Low-voltage power distribution network fluctuation risk research and intelligent early warning method and system

Abstract

The application provides a low-voltage power distribution network change risk judgment and intelligent early warning method and system, relates to the technical field of power system intelligent early warning, and comprises the following steps: fusing and analyzing power grid operation data and environment data through a space-time convolutional neural network to generate a correlation characteristic matrix of power grid operation state and environment risk; dynamically modeling a power grid topology structure by using a physical information dynamic graph attention network to extract a dynamic risk propagation path; combining the dynamic risk propagation path, and calculating a fault probability distribution of each node of the power grid by using a Bayesian network; optimizing an early warning strategy by using a reinforcement learning algorithm to generate an optimal early warning scheme; and scheduling repair resources by using a multi-objective optimization algorithm to generate a repair resource scheduling scheme.The application has the beneficial effect of significantly improving the accuracy of fault early warning and the efficiency of repair, and provides reliable technical support for intelligent operation and maintenance of a low-voltage power distribution network.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of intelligent early warning of power systems, in particular to a low-voltage distribution network change risk judgment and intelligent early warning method and system. BACKGROUND

[0002] In recent years, with the continuous expansion of the scale and the increase of the complexity of the power system, the fault diagnosis and operation and maintenance management of the low-voltage distribution network are facing severe challenges. Especially in remote mountainous areas, coastal areas and other multi-disaster environments, power grid equipment is easily affected by extreme weather such as typhoons, heavy rains, salt fog corrosion, etc., resulting in frequent faults and difficulty in rapid positioning and repair.

[0003] At present, the fault early warning of the low-voltage distribution network mainly relies on manual inspection and simple threshold alarm systems, and these methods have problems such as low efficiency, poor real-time performance, and insufficient adaptability. For example, traditional manual inspection cannot capture the influence of environmental changes on the power grid in real time, and the alarm system based on fixed thresholds is difficult to cope with complex and variable fault modes. In the prior art, some research attempts to make fault prediction through statistical models or rule engines, but these methods lack the modeling capability of the dynamic changes of the power grid topology and the physical laws, resulting in low prediction accuracy and high false alarm rate, which cannot meet the demand of modern power grid intelligent operation and maintenance. SUMMARY

[0004] The purpose of the present application is to provide a low-voltage distribution network change risk judgment and intelligent early warning method and system to improve the above problems. In order to achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0005] In a first aspect, the application provides a low-voltage distribution network change risk judgment and intelligent early warning method, comprising:

[0006] fusing and analyzing the power grid operation data and the environmental data through a spatio-temporal convolutional neural network to generate a correlation feature matrix of the power grid operation state and the environmental risk;

[0007] based on the correlation feature matrix, dynamically modeling the power grid topology structure by using a physical information dynamic graph attention network, and extracting a dynamic risk propagation path;

[0008] combining the dynamic risk propagation path, calculating the fault probability distribution of each node of the power grid by using a Bayesian network;

[0009] based on the fault probability distribution, optimizing the early warning strategy by using a reinforcement learning algorithm to generate an optimal early warning scheme;

[0010] according to the optimal early warning scheme, scheduling the repair resources by using a multi-objective optimization algorithm to generate a repair resource scheduling scheme.

[0011] Preferably, the power grid operation data and environmental data are fused and analyzed by the spatio-temporal convolutional neural network to generate a correlation feature matrix of the power grid operation state and environmental risk, which includes:

[0012] The first data of voltage, current and power factor with a frequency of 1 minute / second are collected in real time by smart meters and sensors, the second data of wind speed, rainfall, temperature and humidity with a frequency of 10 minutes / second are obtained from a meteorological station, the third data of terrain elevation and vegetation distribution are obtained from a geographic information system, and the fourth data of salt spray concentration with a sampling frequency of 1 hour / second are obtained from corrosion monitoring equipment;

[0013] The first data, the second data, the third data and the fourth data are aligned according to time and space, and a spatio-temporal data cube is constructed, and the spatio-temporal data cube is processed by a spatio-temporal convolutional neural network to obtain a correlation feature matrix of the power grid operation state and environmental risk.

[0014] Preferably, based on the correlation feature matrix, a physical information dynamic graph attention network is used to dynamically model the power grid topology structure and extract a dynamic risk propagation path, which includes:

[0015] The topological structure information of the power grid in the power grid management system is obtained, including node data and edge data; the correlation feature matrix is assigned as a node to the topological structure information of the power grid to obtain a power grid topology structure with node features;

[0016] The dynamic graph structure is constructed according to the power grid topology structure with node features, the second data and the fourth data;

[0017] Based on the dynamic graph structure and the power grid operation data including current, voltage and resistance, and taking the law of electrical circuits as a constraint condition, a physical information loss function is constructed, the physical information loss function is integrated into the neural network training process, and a node feature embedded with physical information is obtained;

[0018] The attention coefficients between nodes are calculated by using a dynamic graph attention network algorithm in combination with the node feature embedded with physical information and the dynamic graph structure, and the edge weights are dynamically adjusted according to the attention coefficients;

[0019] Based on the edge weights and the node features, the node state is iteratively updated by a multi-layer PI-DGAT to simulate the dynamic propagation process of the risk in the power grid, and a dynamic risk propagation path is obtained.

