Relay protection method and device based on multi-modal data fusion, terminal equipment and storage medium

Abstract

The invention discloses a relay protection method and device based on multi-modal data fusion, terminal equipment and a storage medium, and belongs to the technical field of electric power, and the method comprises the steps: constructing an electrical quantity view according to electrical data, and constructing a traveling wave residual view according to traveling wave data; performing view fusion on the electrical quantity view and the traveling wave residual view to generate a fault propagation trend graph; constructing a target state sequence according to a fault propagation trend graph and the state data; and inputting the target state sequence into a preset relay protection decision model, so that the relay protection decision model outputs a plurality of action sequences of the protection devices with the aim of completing fault isolation as soon as possible, generates an action instruction of each protection device, and sends the action instruction to the corresponding protection device. Therefore, the problems that an existing relay protection method is poor in robustness and failure in fault isolation when facing complex disturbance are solved.

Description

Technical Field

[0001] This invention relates to the field of power technology, and in particular to a relay protection method, device, terminal equipment, and storage medium based on multimodal data fusion. Background Technology

[0002] With the development of the power system, especially the integration of a high proportion of new energy sources (such as wind power and photovoltaics), the operating environment of the power grid has undergone tremendous changes, exhibiting the following characteristics: high volatility; unstable output from new energy sources leads to frequent fluctuations in power flow and voltage; frequent power flow reversals; the existence of distributed generation allows for bidirectional power flow, altering the traditional unidirectional power supply mode; unstable short-circuit capacity; and changes in the system's equivalent impedance result in fault current levels that differ from traditional conditions. These new characteristics make the operating conditions of the power grid more complex.

[0003] Traditional relay protection methods have revealed significant shortcomings. Existing relay protection decisions largely rely on a single electrical quantity, such as detecting whether the current exceeds a set value or whether the voltage drops. This single source of information cannot fully reflect the complete picture of the fault under complex operating conditions, easily leading to maloperation or failure to operate of the protection device. Therefore, existing relay protection methods have poor robustness when facing complex disturbances and suffer from fault isolation failures. Summary of the Invention

[0004] This invention provides a relay protection method, device, terminal equipment, and storage medium based on multimodal data fusion. The method can solve the problem of fault isolation failure in existing relay protection methods when facing complex disturbances.

[0005] An embodiment of the present invention provides a relay protection method based on multi-modal data fusion, comprising: Acquire electrical data of each node, traveling wave data monitored by each traveling wave sensor, and status data of several protection devices when a power system fault occurs; Based on the electrical data, construct an electrical quantity view, and based on the traveling wave data, construct a traveling wave residual view; The electrical quantity view and the traveling wave residual view are fused to generate a fault propagation trend map characterizing the fault propagation trend of the power system. Based on the fault propagation trend map and the state data, a target state sequence of the power system is constructed; The target state sequence is input into a preset relay protection decision model so that the relay protection decision model outputs a sequence of actions of the protection devices with the goal of completing fault isolation as quickly as possible. Based on the action sequence, an action command for each of the protection devices is generated, and the action command is sent to the corresponding protection device.

[0006] Furthermore, constructing a traveling wave residual view based on the traveling wave data includes: Acquire historical traveling wave data of each traveling wave sensor prior to the occurrence of the fault; According to the preset time window, the traveling wave data of each traveling wave sensor is divided into several target frames, and the short-time energy of each target frame is calculated. The historical traveling wave data is also divided into several historical frames, and the historical short-time energy of each historical frame is calculated. Based on the short-time energy of each target frame and the historical short-time energy of the corresponding historical frame, calculate the energy residual of several target frames for each traveling wave sensor. Based on the actual coordinates of each traveling wave sensor and each node, determine the mapped coordinates of each traveling wave sensor in the electrical quantity view; A traveling wave residual view is constructed based on the mapping coordinates and the energy residuals of several target frames of each traveling wave sensor.

[0007] Furthermore, the step of fusing the electrical quantity view and the traveling wave residual view to generate a fault propagation trend map characterizing the fault propagation trend of the power system includes: The electrical quantity view and the traveling wave residual view are aligned and fused in the spatial dimension to generate spatial alignment features, and the electrical quantity view and the traveling wave residual view are aligned and fused in the time dimension to generate time alignment features. The spatial alignment features and temporal alignment features are concatenated to generate a spatiotemporal fusion feature; The spatiotemporal fusion features are input into a preset spatiotemporal convolutional network so that the spatiotemporal convolutional network can identify the fault distribution features at each moment in the spatial dimension and identify the evolution trend features of the fault over time in the temporal dimension, and generate a fault propagation trend map based on the fault distribution features and evolution trend features.

