A method for dispatching and controlling a distribution network based on a situation awareness technology

Abstract

The present invention relates to the field of artificial intelligence technology, and discloses a method for dispatching and controlling a distribution network of a situation awareness technology, including: collecting multi-source data of the distribution network; constructing an initial distribution network model and time series features of the distribution network based on the multi-source data, and generating feature vectors of nodes in the initial distribution network model; using an improved adjacency matrix to perform vector fusion on the feature vectors to obtain local fusion features, wherein the improved adjacency matrix dynamically adjusts the connection weights between nodes through a dynamic influence function and corrects the node degrees based on load differences; generating dynamic weights of the local fusion features according to the time series features, performing feature dimensionality reduction on the local fusion features according to the dynamic weights, predicting the load change trend of the distribution network, and combining safety constraint conditions of voltage limit and current limit to generate a dispatching strategy for the distribution network, and performing dispatching and control on the distribution network according to the dispatching strategy. The present invention can improve the efficiency of dispatching and controlling the distribution network.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular, to a method for dispatching and controlling a distribution network using situation awareness technology. Background Art

[0002] With the continuous development of the power system, the importance of the distribution network in the entire power supply system has become increasingly prominent. However, the current dispatching and control of the distribution network face many challenges. On the one hand, the structure of the distribution network is becoming more and more complex, with numerous nodes and variable connection relationships, making it difficult for traditional dispatching and control methods to comprehensively and accurately grasp its operating state; on the other hand, the factors affecting the operation of the distribution network are diverse. In addition to the state of the equipment itself and the change of power load, the influence of environmental factors (such as temperature, humidity, light, etc.) on the distribution network has also been gradually emphasized, and the existing dispatching and control means have deficiencies in comprehensively considering these multi-source factors.

[0003] At the same time, in terms of data processing, a large amount of structured data, environmental data, real-time operation data, and equipment data from different channels are difficult to be efficiently integrated and analyzed, resulting in the inability to timely and accurately predict the load change trend, thereby affecting the scientificity and rationality of the dispatching strategy, making the efficiency of the distribution network dispatching and control relatively low, and it is difficult to meet the strict requirements of modern power systems for reliability and stability. Therefore, how to improve the efficiency of the distribution network dispatching and control has become an urgent problem to be solved. Summary of the Invention

[0004] The present invention provides a method for dispatching and controlling a distribution network using situation awareness technology, and its main purpose is to solve the problem of relatively low efficiency in the dispatching and control of the distribution network.

[0005] To achieve the above object, a method for dispatching and controlling a distribution network using situation awareness technology provided by the present invention includes:

[0006] Collect multi-source data of the distribution network, where the multi-source data includes: structured data, environmental data, real-time operation data, and equipment data;

[0007] Based on the structured data and the equipment data, construct an initial distribution network model of the distribution network, perform vector transformation on the attribute information corresponding to the nodes in the initial distribution network model to obtain the feature vectors of the nodes;

[0008] Use an improved adjacency matrix to perform vector fusion on the feature vectors to obtain the local fusion features of the nodes in the distribution network, where the improved adjacency matrix dynamically adjusts the connection weights between nodes through a dynamic influence function and corrects the node degrees based on the load difference;

[0009] Generate the time series characteristics of the distribution network based on the environmental data and the real-time operation data, generate the dynamic weights of the local fusion features according to the time series characteristics, and perform feature dimensionality reduction on the local fusion features according to the dynamic weights to obtain the dimensionality-reduced features of the nodes in the distribution network;

[0010] Predict the load change trend based on the dimensionality-reduced features, combine the safety constraint conditions of voltage limit and current limit, generate the scheduling strategy of the distribution network, and perform scheduling and control on the distribution network according to the scheduling strategy.

[0011] Optionally, the collection of multi-source data of the distribution network includes:

[0012] Collect data of the nodes of the distribution network based on the distributed sensor network to obtain the initial data of the distribution network;

[0013] Clean the initial data to obtain the cleaned data of the distribution network;

[0014] Classify the cleaned data to obtain the classified data of the distribution network, and determine the classified data as the multi-source data of the distribution network.

[0015] Optionally, the construction of the initial distribution network model based on the structure data and the equipment data includes:

[0016] Determine the nodes and connection relationships of the distribution network according to the structure data and the equipment data;

[0017] Generate the topological structure of the distribution network according to the nodes and the connection relationships;

[0018] Generate the initial distribution network model of the distribution network based on the graph theory method and the topological structure.

[0019] Optionally, the vector transformation of the attribute information corresponding to the nodes in the initial distribution network model to obtain the feature vectors of the nodes includes:

[0020] Extract the attribute information of the nodes in the initial distribution network model, where the attribute information includes: voltage level, load type, and equipment status;

[0021] Perform numerical coding on the attribute information to obtain the numerical features of the attribute information;

[0022] Perform feature combination on the numerical features to obtain the feature vectors of the nodes.