[0020] Preferably, in combination with the dynamic risk propagation path, a Bayesian network is used to calculate the fault probability distribution of each node of the power grid, which includes:

[0021] The Bayesian network structure is constructed by using a dynamic risk propagation path matrix, historical fault data and real-time device state data, wherein network nodes correspond to power grid devices, network edges represent the risk propagation relationship between devices, and the edge weight is determined by the propagation probability in the dynamic risk propagation path matrix, so as to generate a Bayesian network topology structure reflecting the dynamic risk propagation logic between power grid devices;

[0022] Combined with the Bayesian network topology structure and historical fault data, the basic fault probability of the root node and the conditional fault probability of the non-root node are calculated by training the conditional probability table, wherein the conditional probability of the non-root node is derived from the state of its parent node and the dynamic propagation probability, so as to obtain the conditional probability distribution of the Bayesian network;

[0023] Real-time device state data is input into the trained Bayesian network as evidence, and a probability inference algorithm is used to calculate the posterior probability of each node under different fault types, and the fault probability distribution of the power grid node is output.

[0024] Preferably, the early warning strategy is optimized by the reinforcement learning algorithm to generate an optimal early warning scheme, which includes:

[0025] The fault probability distribution and user feedback data are obtained, and after integration processing, a state space is obtained, which is used as the input of the reinforcement learning algorithm;

[0026] The state space and action space are used to obtain a Q value table through Q-learning algorithm training processing, wherein the Q value table is used to generate an optimal early warning strategy;

[0027] Based on the Q value table and the current state, the optimal action is selected according to the Q value table, and then the optimal early warning scheme is obtained, wherein the optimal early warning scheme includes an early warning level and a response measure.

[0028] In a second aspect, the application also provides a low-voltage distribution network variable risk research and intelligent early warning system, which comprises:

[0029] A generation module is configured to fuse and analyze power grid operation data and environmental data by a spatio-temporal convolutional neural network to generate an associated feature matrix of power grid operation state and environmental risk;

[0030] An extraction module is configured to dynamically model a power grid topology structure by using a physical information dynamic graph attention network based on the associated feature matrix to extract a dynamic risk propagation path;

[0031] A calculation module is configured to calculate a fault probability distribution of each node of the power grid by using a Bayesian network in combination with the dynamic risk propagation path;

[0032] An optimization module is configured to optimize an early warning strategy by a reinforcement learning algorithm based on the fault probability distribution to generate an optimal early warning scheme;

[0033] The scheduling early warning module is used for scheduling the repair resources according to the optimal early warning scheme by using a multi-objective optimization algorithm.

[0034] In a third aspect, the application further provides a low-voltage power distribution network change risk judgment and intelligent early warning device, comprising:

[0035] The memory is used for storing the computer program.

[0036] The processor is used for executing the computer program to realize the steps of the low-voltage power distribution network change risk judgment and intelligent early warning method.

[0037] In a fourth aspect, the application further provides a readable storage medium, wherein the readable storage medium stores a computer program, and the computer program is executed by a processor to realize the steps of the low-voltage power distribution network change risk judgment and intelligent early warning method.

[0038] The application has the following beneficial effects:

[0039] The application extracts the correlation features of power grid operation state and environmental risk by using a space-time convolutional neural network, models the dynamic changes of power grid topology by combining a physical information dynamic graph attention network, captures the influence of environmental factors (such as typhoon, rainstorm, and salt fog corrosion) on the power grid in real time, and calculates the fault probability distribution of each node through a Bayesian network, so that the position and type of fault occurrence can be accurately predicted, and the problems that the traditional method cannot dynamically adapt to environmental changes and lacks physical law constraints are solved.

[0040] The application integrates circuit laws (such as Ohm's law and Kirchhoff's law) as constraint conditions into the neural network training process, constructs a physical information loss function, ensures that the model output conforms to the basic operation law of the power grid, improves the prediction accuracy of the model, enhances the interpretability of the results, and enables operation and maintenance personnel to intuitively understand the logic and root cause of fault propagation.

[0041] The application optimizes the early warning strategy by using a reinforcement learning algorithm, designs a reward function by combining user feedback data (such as power outage influence range and key user importance), generates an optimal early warning scheme and repair resource scheduling scheme, and based on the repair scheduling scheme of the multi-objective optimization algorithm, the resource utilization rate and risk coverage range can be maximized while ensuring the shortest repair time, which significantly improves the power supply reliability and operation and maintenance efficiency of the power grid.

[0042] Other features and advantages of the application will be described in the following description, and some will become apparent from the description, or will be understood from the practice of the application. The purpose and other advantages of the application can be achieved and obtained by the structures specifically pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be considered as limiting the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0044] Figure 1 The low-voltage power distribution network change risk judgment and intelligent early warning method process schematic diagram described in the embodiments of the present application;

[0045] Figure 2 The low-voltage power distribution network change risk judgment and intelligent early warning system structure schematic diagram described in the embodiments of the present application;

[0046] Figure 3 The low-voltage power distribution network change risk judgment and intelligent early warning device structure schematic diagram described in the embodiments of the present application.

