-30 node power flow probability model data table

[0172]

[0173]

[0174] Note: Node type: 0-element point, 6-end point, 1-midpoint, 2-compound point

[0175] See Table 4, where the data in the columns of Arrival Parameter Table | Starting Parameter Table are repeated; the method for handling redundant information is consistent with that in Table 3.

[0176] 3.3 Method for constructing a probability model of power grid information

[0177] In principle, a grid information probability map of any level can be constructed according to the grid information and the relationships between them.

[0178] Two layers of data models are defined based on the secondary network of the power grid. The primary network diagram is the power grid flow probability diagram, and the secondary network can be a flow path network. The simplest one can be a single line, and the complex one can be a sub-circuit network of a power grid; of course, it also includes a virtual path network.

[0179] Constructing a power grid information probability graph is to construct a quasi-Bayesian network by extracting the power grid topology and component attribute information features and abstracting the network relationship, including the establishment of power grid information probability graphs at different levels.

[0180] III. (1) Main tasks and methods

[0181] Definition of observation point variables:

[0182] According to the purpose of using the power grid data, the observation point variables are determined, such as dynamic information flow, voltage, current, frequency, etc., or static information component setting information, etc.; the observation variable can be a single power grid characteristic observation or a combination vector of multiple power grid characteristic quantities.

[0183] For power grid research with different purposes, the requirements for power grid information granularity and density are also different. For example, even if power flow is selected as the observed variable, its sampling points can be set at the provincial grid level and the county level.

[0184] The observation point variables can be one type of power grid characteristics or different types of characteristics. Variables with different characteristics can be defined in the same power grid information probability map, or they can be layered to establish their own power grid information probability maps.

[0185] Learning the structure definition of the probability graph of power grid information:

[0186] It is mainly based on the connection and correlation of observation point variables in the power grid environment and the correlation that is out of the constraints of the power grid environment.

[0187] After the definition of observation point variables is completed, the association relationship of the observation points can be preliminarily established based on the characteristics and characteristic relationships of the variables in the power grid.

[0188] It can be directly generated from the grid learning data or obtained through mathematical calculations. At the same time, through the learning of grid data, the probability graph structure information is improved, mainly the determination of the influence relationship of the observation point.

[0189] (Semi)parameter (edge-weight) modification and learning:

[0190] During the grid data learning process, the parameters such as the observed variable information and influencing relationship of the grid information probability map will be adjusted and corrected.

[0191] III. (2) Working methods and processes

[0192] Determine model size and depth:

[0193] Define the scale of the network model, which can be equivalent to the scale of the power grid.

[0194] In general research, building only a first-level network model is sufficient to support the study of the power grid.

[0195] The secondary network model definition of the method of the present invention focuses on studying and learning the configuration, planning, adjustment of the power grid and the impact on the observation information of the primary network from the perspective of the physical structure of the power grid.

[0196] Generally speaking, the primary network focuses on studying the operation of the power grid, which is mostly dynamic information; the secondary network focuses on studying the configuration of the power grid, which is mostly static information.

[0197] like Figure 8 , select the scale of all 30 nodes, that is, the observation points {N1, N2, ..., N30}, which is the first-level network selection of the whole network scale. The observation features can be selected as P, V, I, H, etc., such as defining Gp(N30, Ek), where K can be initially preset to a maximum value of Figure 8 All the edges; if from Figure 6 From the electrical diagram, the physical wiring of node 6 can be known, a secondary network about node 6 can be defined, and the internal electrical connection rules and the configuration and configuration adjustments of external observation points can be learned.

[0198] Model granularity determination:

[0199] The granularity of the grid observation information and the associated settings of the levels and observation points is determined according to the purpose of learning and research. A high granularity of grid feature extraction will make the model structure of the grid probability model diagram complex, resulting in complex probability reasoning and difficulty in learning.

[0200] In a large-scale power grid, the number of observation points is huge, but some characteristic points can be selected for observation according to the agreed rules based on the learning purpose. Figure 8 , out of the total 30 nodes, selecting only the observation points {N2, N12, N15, N28} is a low-granularity choice.