[0023] Optionally, the use of the improved adjacency matrix to perform vector fusion on the feature vectors to obtain the local fusion features of the nodes in the distribution network includes:

[0024] Perform matrix multiplication on the feature vector and the improved adjacency matrix to obtain the local fusion features of the nodes in the distribution network, where the improved adjacency matrix is:

[0025] ;

[0026] Where is the improved adjacency matrix, is the degree matrix, is the adjacency matrix, is the degree of node , indicates whether there is an edge connecting node and node , is the node and node the improved connection weight between, represents all node sets adjacent to node , is the node and node the dynamic influence function between, is the node and node the load change difference between, is the node and node the environmental data difference between, is the node and node the device state similarity between, represents the node identifier of the th node, is the th node identifier, is a constant coefficient for adjusting the weight of the adjacency matrix, is the total number of nodes, is a constant coefficient for adjusting the influence of load difference on degree.

[0027] Optionally, generating the dynamic weight of the local fusion feature according to the temporal feature includes:

[0028] Determine the key influencing factors of the local fusion feature based on the temporal feature;

[0029] Generate the dynamic weight of the local fusion feature based on the key influencing factors.

[0030] Optionally, performing feature dimensionality reduction on the local fusion feature according to the dynamic weight to obtain the dimensionality-reduced features of the nodes in the distribution network includes:

[0031] Multiply the dynamic weights and the local fusion features element by element to obtain weighted local fusion features;

[0032] Perform dimensionality reduction on the weighted local fusion features based on the principal component analysis algorithm to obtain the dimensionality-reduced features of the nodes in the distribution network.

[0033] Optionally, the predicting the load change trend according to the dimensionality-reduced features includes:

[0034] Input the dimensionality-reduced features into a pre-trained load change trend prediction model;

[0035] Use the load change trend prediction model to perform pattern recognition and trend inference on the dimensionality-reduced features to obtain the load change trend of the distribution network, where the load change trend includes: the rising, falling, and steady state information of the load at different time scales, the time when the load peak and valley appear, and the corresponding load values.

[0036] Optionally, it is characterized in that the method for generating the scheduling strategy is as follows: Determine the load demand and power supply capacity of the distribution network in different time periods according to the load change trend;

[0037] Generate the scheduling strategy of the distribution network according to the load demand and the power supply capacity, in combination with the safety constraint conditions of voltage limit and current limit.

[0038] Optionally, the scheduling and control of the distribution network according to the scheduling strategy includes:

[0039] Parse the scheduling strategy to obtain the scheduling tasks of the distribution network;

[0040] Convert the scheduling tasks into control instructions for the distribution network;

[0041] Send the control instructions to the control devices of the distribution network based on a preset communication network.

[0042] The present invention constructs an initial distribution network model based on structural data and device data, uses the graph theory method combined with the topological structure to accurately simulate the operating characteristics of the distribution network. At the same time, the attribute information is vector-transformed, and attribute information such as voltage level, load type, and device status is extracted. After numerical coding and feature combination, a feature vector is generated, representing the complex attribute information in a concise and computable vector form. An improved adjacency matrix is introduced for vector fusion. The construction of the improved adjacency matrix comprehensively considers dynamic factors such as the load change difference, environmental data difference, and device status similarity between nodes. The improved connection weight and degree calculation methods enable the matrix to more accurately reflect the actual association between nodes. The obtained local fusion features can more comprehensively and accurately reflect the state of nodes in the distribution network and their relationships with other nodes, enhancing the modeling ability of the complex relationships in the distribution network. The scheduling strategy formulated based on the load change trend and power supply capacity, combined with safety constraint conditions, can achieve the optimal allocation of power resources. Therefore, the present invention proposes a distribution network scheduling and control method based on situation awareness technology, which can solve the problem of low efficiency in distribution network scheduling and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 FIG. is a schematic flowchart of a distribution network scheduling and control method based on situation awareness technology provided by an embodiment of the present invention;

[0044] The realization, functional features, and advantages of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0046] An embodiment of the present application provides a distribution network scheduling and control method based on situation awareness technology. The execution subject of the distribution network scheduling and control method based on situation awareness technology includes, but is not limited to, at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiment of the present application. In other words, the distribution network scheduling and control method based on situation awareness technology can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to: a single server, a server cluster, a cloud server, or a cloud server cluster, etc. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, and artificial intelligence platforms.

[0047] Referring to Figure 1 as shown, it is a schematic flowchart of a distribution network scheduling and control method based on situation awareness technology provided by an embodiment of the present invention. In this embodiment, the distribution network scheduling and control method based on situation awareness technology includes:

[0048] S1. Collect multi-source data of the distribution network, where the multi-source data includes: structural data, environmental data, real-time operation data, and equipment data.

[0049] In the embodiment of the present invention, the collection of multi-source data of the distribution network includes:

[0050] Based on a distributed sensor network, collect data from the nodes of the distribution network to obtain the initial data of the distribution network;

[0051] Perform data cleaning on the initial data to obtain the cleaned data of the distribution network;

[0052] Perform data classification on the cleaned data to obtain the classified data of the distribution network, and determine the classified data as the multi-source data of the distribution network.