[0047] In the figure: 701, generation module; 702, extraction module; 703, calculation module; 704, optimization module; 705, scheduling early warning module; 800, low-voltage power distribution network change risk judgment and intelligent early warning device; 801, processor; 802, memory; 803, multimedia assembly; 804, I / O interface; 805, communication assembly. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clear, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0049] It should be noted that: similar labels and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present application, the terms "first", "second" and the like are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0050] Embodiment 1:

[0051] The embodiment provides a low-voltage power distribution network change risk research and intelligent early warning method.

[0052] Referring to Figure 1 , the method comprises steps S100, S200, S300, S400 and S500.

[0053] S100, the power grid operation data and environmental data are analyzed by a space-time convolutional neural network to generate a correlation feature matrix of the power grid operation state and environmental risk.

[0054] It can be understood that the step S100 includes S101 and S102.

[0055] S101, the first data of voltage, current and power factor with a frequency of 1 minute / time are collected in real time by a smart meter and a sensor, the second data of wind speed, rainfall, temperature and humidity with a frequency of 10 minutes / time are obtained from a meteorological station, the third data of terrain elevation and vegetation distribution are obtained from a geographic information system, and the fourth data of salt fog concentration with a sampling frequency of 1 hour / time are obtained from corrosion monitoring equipment;

[0056] S102, the first data, the second data, the third data and the fourth data are aligned according to time and space, and a space-time data cube is constructed, the space-time data cube is processed by using a space-time convolutional neural network, and a correlation feature matrix of the power grid operation state and environmental risk is obtained, and the calculation formula of the space-time data cube is as follows:

[0057]

[0058] Wherein, X represents the fusion result of the power grid operation data and the environmental data, Each element in the data cube is a real number, N is the number of power grid nodes, T is the time step, represents the number of time points of data collection, and D represents the number of features of each node at each time point.

[0059] It should be noted that a 3x3 convolution kernel is used to extract local spatial features, and the formula is as follows:

[0060] F space =Conv2D(X,K space )

[0061] In the formula, K space is a spatial convolution kernel, and F space is a spatial feature map.

[0062] A 1D convolution kernel is used to capture dynamic changes, and an associated feature vector of each power grid node is output wherein D' is the feature dimension.

[0063] wherein the correlation feature matrix of the power grid operating state and the environmental risk is F, the rows of the matrix represent the power grid nodes, and the columns represent the feature dimensions (such as a voltage fluctuation-wind speed correlation index, a cable aging-salt mist correlation index).

[0064] S200, based on the correlation feature matrix, a physical information dynamic graph attention network is used to dynamically model the power grid topology structure, and a dynamic risk propagation path is extracted.

[0065] It can be understood that the present step S200 includes S201, S202, S203, S204 and S205.

[0066] S201, acquiring topology structure information of the power grid in the power grid management system, including node data and edge data; the correlation feature matrix is assigned as the node to the topology structure information of the power grid, and the topology structure of the power grid with node features is obtained;

[0067] It should be noted that the node data represents the devices (such as transformers, cable joints) in the power grid, and each node contains device ID, type, location and other information; the edge data represents the connection relationship between the devices (such as cable connection), and each edge contains the starting node, the ending node, the connection type (such as cable type), the length and other information. The node data V and the edge data E are exported from the power grid management system through an API interface or a database query. The features of the environmental risk and the power grid operating state are fused into the topology structure, providing multi-dimensional input for dynamic modeling.

[0068] S202, constructing a dynamic graph structure according to the power grid topology structure with node features, the second data and the fourth data;

[0069] It can be understood that in the present embodiment, the edge weight E(t) of the dynamic graph structure G(t) = (V, E(t)) is driven by environmental data:

[0070] E ij (t) = a · W(t) + β · R(t) + γ · S(t)

[0071] In the formula, a, β, γ are weight coefficients, which are obtained by training historical data. For example, high wind speed W(t) can cause cable swing to intensify, and the edge weight decreases. Among them, the short-time (minute-level) edge weight reflects real-time environmental changes, and the long-time (hour-level) edge weight reflects the cumulative effect of device aging. In the present step, the dynamic edge weight enables the model to respond to environmental changes in real time, and improves the accuracy of the risk propagation path.

[0072] S203, based on the dynamic graph structure and power grid operation data, wherein the power grid operation data includes current, voltage and resistance, and the circuit law is taken as a constraint condition, a physical information loss function is constructed, the physical information loss function is integrated into a neural network training process, and a physical information embedded node feature is obtained, and a calculation formula thereof is as follows:

[0073]

[0074] In the formula, The physical information loss function is I i The current of node i is V i The voltage of node i is R i The resistance of node i is The neighbor set of node j is represented.