[0201] Model hierarchy determination:

[0202] For the same-level network subdivision, it is a division and utilization of the subnet of the network model constructed at this level.

[0203] The model level can be set according to different grid information as different observed characteristic variables, such as establishing grid current probability graphs or grid voltage probability graphs respectively.

[0204] The model hierarchy can also divide the network into blocks and establish multiple independent models, such as provincial power grid flow probability diagrams and regional power grid flow probability diagrams.

[0205] The division methods can be used interchangeably. At the same time, the use of power grid data is also multi-purpose.

[0206] like Figure 8, select all 30 nodes, that is, observation points {N1, N2, ..., N30}, and select the power grid flow probability model with observation characteristics P; or select all 30 nodes {N1, N2, ..., N30} and select the power grid voltage probability model with observation characteristics V.

[0207] or Figure 6 , divide the physical network into two networks according to the dotted line, and define the power grid information model of the two networks respectively. And set the probability model between subnets according to the physical characteristics. This application does not explain, but can be processed according to the granularity selection method.

[0208] Initial model establishment:

[0209] After determining the scale and depth of the model, as well as the preliminary granularity and level, the model can be established.

[0210] like Figure 8 and Table 3, in both graphical and data table form.

[0211] The model is initially established based on the definition of the grid calculation points and the path network between the points. The data is accurate and easy to obtain; the network is relatively simple. In this way, the cost of building and maintaining the model and the complexity of probability solution and reasoning are relatively reduced.

[0212] The learning of the first-level network mainly focuses on the learning of network weight parameters and the correction and improvement of network parameters.

[0213] like Fig. 9 and Table 4, in graphical and data table form.

[0214] The choice of model size in practice is related to the power grid data and the purpose of learning. If the node information of the probability model is the power grid flow variable, the influence weight of the network is determined by the network circuit.

[0215] The characteristics of power grid data are highly identifiable, which supports the establishment of an initial Bayesian-like network and provides more meaningful power grid learning data, making the iterative interactive process of model improvement and parameter learning more effective.

[0216] Therefore, representing the power grid data through the power grid probability visualization model of the present invention is the best support for building a learning model. Through the application of mature technologies and the guidance of domain experts, variables in the research problem domain are selected and extracted, and appropriate strategies are provided to select important factors from the variables provided by experts. For the key points and difficult problems of building a Bayesian-like network for the power grid, a directed acyclic graph is constructed based on the physical structure of the power grid and the distribution parameters of each node in the graph are given, that is, each node, that is, the power grid characteristic, corresponds to a conditional probability distribution under a power grid environment.

[0217] (IV) Application method of power grid probability map

[0218] The network resistance probability model diagram obtained through machine / deep learning can obtain the network resistance value under given energy flow conditions, and the road resistance adjustment model obtained through secondary learning of this data value can learn the configuration adjustment and planning expectation method of the road network.

[0219] The power grid information probability visualization model designed by the present invention can achieve ideal performance for all supported machine learning algorithms and can support different methods for constructing Bayesian-like networks.

[0220] The constructed power grid information probability graph is a quasi-Bayesian network. The power grid information probability is independent of certain conditions, and its network parameter learning method is relatively simple.

[0221] Take the power grid flow probability model and the power grid configuration probability model as examples. As follows.

[0222] The power grid flow probability model is the first-level network:

[0223] First, the initial model is constructed. The Bayesian network structure and data are defined by the power grid flow and power grid topology, and the network resistance probability is used as a parameter. The variables of the Bayesian network are preset by domain experts, and the Bayesian network structure is initially set using expert knowledge, and its distribution parameters are specified by learning. Then, the deviation of the constructed network is corrected through data learning to make up for the limited knowledge of domain experts.

[0224] Secondly, learning and correction. For the grid information probability model graph established based on grid data, the node variables and edge probability weights defined in this way require training data to learn and correct the structure and parameters of the Bayesian network. This is also the key to learning the structure and parameters of the Bayesian network of the grid from data.