[0053] Specifically, use a sensor network distributed at each node position of the distribution network to collect data. These sensors are responsible for monitoring and recording various information at the nodes where they are located, and integrate the collected information to form an initial data set of the distribution network.

[0054] Specifically, the situation awareness technology is an advanced technology that integrates multi-source data to achieve comprehensive, real-time monitoring and accurate prediction of the operating state of complex systems, and plays a key role in the field of distribution network dispatching and control.

[0055] Specifically, install various sensors at key node positions such as transformers, poles, and distribution boxes in the distribution network, such as voltage sensors, current sensors, temperature sensors, humidity sensors, etc. Among them, the sensors transmit the collected data to the data collection terminal through wired or wireless communication technologies, such as optical fibers, ZigBee, 4G / 5G, etc., and the terminal summarizes and forms the initial data.

[0056] For example, the voltage sensor collects the voltage value of a certain node at a specific moment as 225V; the current sensor records the current of the same node at the same moment as 50A; the temperature sensor measures the surface temperature of the equipment at the node as 30°C.

[0057] Furthermore, the initial data may contain problems such as noise, incorrect data, and duplicate data. Data cleaning is to process these problem data, remove or correct inaccurate, incomplete, and duplicate data, and improve the data quality.

[0058] Specifically, a variety of data cleaning algorithms and techniques are adopted. Among them, for outliers, the 3σ criterion based on statistical methods is used for identification and processing. If a data point deviates from the mean by more than 3 times the standard deviation, it is determined as an outlier and corrected or deleted. For duplicate data, through the data comparison algorithm, duplicate records are found and deleted based on the unique identifier or feature combination of the data. For missing values, mean filling, median filling or prediction methods based on machine learning are used for filling.

[0059] For example, if the current data shows an obvious outlier of 1000A at a certain moment (the normal range is 0 - 100A), it is identified as an outlier using the 3σ criterion and may be corrected to a value near the mean of the current in this period. If there are multiple completely identical voltage data records, the duplicate records are deleted.

[0060] Furthermore, according to attributes such as the nature, source or use of the data, the cleaned data is divided into different categories, such as structural data, environmental data, real-time operation data, and equipment data, etc., so as to conduct targeted processing and analysis subsequently.

[0061] Specifically, data classification rules are formulated, such as classifying according to the physical meaning of the data and the monitoring object.

[0062] Specifically, through writing program scripts or using data management software, the cleaned data is automatically classified according to the classification rules.

[0063] For example: data such as node connection relationships, line parameters, lengths of each line in the distribution network, and conductor models are classified as structural data; real-time collected active power and reactive power data are classified as real-time operation data; data such as the model and manufacturing year of the equipment are classified as equipment data.

[0064] S2. Based on the structural data and the equipment data, construct the initial distribution network model of the distribution network, perform vector transformation on the attribute information corresponding to the nodes in the initial distribution network model, and obtain the feature vectors of the nodes.

[0065] In the embodiment of the present invention, the constructing the initial distribution network model of the distribution network based on the structural data and the equipment data includes:

[0066] Determine the nodes and connection relationships of the distribution network according to the structural data and the equipment data;

[0067] Generate the topological structure of the distribution network according to the nodes and the connection relationships;

[0068] Generate the initial distribution network model of the distribution network based on the graph theory method and the topological structure.

[0069] Specifically, the structural data contains the topological architecture information of the distribution network, such as line connection conditions, substation locations, etc.; the equipment data records the installation locations and connection methods of various power equipment (such as transformers, switches, etc.). By analyzing this data, the key positions in the distribution network can be identified as nodes, and the electrical connection relationships between the nodes can be determined.

[0070] Specifically, organize and analyze the structural data and equipment data to establish data associations.

[0071] Specifically, determine the connections of the nodes through the start and end point information of the lines; based on the access point information of the equipment, clarify the node location where the equipment is located and its connections with other nodes. This process can be completed using database queries and data matching algorithms.

[0072] For example: The structural data records that line L1 connects nodes N1 and N2, and the equipment data shows that transformer T1 is installed on node N1. In this way, the nodes N1, N2 and their connection relationship are determined, and at the same time, the association between transformer T1 and node N1 is clarified.

[0073] Specifically, after determining the nodes and connection relationships, represent this information in an intuitive way to form the topological structure of the distribution network. Among them, the topological structure describes the mutual connection methods and hierarchical relationships between the nodes, and is an abstract representation of the overall architecture of the distribution network.

[0074] Specifically, the topological structure can be represented in the form of a graph, where the nodes are represented by the vertices in the graph and the connection relationships are represented by the edges in the graph.

[0075] Furthermore, use relevant algorithms and data structures in graph theory, such as adjacency matrices, adjacency lists, etc., to store and represent the topological structure. These algorithms can be implemented through programming to convert the nodes and connection relationships into corresponding graph structures.

[0076] Specifically, assume there are nodes N1, N2, N3, and the connection relationships are N1 - N2, N2 - N3. When using an adjacency matrix to represent the topological structure, the elements in the matrix are assigned 1 (connected) or 0 (not connected) according to the connection conditions between the nodes.