[0075] It should be noted that when the historical fault data is insufficient, the physical constraint can prevent the model from overfitting, improve the prediction ability of rare faults (such as multi-point grounding caused by salt mist corrosion), and through the physical loss term, the risk propagation path output by the model can be associated with specific physical quantities (such as current anomalies), so that the operation and maintenance personnel can locate the root cause, ensure that the risk propagation model conforms to the basic operation law of the power grid, and avoid the prediction results that violate the physical law.

[0076] S204, combining the physical information embedded node feature and the dynamic graph structure, using a dynamic graph attention network algorithm to calculate the attention coefficient between nodes, and dynamically adjusting the edge weight according to the attention coefficient, distinguishing the intensity of risk propagation between different nodes through the attention mechanism, and avoiding the error caused by the uniform propagation assumption.

[0077] S205, based on the edge weight and the node feature, the node state is iteratively updated through a multi-layer PI-DGAT, the dynamic propagation process of the risk in the power grid is simulated, and a dynamic risk propagation path is obtained.

[0078] It should be noted that the risk propagation is realized through the multi-layer PI-DGAT, and the node state update formula is as follows:

[0079]

[0080] In the formula, The state of node i at time t is The neighbor set of node i is represented, and ij The attention coefficient of node i and node j at time t is represented, and W (l) The weight matrix of the lth layer is represented, The feature representation of node j at the lth layer is represented.

[0081] Therefore, step S200 realizes the fine description of the risk propagation of the low-voltage distribution network by dynamic graph structure modeling, physical information embedding, attention mechanism and typhoon data fusion, and combines the environmental factors, physical laws and deep learning organically, which not only guarantees the physical rationality of the model, but also improves the prediction accuracy under complex environment.

[0082] S300, combining the dynamic risk propagation path, the failure probability distribution of each node of the power grid is calculated by using the Bayesian network.

[0083] It can be understood that steps S300 include S301, S302 and S303.

[0084] S301, the dynamic risk propagation path matrix, historical failure data and real-time device state data are used to construct the Bayesian network structure, wherein the network nodes correspond to the power grid devices, the network edges represent the risk propagation relationship between the devices, and the edge weight is determined by the propagation probability in the dynamic risk propagation path matrix. The Bayesian network topology structure reflecting the dynamic risk propagation logic between the power grid devices is generated; wherein the dynamic propagation path matrix is:

[0085] P ij = the risk propagation probability from node i to node j

[0086] It is generated by a spatio-temporal convolutional neural network (ST-CNN) and a dynamic graph attention network (DGAT), and reflects the real-time influence of environmental factors (wind speed, salt spray concentration) on risk propagation.

[0087] S302, combining the Bayesian network topology structure and the historical failure data, the basic failure probability of the root node and the conditional failure probability of the non-root node are calculated by training the conditional probability table, wherein the conditional probability of the non-root node is derived from the state of its parent node and the dynamic propagation probability, and the conditional probability distribution of the Bayesian network is obtained; the conditional probability formula of the non-root node is:

[0088]

[0089] xi∈{0,1}: state of parent node (0 normal, 1 failure), is the propagation probability from parent node k i to current node j.

[0090] S303, the real-time device state data is input as evidence into the trained Bayesian network, and the posterior probability of each node under different failure types is calculated by using the probability inference algorithm, and the failure probability distribution of the power grid node is output.

[0091] ​It should be noted that, based on the Bayesian network topology and historical fault data, a probability table (CPT) is used as the training condition. The fault probability of the root node is obtained from the statistical analysis of historical fault data, while the fault probability of non-root nodes is determined by the parent node state and the dynamic propagation probability P. ij Joint calculations are performed to obtain the conditional probability table of the trained Bayesian network, which describes the fault propagation relationship between nodes. Finally, using the trained Bayesian network and real-time equipment status data, probabilistic inference is performed to output the fault probability distribution of each node in the power grid, providing a basis for fault early warning and emergency repair decisions. The fault probability distribution matrix is ​​as follows:

[0092] Q i，c =P(X) i =1|Fault type=c)

[0093] It can be further correlated with the degree of impact of the fault (such as the number of users affected by power outages and economic losses) to generate priority-ranked emergency repair plans.

[0094] S400: Based on the fault probability distribution, the early warning strategy is optimized through reinforcement learning algorithm to generate the optimal early warning scheme.

[0095] It is understood that step S400 includes S401, S402, and S403, wherein:

[0096] S401. Obtain the fault probability distribution and user feedback data, integrate and process them to obtain the state space, and use the state space as the input of the reinforcement learning algorithm.

[0097] S402. Using the state space and action space, the Q-learning algorithm is used to train and process the Q-value table, which is used to generate the optimal early warning strategy.

[0098] S403. Based on the Q-value table and the current state, select the optimal action according to the Q-value table, and then obtain the optimal early warning plan. The optimal early warning plan includes the early warning level and the corresponding countermeasures. The calculation formula is as follows:

[0099]

[0100] In the formula, For the optimal action, Q(s) t , a) is the state-action value function.