[0225] The grid configuration probability model is the secondary grid:

[0226] Initial construction of the secondary network. The paths between the power grid flow calculation points constitute the component data to define the internal network structure and data. The static and dynamic data of the components are used as parameters. Through circuit relationships and calculations, the variables of the internal network are obtained and its distribution parameters are specified. This process requires the inclusion of important information about the power grid composition and components, the preset node characteristic variables of the power grid configuration probability model, and the use of expert knowledge and the topological model of the physical power grid to preliminarily specify the network structure.

[0227] Learning and correction of the secondary network. The secondary network model is corrected by learning the network parameter configuration, structural correction rules or circuit physical calculation methods from data through machine learning methods.

[0228] After the grid probability visualization model is established, it can be used to construct a visualized grid topology relationship, including:

[0229] Measure and collect location information of power grid equipment points, power grid lines, and power grid area surfaces, and associate them with the power grid probability visualization model to establish a complete equipment space data table;

[0230] Collect business information on power grid equipment points, power grid lines and power grid area surfaces, fill in business information data, and establish equipment attribute data tables; topologically associate the attribute data of the corresponding equipment with the spatial data through the equipment attribute data table and the equipment spatial data table to form an equipment topological relationship information table;

[0231] Through the equipment topology relationship information table, the GIS system is used to build the power grid topology relationship, and the topology information and equipment distribution of the entire power grid are displayed on the map.

[0232] The grid probabilistic visualization model can also be used to build a three-dimensional power network, including:

[0233] Establish power grid construction objects based on the power grid probability visualization model and the power grid database, record the timestamp information of the actual power grid construction, and build a timestamp-based power grid construction object achievement database;

[0234] Design the linear reference mechanism and network topology model mechanism of the power grid probability visualization model, extract the power grid equipment models at different power grid construction stages through timestamp information for spatial analysis, establish a geocoding scheme that reflects the spatial topological relationship, and associate the equipment objects in different time periods and the same spatial range with a database based on spatial topology to obtain a database of association results based on spatial topology analysis;

[0235] The time-stamp-based database indexing technology is used to establish spatiotemporal index information for the associated achievement database, and the spatiotemporal index information is stored in the associated achievement database to complete the construction of the three-dimensional power network.

[0236] The grid probabilistic visualization model can also be used to conduct real-time and online analysis of the grid, including:

[0237] Based on the probabilistic visualization model of the power grid, trajectory pattern mining methods, machine learning methods and electromechanical hybrid simulation methods are used to determine the spatiotemporal dynamic operation characteristics and evolution laws of the power grid;

[0238] For the static scenario of power grid, a static stability domain with nonlinear boundaries is constructed to identify the comprehensive stability margin of power grid nodes;

[0239] According to the dynamic scenario of power grid, a dynamic stability domain is constructed to evaluate the dynamic stability of the power grid;

[0240] In view of the future stability of the power grid, deep recurrent neural network technology is used to predict the future status of the power grid based on the long-term and short-term memory information of the power grid node state parameters and the status information of spatially adjacent nodes.

[0241] Embodiment 2:

[0242] Based on the same inventive concept, the present application also provides a power grid probability visualization model construction system for machine learning. Since the principles of these devices for solving technical problems are similar to the power grid probability visualization model construction method for machine learning, the repeated parts will not be repeated.

[0243] The basic structure of the power grid probability visualization model construction system for machine learning is as follows: Fig.10 As shown, it includes: an initialization module, a node information module, an edge information module and a graph module;

[0244] An initialization module for determining the scale, depth, granularity and level of the grid information probability map according to the machine learning purpose;

[0245] The node information module is used to extract grid energy flow information from the physical grid and construct node information based on scale, depth, granularity and hierarchy;

[0246] The edge information module is used to extract the power grid energy flow environment and path information from the path information in the physical network relationship and construct the edge information;

[0247] Graphics module, used to build a probabilistic visualization model of the power grid based on node information and edge information.