[0077] Specifically, graph theory is a mathematical method used to study the properties and applications of graphs. Using the concepts and algorithms in graph theory, combined with the generated topological structure, model the distribution network. Through graph theory, the nodes and connection relationships of the distribution network can be mathematically described and analyzed, so as to construct an initial model that can reflect the operating characteristics of the distribution network.

[0078] Specifically, the physical characteristics of the distribution network (such as voltage, current, power, etc.) are associated with the nodes and edges of the graph to establish mathematical equations and models. Professional modeling software or programming languages (such as Python combined with the NetworkX library) can be used to implement graph theory algorithms and model construction.

[0079] For example: Use the Dijkstra algorithm to calculate the shortest path between two points in the distribution network to optimize the power transmission path. By establishing equations for node voltage and current, the electrical characteristics of the distribution network are described.

[0080] Generally speaking, the initial distribution network model generated based on graph theory and topological structure can accurately simulate the operating state and characteristics of the distribution network, provide effective model support for subsequent load forecasting, dispatching strategy formulation, etc., and improve the management and operation efficiency of the distribution network.

[0081] In the embodiment of the present invention, the vector transformation of the attribute information corresponding to the nodes in the initial distribution network model to obtain the feature vectors of the nodes includes:

[0082] Extract the attribute information of the nodes in the initial distribution network model, where the attribute information includes: voltage level, load type, and equipment status;

[0083] Perform numerical encoding on the attribute information to obtain the numerical features of the attribute information;

[0084] Perform feature combination on the numerical features to obtain the feature vectors of the nodes.

[0085] Specifically, identify and extract key attribute information related to the nodes from the initial distribution network model. The voltage level reflects the voltage level of the node in the distribution network; the load type (such as residential load, industrial load, etc.) reflects the load characteristics connected to the node; the equipment status (such as normal operation, failure, etc.) describes the working conditions of the equipment on the node.

[0086] Specifically, in the initial distribution network model, through data query and screening operations, the corresponding attribute information is extracted according to the identification information of the nodes, and database query statements or data processing programs can be used to complete this task.

[0087] For example: For node N1, extract its voltage level as 10 kV, load type as residential load, and equipment status as normal operation from the model.

[0088] Specifically, since the attribute information usually exists in text or classification form, in order to facilitate computer processing and model analysis, it is necessary to convert this attribute information into numerical form. Through numerical encoding, different attribute values can be mapped to specific numerical values to form the numerical features of the attribute information.

[0089] Specifically, different encoding methods are adopted for different types of attribute information. For the voltage level, its numerical value can be directly used; for the load type, one-hot encoding can be adopted, and each load type is represented by a binary vector, with only the corresponding position being 1 and the rest being 0; for the equipment status, 0 and 1 can be used to represent the normal and faulty states respectively.

[0090] Specifically, assume that there are three types of load: residential load, industrial load, and commercial load. The residential load is represented by [1, 0, 0], the industrial load is represented by [0, 1, 0], and the commercial load is represented by [0, 0, 1]. If the load type of node N1 is residential load, then its encoded numerical feature is [1, 0, 0].

[0091] Specifically, the numerical features of the encoded attribute information are combined together in a certain order to form a vector, which is the feature vector of the node. Among them, the feature vector can comprehensively reflect various attribute information of the node, facilitating unified processing and analysis in subsequent models.

[0092] Specifically, the numerical value of the voltage level, the vector after encoding the load type, the numerical value of the equipment status, etc. are arranged in sequence according to a pre-determined order to form a vector. For example, the voltage level is arranged first, then the encoded vector of the load type, and finally the numerical value of the equipment status.

[0093] Assume that the voltage level of node N1 is 10 kV, the load type encoding is [1, 0, 0], and the equipment status is normal (represented by 1), then its feature vector is [10, 1, 0, 0, 1].

[0094] Generally speaking, the generated node feature vector can comprehensively and concisely represent the attribute information of the node.

[0095] S3. Use the improved adjacency matrix to perform vector fusion on the feature vector to obtain the local fusion feature of the nodes in the distribution network.

[0096] In the embodiment of the present invention, the step of using the improved adjacency matrix to perform vector fusion on the feature vector to obtain the local fusion feature of the nodes in the distribution network includes:

[0097] Perform matrix multiplication on the feature vector and the improved adjacency matrix to obtain the local fusion feature of the nodes in the distribution network.

[0098] Specifically, perform matrix multiplication on the feature vector and the improved adjacency matrix. The construction of the improved adjacency matrix takes into account various factors between nodes, such as load change differences, environmental data differences, and equipment status similarities, etc., to more accurately reflect the association between nodes.

[0099] Specifically, the improved adjacency matrix is as follows:

[0100] ;

[0101] where is the improved adjacency matrix, is the degree matrix, is the adjacency matrix, is the node degree, indicates whether there is an edge connecting node and node , is the node and node improved connection weight between, represents all node sets adjacent to node , is the node and node dynamic influence function between, is the node and node load change difference between, is the node and node environmental data difference between, is the node and node device state similarity between, represents the node identifier of the th node, is the th node identifier of the node, is the constant coefficient for adjusting the adjacency matrix weight, is the total number of nodes, is the constant coefficient for adjusting the influence of load difference on degree.