[0101] S500: Based on the optimal early warning scheme, use a multi-objective optimization algorithm to schedule emergency repair resources and generate an emergency repair resource scheduling scheme.

[0102] Understandably, in this step, emergency repair resource data and geographic information data are acquired, integrated, and processed to obtain input data for the emergency repair resource scheduling problem. Using the input data and optimal early warning scheme of the emergency repair resource scheduling problem, a Pareto optimal solution set is obtained through multi-objective optimization algorithm (such as NSGA-II). Based on the Pareto optimal solution set and user feedback data, the optimal emergency repair resource scheduling scheme is obtained through scheme selection processing, including the scheduling plan and route planning of emergency repair personnel, equipment, and vehicles.

[0103] Example 2:

[0104] like Figure 2 As shown, this embodiment provides a low-voltage distribution network change risk assessment and intelligent early warning system. See [link to documentation]. Figure 2 The system includes:

[0105] Generation module 701: Used to perform fusion analysis on power grid operation data and environmental data through spatiotemporal convolutional neural networks to generate a correlation feature matrix between power grid operation status and environmental risk;

[0106] Extraction module 702: Used to dynamically model the power grid topology based on the correlation feature matrix and extract dynamic risk propagation paths using a physical information dynamic graph attention network;

[0107] Calculation module 703: Used to combine dynamic risk propagation paths and use Bayesian networks to calculate the fault probability distribution of each node in the power grid;

[0108] Optimization module 704: Used to optimize the early warning strategy based on the fault probability distribution and generate the optimal early warning scheme through reinforcement learning algorithm;

[0109] Dispatch and early warning module 705: Used to schedule emergency repair resources based on the optimal early warning scheme and using a multi-objective optimization algorithm.

[0110] Specifically, the generation module 701 includes:

[0111] The data acquisition unit is used to collect first data such as voltage, current and power factor at a frequency of 1 minute / time through smart meters and sensors, second data such as wind speed, rainfall, temperature and humidity at a frequency of 10 minutes / time from meteorological stations, third data such as terrain elevation and vegetation distribution at geographic information systems, and fourth data such as salt spray concentration at a sampling frequency of 1 hour / time from corrosion monitoring equipment.

[0112] The first construction unit is configured to align the first data, the second data, the third data and the fourth data in time and space, construct a spatio-temporal data cube, process the spatio-temporal data cube by using a spatio-temporal convolutional neural network, and obtain a correlation feature matrix of the power grid operation state and the environmental risk. The calculation formula of the spatio-temporal data cube is as follows:

[0113]

[0114] wherein X represents a fusion result of the power grid operation data and the environmental data, each element in the data cube is a real number, N is the number of power grid nodes, T is a time step, D represents the number of feature quantities of each node at each time point.

[0115] Specifically, the extraction module 702 comprises:

[0116] The first acquisition unit is configured to acquire topological structure information of a power grid in a power grid management system, wherein the topological structure information comprises node data and edge data; and assign the correlation feature matrix as node values to the topological structure information of the power grid to obtain a power grid topological structure with node features.

[0117] The second construction unit is configured to construct a dynamic graph structure according to the power grid topological structure with node features, the second data and the fourth data.

[0118] The third construction unit is configured to construct a physical information loss function based on the dynamic graph structure and the power grid operation data, wherein the power grid operation data comprises current, voltage and resistance, and the circuit law is taken as a constraint condition, and the physical information loss function is integrated into a neural network training process to obtain node features with physical information embedding, and the calculation formula is as follows:

[0119]

[0120] In the formula, denotes the physical information loss function, I i denotes the current of node i, V i denotes the voltage of node i, R i denotes the resistance of node i, denotes a neighbor set of node j;

[0121] The first calculation unit is configured to combine the node features with physical information embedding and the dynamic graph structure, calculate attention coefficients between nodes by using a dynamic graph attention network algorithm, and dynamically adjust edge weights according to the attention coefficients.

[0122] The simulation unit is configured to simulate a dynamic propagation process of the risk in the power grid based on the edge weight and the node feature by iteratively updating a node state through the multi-layer PI-DGAT, and obtain a dynamic risk propagation path.

[0123] Specifically, the computing module 703 includes:

[0124] The fourth constructing unit is configured to construct a Bayesian network structure by using the dynamic risk propagation path matrix, the historical fault data and the real-time device state data, wherein a network node corresponds to a device in the power grid, a network edge represents a risk propagation relationship between devices, and an edge weight is determined by a propagation probability in the dynamic risk propagation path matrix, so as to generate a Bayesian network topology structure reflecting a dynamic risk propagation logic between devices in the power grid.

[0125] The second computing unit is configured to calculate a basic fault probability of a root node and a conditional fault probability of a non-root node by training a conditional probability table in combination with the Bayesian network topology structure and the historical fault data, wherein the conditional probability of the non-root node is derived from a state of a parent node and a dynamic propagation probability, so as to obtain a conditional probability distribution of the Bayesian network.