[0248] The detailed structure of the power grid probability visualization model construction system for machine learning is as follows: Fig.11 shown.

[0249] The path information includes:

[0250] The connection relationship and association relationship of energy flow information values ​​of observation points equipped with energy flow information sensors in the power grid in the power grid and the association relationship that is free from the constraints of the power grid environment.

[0251] The energy flow information includes: power flow, voltage, current and frequency.

[0252] Among them, the nodes in the power grid information probability graph include: source point, midpoint, end point and indeterminate point;

[0253] Among them, the element points include power generation points, power supply points and power sources;

[0254] The end points include power consumption points and loads;

[0255] The midpoint includes an observation point set up for observing power grid information;

[0256] An indefinite point includes a composite point consisting of at least two of the origin, the middle point and the end point.

[0257] Among them, the edges in the power grid information probability graph correspond to the components, connections and road networks in the power grid, the edges have directions, and the edges include resistance probabilities;

[0258] The resistance probability represents the influence weight of impedance and loss on the energy flow information flowing through the edge where the resistance probability exists.

[0259] The value of the node is obtained by the resistance probability of the edge reaching the node and the value of the starting point of the edge, or is directly extracted from the grid observation point corresponding to the node when constructing a grid information probability model.

[0260] Among them, the scale of the power grid information probability map is set according to the scale of the power grid;

[0261] The depth of the power grid information probability map is set according to the research requirements, and the depth includes the first-level network map and the second-level network map;

[0262] The granularity of the power grid information probability graph is set according to the learning and research purposes, and the granularity includes the number of selected nodes;

[0263] The levels of the power grid information probability diagram are set according to the power grid information, and the levels include the levels divided according to the energy flow information and the levels divided according to the scale of the power grid;

[0264] Among them, the second-level network diagram represents the electrical path between nodes in the first-level network diagram.

[0265] The system further comprises a first-level network diagram construction module, which comprises: a first-level diagram setting unit and a first-level diagram correction unit;

[0266] A first-level graph setting unit, used to determine the structure and node values ​​of the first-level network graph according to the energy flow information and the power grid topology, and to set the value of the edge resistance probability;

[0267] The first-level graph correction unit is used to correct the structure, node values ​​and edge resistance probability values ​​of the first-level network graph through machine learning methods.

[0268] The system further comprises a secondary network diagram construction module, which comprises: a secondary diagram setting unit and a secondary diagram correction unit;

[0269] A secondary graph setting unit, used to determine the structure of the secondary network graph and the value of the resistance probability of the edge according to the path constituent elements between the power grid observation points;

[0270] The secondary graph correction unit is used to correct the secondary network graph structure and edge blocking probability through machine learning methods.

[0271] Among them, the system also includes a correction module;

[0272] The correction module is used to correct the values ​​of nodes and edge weights in the power grid information probability graph using a machine learning method.

[0273] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0274] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0275] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0276] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0277] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application rather than to limit its protection scope. Although the present application has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that after reading the present application, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the application, but these changes, modifications or equivalent substitutions are all within the protection scope of the claims to be approved.

Claims

1. A method for constructing a power grid probability visualization model for machine learning, characterized by: include: Determine the scale, depth, granularity and level of the grid information probability map based on machine learning purposes; Based on scale, depth, granularity and level, grid energy flow information is extracted from the physical grid to construct node information; From the path information in the physical network relationship, the design extracts the grid energy flow environment and path information and constructs the edge information; Construct a probabilistic visualization model of power grid based on node information and edge information; The scale of the power grid information probability map is set according to the scale of the power grid; The depth of the power grid information probability map is set according to research requirements, and the depth includes a primary network map and a secondary network map; The granularity of the power grid information probability graph is set according to the learning and research purpose, and the granularity includes the number of selected nodes; The levels of the power grid information probability map are set according to the power grid information, and the levels include levels divided according to the power grid energy flow information and levels divided according to the scale of the power grid; Wherein, the secondary network diagram represents the electrical paths between nodes in the primary network diagram; The construction of the primary network model includes: Determine the structure and node values ​​of the primary network model according to the power grid energy flow information and the power grid topology, and set the value of the edge resistance probability; Correcting the structure, node values ​​and edge resistance probability values ​​of the primary network model by machine learning methods; The construction of the secondary network model includes: Determine the structure of the secondary network model and the value of the edge resistance probability according to the path constituent elements between the power grid observation points; The secondary network model structure and the edge blocking probability are modified by machine learning methods.