[0102] Specifically, the degree matrix , where is the node degree. The degree matrix is a diagonal matrix, and the element on the diagonal represents the degree of node , and the non - diagonal elements are all 0.

[0103] Specifically, the degree reflects the connection tightness between a node and other nodes. The degree of node is calculated by the formula , where represents the node The set of all adjacent nodes, where the degree calculation not only considers the connection weights between nodes , but also introduces the load change difference and the adjustment coefficient , so that the degree of a node can more dynamically reflect its importance and connection situation in the distribution network.

[0104] Specifically, the adjacency matrix is used to represent the connection relationship between nodes.

[0105] Specifically, the elements of the traditional adjacency matrix usually represent whether there is an edge connecting node and node (such as represents there is a connection, represents there is no connection), while here the improved connection weight is used, where, is a constant coefficient for adjusting the weight of the adjacency matrix, used to adjust the overall weight of the adjacency matrix, is the node and node between the dynamic influence function, which comprehensively considers the node and node between the load change difference , environmental data difference and equipment status similarity , so that the connection weight between nodes can be adjusted according to these dynamic factors, more accurately reflecting the actual association between nodes.

[0106] Specifically, the improved adjacency matrix dynamically adjusts the connection weight between nodes through the dynamic influence function and corrects the node degree based on the load difference.

[0107] Specifically, assume there are 3 nodes in the distribution network, namely , the eigenvector of the node .

[0108] First, determine the element of the adjacency matrix : Assume node 1 is connected to node 2, node 2 is connected to node 3, and node 1 is not connected to node 3, then the traditional adjacency matrix , .

[0109] Then calculate the improved connection weight : Assume , then . Similarly, calculate the values of other , and obtain the adjacency matrix .

[0110] Next, calculate the degree matrix : Assume , for node 1, , then . Similarly calculate and , to obtain .

[0111] Finally, calculate the improved adjacency matrix .

[0112] Furthermore, multiply the eigenvector by the improved adjacency matrix , as , then , is the local fusion feature of the node.

[0113] Specifically, using the improved adjacency matrix to fuse the eigenvectors takes into account dynamic factors such as the load change difference, environmental data difference, and device state similarity between nodes, enabling the obtained local fusion feature to more comprehensively and accurately reflect the actual state of the node in the distribution network and its association with other nodes.

[0114] S4. Generate the time series features of the distribution network according to the environmental data and the real-time operation data, generate the dynamic weights of the local fusion features according to the time series features, and perform feature dimensionality reduction on the local fusion features according to the dynamic weights to obtain the dimensionality-reduced features of the nodes in the distribution network.

[0115] In the embodiment of the present invention, generating the dynamic weights of the local fusion features according to the time series features includes:

[0116] Determine the key influencing factors of the local fusion features based on the time series features;

[0117] Generate the dynamic weights of the local fusion features based on the key influencing factors.

[0118] Specifically, environmental data (such as temperature, humidity, light, etc.) and real-time operation data (such as voltage, current, power, etc.) are time-varying sequence data. By processing and analyzing these data, the time series features contained therein are extracted, and these features can reflect the operation state and change trend of the distribution network at different time points.

[0119] Specifically, time series analysis methods such as the sliding window method, autoregressive integrated moving average model (ARIMA), etc. can be adopted. For example, using the sliding window method, the environmental data and real-time operation data are divided with a certain time window (such as 1 hour, 24 hours, etc.), and the statistical features of the data within each window, such as the mean, variance, maximum value, minimum value, etc., are calculated as the time series features.

[0120] Specifically, assume that the temperature in the environmental data is recorded every hour, and the power in the real-time operation data is also recorded every hour. Using a 24-hour sliding window, calculate the mean of the temperature and the variance of the power within each window. These obtained statistical values are the time series features. For example, the mean temperature in the first 24-hour window is 25°C, and the power variance is 100 (kW²).

[0121] Specifically, after obtaining the time series features of the distribution network, analyze the correlation between these features and the local fusion features, and find out the factors that have a greater impact on the local fusion features. These key influencing factors can reveal the internal relationship between the operation state of the distribution network and the local fusion features, providing a basis for generating dynamic weights.

[0122] Specifically, methods such as correlation analysis and causal analysis can be used. For example, by calculating the Pearson correlation coefficient between the time series features and the local fusion features, find out the time series features with a relatively large absolute value of the correlation coefficient as the key influencing factors.

[0123] Specifically, through correlation analysis, it is found that the correlation coefficient between the time series feature of temperature and a certain component in the local fusion features is 0.8, the correlation coefficient between the time series feature of power and this component is 0.6, and the correlation coefficient between the time series feature of humidity and this component is 0.2. Then it can be considered that temperature and power are the key influencing factors of this local fusion feature.