[0126] The third computing unit is configured to input the real-time device state data as evidence into the trained Bayesian network, and calculate a posterior probability of each node under different fault types by using a probability inference algorithm, so as to output a fault probability distribution of the nodes in the power grid.

[0127] Specifically, the optimization module 704 includes:

[0128] The second obtaining unit is configured to obtain the fault probability distribution and user feedback data, integrate and process the fault probability distribution and the user feedback data, and obtain a state space, wherein the state space is used as an input of a reinforcement learning algorithm.

[0129] The processing unit is configured to obtain a Q value table by training and processing the state space and an action space through a Q-learning algorithm, wherein the Q value table is used to generate an optimal early warning strategy.

[0130] The obtaining unit is configured to select an optimal action according to the Q value table based on the Q value table and a current state, and obtain an optimal early warning scheme, wherein the optimal early warning scheme includes an early warning level and a response measure, and a calculation formula of the optimal early warning scheme is as follows:

[0131]

[0132] In the formula, s is the current state, a is the optimal action, and Q(s is the optimal action, and Q(s t is a state-action value function.

[0133] It should be noted that the specific manner in which the various modules perform operations in the system of the above embodiments has been described in detail in the embodiments related to the method, and will not be described in detail here.

[0134] Embodiment 3:

[0135] Corresponding to the above method embodiment, the present embodiment also provides a low-voltage power distribution network change risk judgment and intelligent early warning device. The low-voltage power distribution network change risk judgment and intelligent early warning device described below can be correspondingly referred to with the low-voltage power distribution network change risk judgment and intelligent early warning method described above.

[0136] Figure 3 Fig. 8 is a block diagram of a low-voltage power distribution network change risk judgment and intelligent early warning device 800 according to an example embodiment. As shown in Fig. 8, the low-voltage power distribution network change risk judgment and intelligent early warning device 800 includes a processor 801 and a memory 802. The low-voltage power distribution network change risk judgment and intelligent early warning device 800 also includes one or more of a multimedia component 803, an I / O interface 804, and a communication component 805. Figure 3

[0137] ​The processor 801 is configured to control the overall operation of the low-voltage power distribution network change risk judgment and intelligent early warning device 800 to complete all or part of the steps in the low-voltage power distribution network change risk judgment and intelligent early warning method described above. The memory 802 is configured to store various types of data to support the operation of the low-voltage power distribution network change risk judgment and intelligent early warning device 800. For example, the data can include instructions for any application or method operating on the low-voltage power distribution network change risk judgment and intelligent early warning device 800, and application-related data, such as contact data, sent and received messages, pictures, audio, video, and the like. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The multimedia component 803 can include a screen and an audio component. The screen can be a touch screen, for example, and the audio component is configured to output and / or input audio signals. For example, the audio component can include a microphone configured to receive external audio signals. The received audio signals can be further stored in the memory 802 or transmitted through the communication component 805. The audio component also includes at least one speaker configured to output audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, which can be a keyboard, mouse, or button, etc. The buttons can be virtual buttons or physical buttons. The communication component 805 is configured to perform wired or wireless communication between the low-voltage power distribution network change risk judgment and intelligent early warning device 800 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, near field communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 805 can include a Wi-Fi module, a Bluetooth module or an NFC module.

[0138] In an example embodiment, the low-voltage power distribution network change risk judgment and intelligent early warning device 800 can be implemented by one or more of an Application Specific Integrated Circuit (ASIC), a Digital Signal Processor (DSP), a Digital Signal Processing Device (DSPD), a Programmable Logic Device (PLD), a Field Programmable Gate Array (FPGA), a controller, a microcontroller, a microprocessor, or other electronic elements, for executing the low-voltage power distribution network change risk judgment and intelligent early warning method described above.

[0139] In another example embodiment, a computer-readable storage medium including program instructions is also provided, which, when executed by a processor, implement the steps of the low-voltage power distribution network change risk judgment and intelligent early warning method described above. For example, the computer-readable storage medium can be the memory 802 described above including program instructions, which can be executed by the processor 801 of the low-voltage power distribution network change risk judgment and intelligent early warning device 800 to complete the low-voltage power distribution network change risk judgment and intelligent early warning method described above.

[0140] Embodiment 4:

[0141] Corresponding to the method embodiments above, in this embodiment, a readable storage medium is also provided, which can be referred to in conjunction with the low-voltage power distribution network change risk judgment and intelligent early warning method described above.

[0142] The computer program stored on the readable storage medium, when executed by a processor, implements the steps of the low-voltage power distribution network change risk judgment and intelligent early warning method of the method embodiments described above.

[0143] The readable storage medium can be specifically a U disk, a mobile hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and various readable storage media that can store program codes.

[0144] In summary, the correlation features of power grid operation state and environmental risk are extracted by the spatio-temporal convolutional neural network, the dynamic changes of power grid topology are modeled by the physical information dynamic graph attention network, and the circuit law is integrated as a constraint condition to ensure the physical rationality of the model. On this basis, the fault probability distribution of each node is calculated by the Bayesian network, and the optimal repair resource scheduling scheme is generated by optimizing the early warning strategy through the reinforcement learning algorithm, which not only can capture the influence of environmental factors on the power grid in real time, but also can accurately predict the risk propagation path before the fault occurs, significantly improving the accuracy of fault early warning and the efficiency of repair, and providing reliable technical support for the intelligent operation and maintenance of low-voltage distribution network.