2. The method according to claim 1, characterized in that The path information includes: The connection relationship and association relationship of energy flow information values ​​of observation points equipped with energy flow information sensors in the power grid in the power grid and the association relationship that is free from the constraints of the power grid environment.

3. The method according to claim 1, characterized in that The power grid energy flow information includes: power flow, voltage, current and frequency.

4. The method according to claim 1, characterized in that The nodes in the power grid information probability graph include: source points, midpoints, end points and indefinite points; Among them, the element points include power generation points, power supply points and power sources; The end points include power consumption points and loads; The midpoint includes an observation point set up for observing power grid information; An indefinite point includes a composite point consisting of at least two of the origin, the middle point and the end point.

5. The method according to claim 1, characterized in that The edges in the power grid information probability graph correspond to the elements, connections and road networks in the power grid, the edges have directions, and the edges include resistance probabilities; The resistance probability represents the influence weight of impedance and loss on the power grid energy flow information flowing through the edge where the resistance probability is located.

6. The method according to claim 5, characterized in that The value of the node is obtained by the barrier probability of the edge reaching the node and the value of the starting point of the edge, or is directly extracted from the power grid observation point corresponding to the node when constructing the power grid information probability model.

7. The method according to claim 1, characterized in that After the power grid probability visualization model is constructed based on the node information and the edge information, the method further includes: The values ​​of nodes and the weights of edges in the power grid information probability graph are corrected using a machine learning method.

8. A power grid probability visualization model construction system for machine learning, characterized in that: include: Initialization module, node information module, edge information module and graph module; The initialization module is used to determine the scale, depth, granularity and level of the power grid information probability map according to the machine learning purpose; The node information module is used to extract grid energy flow information from the physical grid to construct node information based on scale, depth, granularity and hierarchy; The edge information module is used to design and extract the power grid energy flow environment and path information from the path information in the physical network relationship to construct the edge information; The graphics module is used to construct a power grid probability visualization model based on node information and edge information; Among them, the scale of the power grid information probability map is set according to the scale of the power grid; The depth of the power grid information probability map is set according to the research requirements, and the depth includes the first-level network map and the second-level network map; The granularity of the power grid information probability graph is set according to the learning and research purposes, and the granularity includes the number of selected nodes; The levels of the power grid information probability map are set according to the power grid information, and the levels include the levels divided according to the power grid energy flow information and the levels divided according to the scale of the power grid; Among them, the second-level network diagram represents the electrical paths between nodes in the first-level network diagram; The system further comprises a first-level network diagram construction module, which comprises: a first-level diagram setting unit and a first-level diagram correction unit; A first-level graph setting unit, used to determine the structure and node values ​​of the first-level network graph according to the grid energy flow information and the grid topology, and to set the value of the edge resistance probability; A first-level graph correction unit, used to correct the structure, node values ​​and edge resistance probability values ​​of the first-level network graph through a machine learning method; The system further comprises a secondary network diagram construction module, which comprises: a secondary diagram setting unit and a secondary diagram correction unit; A secondary graph setting unit, used to determine the structure of the secondary network graph and the value of the resistance probability of the edge according to the path constituent elements between the power grid observation points; The secondary graph correction unit is used to correct the secondary network graph structure and edge blocking probability through machine learning methods.

9. The system according to claim 8, characterized in that Also includes correction modules; The correction module is used to correct the values ​​of nodes and edge weights in the power grid information probability graph using a machine learning method.