[0124] Specifically, according to the determined key influencing factors, assign corresponding weights to each component of the local fusion features. These weights will be dynamically adjusted with the change of the key influencing factors to reflect the impact of the operation state of the distribution network at different times on the local fusion features.

[0125] Specifically, linear regression, machine learning algorithms (such as neural networks, decision trees, etc.) can be adopted to establish the mapping relationship between the key influencing factors and the dynamic weights. For example, using a linear regression model, taking the key influencing factors as the input and the weights of the local fusion features as the output, and obtaining the calculation formula of the weights through training the model.

[0126] Specifically, assume that the dynamic weights are obtained through linear regression related to temperature and power is , when the temperature at a certain moment ,power When the dynamic weight is calculated .

[0127] In detail, the generated dynamic weights can be dynamically adjusted according to the real-time operating status of the distribution network, so that the local fusion features can more accurately reflect the actual situation of the distribution network in subsequent processing, thereby improving the accuracy of the dispatching and control of the distribution network.

[0128] In an embodiment of the present invention, the step of performing feature dimensionality reduction on the local fusion feature according to the dynamic weight to obtain the dimensionality reduction feature of the node in the distribution network includes:

[0129] Multiplying the dynamic weight by the local fusion feature element by element to obtain a weighted local fusion feature;

[0130] The weighted local fusion features are subjected to dimensionality reduction processing based on a principal component analysis algorithm to obtain dimensionality reduction features of the nodes in the distribution network.

[0131] In detail, the generated dynamic weight is multiplied element-by-element with each component of the local fusion feature, and the local fusion feature is weighted, which can highlight the local fusion feature components corresponding to the key influencing factors, suppress unimportant components, and improve the effectiveness of the features.

[0132] In detail, if the local fusion feature is represented as a vector , the dynamic weight is represented as a vector , then the weighted local fusion feature , element-wise multiplication can be achieved using array operations in a programming language such as Python.

[0133] In detail, assuming that the local fusion feature , dynamic weight , then the weighted local fusion feature .

[0134] In detail, principal component analysis is a commonly used data dimensionality reduction method, which projects high-dimensional data into low-dimensional space by finding the principal components of the data, while retaining the main information of the data as much as possible. Performing PCA dimensionality reduction on the weighted local fusion features can reduce the dimension of the features, reduce the computational complexity, and retain the key information of the distribution network operation status.

[0135] Specifically, the weighted local fusion features are first standardized to make their mean 0 and variance 1. Then the covariance matrix of the data is calculated, and the eigenvalue decomposition of the covariance matrix is performed to obtain the eigenvalues and eigenvectors. The top feature with the largest eigenvalue is selected. The eigenvectors are used as the principal components, and the weighted local fusion features are projected onto these principal components to obtain the features after dimensionality reduction.

[0136] Specifically, assume that the weighted local fusion features are a 10-dimensional vector. After dimensionality reduction by PCA, the first 3 principal components are selected, and the 10-dimensional vector is projected onto these 3 principal components to obtain a 3-dimensional feature vector after dimensionality reduction.

[0137] S5. Predict the load change trend based on the features after dimensionality reduction, combine the safety constraint conditions of voltage limit and current limit, generate the scheduling strategy for the distribution network, and perform scheduling and control on the distribution network according to the scheduling strategy.

[0138] In the embodiment of the present invention, the predicting the load change trend based on the features after dimensionality reduction includes:

[0139] Input the features after dimensionality reduction into a pre-trained load change trend prediction model;

[0140] Use the load change trend prediction model to perform pattern recognition and trend inference on the features after dimensionality reduction to obtain the load change trend of the distribution network, where the load change trend includes: the rising, falling, and stable state information of the load at different time scales, the time when the load peak and valley values appear, and the corresponding load values.

[0141] Specifically, the features after dimensionality reduction obtained through the previous processing contain the key information of the operation state of the distribution network. These features after dimensionality reduction are provided as inputs to a pre-trained load change trend prediction model, and the learning ability and prediction function of the model are used to analyze the load change situation of the distribution network.

[0142] Specifically, in practical applications, a programming language (such as Python) can be used in combination with a machine learning or deep learning framework (such as TensorFlow, PyTorch, etc.) to implement the feature input operation, organize the features after dimensionality reduction in the format required by the model (such as matrix or tensor form), and then call the prediction interface of the model to input it.

[0143] Specifically, the load change trend prediction model will analyze and process the input features after dimensionality reduction, identify the patterns and rules contained therein. Through these patterns and rules, infer the load change trend of the distribution network at different time scales, including the rising, falling, and stable state information of the load, as well as the time when the load peak and valley values appear and the corresponding load values.

[0144] Specifically, the model can adopt various technologies, such as neural networks, time series analysis algorithms, etc. Taking neural networks as an example, during the training process, the model learns the mapping relationship between the dimensionality-reduced features and the load change trend. When the dimensionality-reduced features are input, the neural network calculates through forward propagation and outputs the predicted load change trend information.

[0145] Specifically, the model predicts that within the next 24 hours, the load is in an increasing state from 9 to 11 am and reaches a load peak of 5000 kW at 11 am; the load is in a decreasing state from 3 to 5 pm and reaches a load valley of 3000 kW at 5 pm; the load is relatively stable during the remaining time periods.