[0145] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0146] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application. The protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method for assessing and intelligently predicting changes in low-voltage distribution network risks, characterized in that, include: By fusing and analyzing power grid operation data and environmental data through a spatiotemporal convolutional neural network, a correlation feature matrix between power grid operation status and environmental risk is generated. Based on the correlation feature matrix, a dynamic graph attention network of physical information is used to dynamically model the power grid topology and extract dynamic risk propagation paths; By combining dynamic risk propagation paths, a Bayesian network is used to calculate the fault probability distribution of each node in the power grid; Based on the fault probability distribution, the early warning strategy is optimized through reinforcement learning algorithm to generate the optimal early warning scheme. Based on the optimal early warning scheme, a multi-objective optimization algorithm is used to schedule emergency repair resources and generate an emergency repair resource scheduling scheme. The method involves fusing and analyzing power grid operation data and environmental data using a spatiotemporal convolutional neural network to generate a correlation feature matrix between power grid operation status and environmental risks, including: The system collects first data on voltage, current and power factor in real time at a frequency of 1 minute / time using smart meters and sensors; second data on wind speed, rainfall, temperature and humidity at a frequency of 10 minutes / time using weather stations; third data on terrain elevation and vegetation distribution at a frequency of 3 minutes / time using geographic information systems; and fourth data on salt spray concentration at a sampling frequency of 1 hour / time using corrosion monitoring equipment. The first, second, third, and fourth data points are aligned in time and space to construct a spatiotemporal data cube. A spatiotemporal convolutional neural network is then used to process the spatiotemporal data cube to obtain the correlation feature matrix between the power grid operating status and environmental risks. The calculation formula for the spatiotemporal data cube is as follows: Where X represents the fusion result of power grid operation data and environmental data, Each element in the data cube is a real number, N is the number of power grid nodes, T is the time step, T represents the number of data acquisition points, and D represents the number of features for each node at each time point. The method, based on the correlation feature matrix, utilizes a physical information dynamic graph attention network to dynamically model the power grid topology and extract dynamic risk propagation paths, including: Obtain the topology information of the power grid in the power grid management system, including node data and edge data; assign the correlation feature matrix as node values ​​to the topology information of the power grid to obtain the power grid topology with node features; A dynamic graph structure is constructed based on the power grid topology with node characteristics, the second data, and the fourth data. Based on a dynamic graph structure and power grid operation data, including current, voltage, and resistance, and using circuit laws as constraints, a physical information loss function is constructed. This physical information loss function is then integrated into the neural network training process to obtain the node features embedded with physical information. The calculation formula is as follows: In the formula, Represents the physical information loss function. Represents a node i The current, Represents a node i voltage, Represents a node i The resistance, Represents a node j The set of neighbors; By combining the node features embedded with physical information and the dynamic graph structure, the attention coefficient between nodes is calculated using the dynamic graph attention network algorithm, and the edge weights are dynamically adjusted based on the attention coefficient. Based on edge weights and node characteristics, the node state is updated iteratively through multi-layer PI-DGAT to simulate the dynamic propagation process of risk in the power grid and obtain the dynamic risk propagation path.

2. The method for assessing and intelligently predicting changes in low-voltage distribution network risks according to claim 1, characterized in that, The method, which combines dynamic risk propagation paths and uses Bayesian networks to calculate the fault probability distribution of each node in the power grid, includes: By utilizing the dynamic risk propagation path matrix, historical fault data, and real-time equipment status data, a Bayesian network structure is constructed, where network nodes correspond to power grid equipment, network edges represent the risk propagation relationship between equipment, and edge weights are determined by the propagation probability in the dynamic risk propagation path matrix, thereby generating a Bayesian network topology that reflects the dynamic risk propagation logic between power grid equipment. Combining the Bayesian network topology and historical fault data, the basic fault probability of the root node and the conditional fault probability of the non-root node are calculated by training the conditional probability table. The conditional probability of the non-root node is derived by jointly deducing the state of its parent node and the dynamic propagation probability, thus obtaining the conditional probability distribution of the Bayesian network. Real-time equipment status data is used as evidence input into the trained Bayesian network, and a probabilistic inference algorithm is used to calculate the posterior probability of each node under different fault types, outputting the fault probability distribution of the power grid nodes.

3. The method for assessing and intelligently predicting changes in low-voltage distribution network risks according to claim 1, characterized in that, The optimization of the early warning strategy through reinforcement learning algorithms to generate the optimal early warning scheme includes: The fault probability distribution and user feedback data are obtained, integrated and processed to obtain the state space, which is then used as the input to the reinforcement learning algorithm. Using the state space and action space, a Q-value table is obtained through training with the Q-learning algorithm. The Q-value table is used to generate the optimal early warning strategy. Based on the Q-value table and the current state, the optimal action is selected according to the Q-value table, thus obtaining the optimal early warning plan. The optimal early warning plan includes the early warning level and the corresponding countermeasures, and its calculation formula is as follows: In the formula, For optimal action, This is a state-action value function.

4. A low-voltage distribution network change risk assessment and intelligent early warning system, based on the low-voltage distribution network change risk assessment and intelligent early warning method described in claim 1, characterized in that, include: Generation module: Used to fuse and analyze power grid operation data and environmental data through spatiotemporal convolutional neural networks to generate a correlation feature matrix between power grid operation status and environmental risks; Extraction module: Used to dynamically model the power grid topology based on the correlation feature matrix and extract dynamic risk propagation paths using a physical information dynamic graph attention network; Calculation module: Used to combine dynamic risk propagation paths and use Bayesian networks to calculate the fault probability distribution of each node in the power grid; Optimization module: Used to optimize the early warning strategy based on the fault probability distribution and generate the optimal early warning scheme through reinforcement learning algorithm; Dispatch and early warning module: Used to schedule emergency repair resources based on the optimal early warning plan and using a multi-objective optimization algorithm; The generation module includes: The data acquisition unit is used to collect first data such as voltage, current and power factor at a frequency of 1 minute / time through smart meters and sensors, second data such as wind speed, rainfall, temperature and humidity at a frequency of 10 minutes / time from meteorological stations, third data such as terrain elevation and vegetation distribution at geographic information systems, and fourth data such as salt spray concentration at a sampling frequency of 1 hour / time from corrosion monitoring equipment. The first building unit aligns the first, second, third, and fourth data points according to time and space, constructs a spatiotemporal data cube, and processes the spatiotemporal data cube using a spatiotemporal convolutional neural network to obtain the correlation feature matrix between the power grid operating status and environmental risks. The calculation formula for the spatiotemporal data cube is as follows: Where X represents the fusion result of power grid operation data and environmental data, Let N represent the number of power grid nodes, T represent the time step, D represent the number of time points for data acquisition, and D represent the number of features of each node at each time point. The extraction module includes: The first acquisition unit is used to acquire the topology information of the power grid in the power grid management system, including node data and edge data; and to assign the correlation feature matrix as a node to the topology information of the power grid to obtain the power grid topology with node features. The second building unit is used to construct a dynamic graph structure based on the power grid topology with node characteristics, the second data, and the fourth data. The third building unit is used to construct a physical information loss function based on a dynamic graph structure and power grid operation data, including current, voltage, and resistance, and using circuit laws as constraints. This physical information loss function is then integrated into the neural network training process to obtain the node features embedded with physical information. The calculation formula is as follows: In the formula, Represents the physical information loss function. Represents a node i The current, Represents a node i voltage, Represents a node i The resistance, Represents a node j The set of neighbors; The first computing unit is used to combine the node features embedded with physical information and the dynamic graph structure, calculate the attention coefficients between nodes using the dynamic graph attention network algorithm, and dynamically adjust the edge weights based on the attention coefficients. Simulation Unit: Used to simulate the dynamic propagation process of risk in the power grid by iteratively updating the node state through multiple layers of PI-DGAT based on edge weights and node characteristics, and obtain the dynamic risk propagation path.

5. The low-voltage distribution network change risk assessment and intelligent early warning system according to claim 4, characterized in that, The computing module includes: The fourth building unit is used to construct a Bayesian network structure using the dynamic risk propagation path matrix, historical fault data, and real-time equipment status data. The network nodes correspond to power grid equipment, the network edges represent the risk propagation relationship between equipment, and the edge weights are determined by the propagation probabilities in the dynamic risk propagation path matrix. This generates a Bayesian network topology that reflects the dynamic risk propagation logic between power grid equipment. The second calculation unit is used to combine the Bayesian network topology and historical fault data, and calculate the basic fault probability of the root node and the conditional fault probability of the non-root node by training the conditional probability table. The conditional probability of the non-root node is derived by jointly deducing the state of its parent node and the dynamic propagation probability, thus obtaining the conditional probability distribution of the Bayesian network. The third computational unit is used to input real-time equipment status data as evidence into the trained Bayesian network, and uses a probabilistic inference algorithm to calculate the posterior probability of each node under different fault types, and outputs the fault probability distribution of the power grid nodes.

6. The low-voltage distribution network change risk assessment and intelligent early warning system according to claim 4, characterized in that, The optimization module includes: The second acquisition unit is used to acquire the fault probability distribution and user feedback data. After integration and processing, the state space is obtained, and the state space is used as the input of the reinforcement learning algorithm. Processing unit: Used to train and process the state space and action space using the Q-learning algorithm to obtain the Q-value table, which is used to generate the optimal early warning strategy; The acquisition unit is used to select the optimal action based on the Q-value table and the current state, thereby obtaining the optimal early warning plan. The optimal early warning plan includes the early warning level and the corresponding countermeasures, and its calculation formula is as follows: In the formula, For optimal action, This is a state-action value function.

Images