[0146] In the embodiment of the present invention, the method for generating the scheduling strategy is as follows: determining the load demand and power supply capacity of the distribution network in different time periods according to the load change trend;

[0147] According to the load demand and the power supply capacity, combined with the safety constraint conditions of voltage limitation and current limitation, generating the scheduling strategy of the distribution network. Specifically, according to the predicted load change trend, analyzing the load magnitude in each time period of the distribution network, that is, the load demand. At the same time, combined with the resource conditions such as the power generation equipment and transmission lines of the distribution network, determining the power supply capacity of the distribution network in different time periods.

[0148] Specifically, a physical model of the distribution network can be established, considering factors such as the rated power of the power generation equipment and the transmission capacity of the transmission lines, and calculating the power supply capacity in combination with the load change trend. For the load demand, directly extracting the load values in different time periods from the load change trend.

[0149] Specifically, according to the load change trend prediction, the load demand is 4500 - 5000 kW from 9 to 11 am, and at this time, the power supply capacity that the power generation equipment and transmission lines of the distribution network can provide is 5500 kW.

[0150] Specifically, the pre-obtained safety constraint conditions (voltage limitation and current limitation) are important parameters to ensure the safe operation of the distribution network. Combining the load demand and the power supply capacity, and formulating a reasonable scheduling strategy on the premise of meeting the safety constraint conditions to achieve the optimal allocation of power resources.

[0151] Specifically, optimization algorithms such as linear programming and integer programming can be adopted. Taking the load demand, power supply capacity and safety constraint conditions as constraint conditions, and taking the minimum power supply cost, load balance, etc. as the objective function, and obtaining the scheduling strategy by solving the optimization problem.

[0152] Specifically, within a certain time period, the load demand is 4000 kW, the power supply capacity is 5000 kW, the voltage limit is 10 kV ± 5%, and the current limit is 200 A. Through calculation by the optimization algorithm, it is necessary to allocate 2000 kW of power from Power Station A and 2000 kW of power from Power Station B to meet the load demand and ensure that the voltage and current are within the safe range.

[0153] In the embodiment of the present invention, the dispatching and control of the distribution network according to the dispatching strategy includes:

[0154] Performing task parsing on the dispatching strategy to obtain the dispatching tasks of the distribution network;

[0155] Performing instruction conversion on the dispatching tasks to obtain the control instructions of the distribution network;

[0156] Sending the control instructions to the control equipment of the distribution network based on a preset communication network.

[0157] Specifically, the dispatching strategy is usually a comprehensive plan and needs to be decomposed into specific and executable dispatching tasks. These tasks clarify what operations to perform on which equipment at what time to achieve the goals of the dispatching strategy.

[0158] Specifically, task parsing rules can be formulated to decompose according to dimensions such as equipment, time, and operation type based on the content of the dispatching strategy. For example, parsing the dispatching strategy "Adjust the output power of Substation A to 3000 kW at 10 am" into a specific dispatching task.

[0159] Specifically, the dispatching strategy is "Adjust the output power of 3 power stations according to the load change within the next 24 hours". After task parsing, specific dispatching tasks are obtained such as "Increase the output power of Power Station 1 by 500 kW at 9 am" and "Reduce the output power of Power Station 2 by 300 kW at 3 pm".

[0160] Specifically, the dispatching tasks are described in natural language or business logic and need to be converted into control instructions that can be recognized and executed by the distribution network control equipment. These instructions usually adopt specific communication protocols and formats.

[0161] Specifically, instruction conversion rules are formulated according to the type and communication protocol of the distribution network control equipment. For example, for control equipment using the Modbus protocol, convert the dispatching tasks into instructions conforming to the Modbus protocol format.

[0162] Specifically, the scheduling task is "closing Circuit Breaker A". After instruction conversion, a control instruction that conforms to the communication protocol is obtained, such as "01050001FF008C3A" (assuming it is a Modbus RTU protocol instruction).

[0163] Specifically, through a preset communication network (such as optical fiber communication, wireless communication, etc.), the generated control instruction is transmitted to the control devices of the distribution network (such as circuit breakers, transformer tap controllers, etc.). After receiving the instruction, the control devices perform corresponding operations to achieve the scheduling and control of the distribution network.

[0164] Specifically, communication links are established using communication devices (such as communication gateways, routers, etc.), and the control instructions are encapsulated and transmitted according to the requirements of the communication protocol. During the transmission process, the reliability and security of the data need to be considered, such as using encryption technology, retransmission mechanisms, etc.

[0165] Specifically, the control instruction is sent to the control device of a substation at a relatively long distance through an optical fiber communication network. After receiving the instruction, the control device adjusts the tap position of the transformer according to the instruction requirements.

[0166] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms.

[0167] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence is the theory, method, and technology that uses a digital computer or a machine controlled by a digital computer to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results.

[0168] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for dispatching and controlling a distribution network based on situation awareness technology, characterized in that The method includes: Collecting multi-source data of the distribution network, where the multi-source data includes: structure data, environmental data, real-time operation data, and equipment data; Constructing an initial distribution network model of the distribution network based on the structure data and the equipment data, performing vector transformation on the attribute information corresponding to the nodes in the initial distribution network model to obtain the feature vectors of the nodes; Performing matrix multiplication on the feature vectors and an improved adjacency matrix to obtain the local fusion features of the nodes in the distribution network, where the improved adjacency matrix is: ； Among them, is the improved adjacency matrix, is the degree matrix, is the adjacency matrix, is the node degree, indicates whether there is an edge connecting node and node , is the improved connection weight between node and node , represents the set of all nodes adjacent to node , is the dynamic influence function between node and node , is the load change difference between node and node , is the environmental data difference between node and node , is the equipment state similarity between node and node , represents the node identifier of the th node, is the node identifier of the th node, is the constant coefficient for adjusting the weight of the adjacency matrix, is the total number of nodes, is the constant coefficient for adjusting the influence of the load difference on the degree. Among them, the improved adjacency matrix dynamically adjusts the connection weights between nodes through the dynamic influence function and corrects the node degrees based on the load difference; Generating time-series features of the distribution network according to the environmental data and the real-time operation data, generating dynamic weights of the local fusion features according to the time-series features, and performing feature dimensionality reduction on the local fusion features according to the dynamic weights to obtain the dimensionality-reduced features of the nodes in the distribution network; Predicting the load change trend according to the dimensionality-reduced features, combining the safety constraint conditions of voltage limit and current limit, generating a scheduling strategy for the distribution network, and performing scheduling and control on the distribution network according to the scheduling strategy.

2. The method for dispatching and controlling a distribution network of the situation awareness technology according to claim 1, wherein, The collecting multi-source data of the distribution network includes: Collecting data of the nodes of the distribution network based on a distributed sensor network to obtain the initial data of the distribution network; Performing data cleaning on the initial data to obtain the cleaned data of the distribution network; Performing data classification on the cleaned data to obtain the classified data of the distribution network, and determining the classified data as the multi-source data of the distribution network.

3. The distribution network dispatching and control method for situation awareness technology according to claim 1, characterized in that, The constructing an initial distribution network model of the distribution network based on the structure data and the equipment data includes: Determining the nodes and connection relationships of the distribution network according to the structure data and the equipment data; Generating the topological structure of the distribution network according to the nodes and the connection relationships; Generating an initial distribution network model of the distribution network based on graph theory and the topological structure.

4. The method for dispatching and controlling a distribution network of a situation awareness technology according to claim 1, wherein, The performing vector transformation on the attribute information corresponding to the nodes in the initial distribution network model to obtain the feature vectors of the nodes includes: Extracting the attribute information of the nodes in the initial distribution network model, where the attribute information includes: voltage level, load type, and equipment status; Performing numerical encoding on the attribute information to obtain the numerical features of the attribute information; Performing feature combination on the numerical features to obtain the feature vectors of the nodes.

5. The method for dispatching and controlling a distribution network of the situation awareness technology according to claim 1, characterized in that The generating dynamic weights of the local fusion features according to the time-series features includes: Determining the key influencing factors of the local fusion features based on the time-series features; Generating dynamic weights of the local fusion features based on the key influencing factors.

6. The distribution network dispatching and control method of the situation awareness technology according to claim 1, characterized in that The performing feature dimensionality reduction on the local fusion features according to the dynamic weights to obtain the dimensionality-reduced features of the nodes in the distribution network includes: Multiplying the dynamic weights and the local fusion features element by element to obtain the weighted local fusion features; Performing dimensionality reduction processing on the weighted local fusion features based on the principal component analysis algorithm to obtain the dimensionality-reduced features of the nodes in the distribution network.

7. The distribution network dispatching and control method for situation awareness technology according to claim 1, characterized in that, The predicting the load change trend according to the dimensionality-reduced features includes: Inputting the dimensionality-reduced features into a pre-trained load change trend prediction model; Performing pattern recognition and trend inference on the dimension-reduced features by using the load change trend prediction model to obtain the load change trend of the distribution network, where the load change trend includes: the rising, falling, and steady state information of the load at different time scales, the time when the load peak and valley values appear, and the corresponding load values.

8. The distribution network scheduling and control method of the situation awareness technology according to any one of claims 1 to 7, characterized in that, The method for generating the scheduling strategy is as follows: determining the load demand and power supply capacity of the distribution network in different time periods according to the load change trend; Generating the scheduling strategy of the distribution network according to the load demand and the power supply capacity, in combination with the safety constraint conditions of voltage limit and current limit.

9. The distribution network dispatching and control method of the situation awareness technology according to claim 1, characterized in that The scheduling and control of the distribution network according to the scheduling strategy includes: Performing task parsing on the scheduling strategy to obtain the scheduling tasks of the distribution network; Performing instruction conversion on the scheduling tasks to obtain the control instructions of the distribution network; Sending the control instructions to the control equipment of the distribution network based on a preset communication network.

Images