[0008] Furthermore, constructing the target state sequence of the power system based on the fault propagation trend map and the state data includes: Based on the fault propagation trend map and the state data, an initial state space is constructed; Using a pre-defined temporal latent space environment model, the initial state space is abstracted and its dimensionality reduced to generate a target state sequence.

[0009] Furthermore, the construction of the relay protection decision model includes: Obtain the simulation operation model of the power system; With the goal of completing fault isolation as quickly as possible, a reward function and an initial policy network are constructed. Construct an intelligent agent and use the intelligent agent to repeatedly optimize the initial policy network until a relay protection decision model is generated; The optimization operation includes: Obtain the policy network to be optimized; wherein, initially, the policy network to be optimized is the initial policy network; Fault points and the states of protection devices are randomly set in the simulation operation model to construct several state sequence samples. The agent generates an action sequence consisting of the actions of several protection devices based on each state sequence sample and the strategy network to be optimized, and interacts with the simulation operation model based on the action sequence. Based on the reward function and the response state sequence of the simulation running model after the interaction, the corresponding reward function value is generated, and several experience sequences are constructed based on the state sequence sample, action sequence, reward function value and response state sequence. Based on the empirical sequence, the policy network to be optimized is optimized to generate a policy network to be evaluated. When several reward function values ​​tend to converge, the strategy network to be evaluated is used as the relay protection decision model; otherwise, the strategy network to be evaluated is used as the strategy network to be optimized in the next round of optimization operation.

[0010] An embodiment of the present invention also provides a relay protection device based on multi-modal data fusion, comprising: The data acquisition module is used to acquire electrical data of each node, traveling wave data monitored by each traveling wave sensor, and status data of several protection devices when a power system fault occurs. A view construction module is used to construct an electrical quantity view based on the electrical data, and to construct a traveling wave residual view based on the traveling wave data; The view fusion module is used to fuse the electrical quantity view and the traveling wave residual view to generate a fault propagation trend map characterizing the fault propagation trend of the power system. A sequence construction module is used to construct the target state sequence of the power system based on the fault propagation trend map and the state data. The relay decision module is used to input the target state sequence into a preset relay protection decision model so that the relay protection decision model can output a series of action sequences of the protection devices with the goal of completing fault isolation as quickly as possible. The relay protection module is used to generate an action command for each of the protection devices according to the action sequence, and send the action command to the corresponding protection device.

[0011] Furthermore, the view construction module constructs a traveling wave residual view based on the traveling wave data, including: Acquire historical traveling wave data of each traveling wave sensor prior to the occurrence of the fault; According to the preset time window, the traveling wave data of each traveling wave sensor is divided into several target frames, and the short-time energy of each target frame is calculated. The historical traveling wave data is also divided into several historical frames, and the historical short-time energy of each historical frame is calculated. Based on the short-time energy of each target frame and the historical short-time energy of the corresponding historical frame, calculate the energy residual of several target frames for each traveling wave sensor. Based on the actual coordinates of each traveling wave sensor and each node, determine the mapped coordinates of each traveling wave sensor in the electrical quantity view; A traveling wave residual view is constructed based on the mapping coordinates and the energy residuals of several target frames of each traveling wave sensor.

[0012] Furthermore, the view fusion module performs view fusion on the electrical quantity view and the traveling wave residual view to generate a fault propagation trend map characterizing the fault propagation trend of the power system, including: The electrical quantity view and the traveling wave residual view are aligned and fused in the spatial dimension to generate spatial alignment features, and the electrical quantity view and the traveling wave residual view are aligned and fused in the time dimension to generate time alignment features. The spatial alignment features and temporal alignment features are concatenated to generate a spatiotemporal fusion feature; The spatiotemporal fusion features are input into a preset spatiotemporal convolutional network so that the spatiotemporal convolutional network can identify the fault distribution features at each moment in the spatial dimension and identify the evolution trend features of the fault over time in the temporal dimension, and generate a fault propagation trend map based on the fault distribution features and evolution trend features.

[0013] This application also provides a terminal device, including: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement a relay protection method based on multimodal data fusion as described in the above embodiments of the invention.

[0014] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a relay protection method based on multimodal data fusion as described in the above embodiments of the invention.

[0015] The following benefits can be obtained by implementing the present invention: This invention provides a relay protection method, device, terminal equipment, and storage medium based on multimodal data fusion. The method constructs an electrical quantity view based on electrical data and a traveling wave residual view based on traveling wave data. The electrical quantity view and the traveling wave residual view are fused to generate a fault propagation trend map characterizing the fault propagation trend of the power system. Therefore, by fusing two types of data with different characteristics—electrical quantities and traveling waves—this invention can more comprehensively and accurately assess fault conditions under complex operating conditions, capture fault propagation trends, and generate a fault propagation trend map. Based on the fault propagation trend map and the state data, a target state sequence of the power system is constructed. This target state sequence is input into a preset relay protection decision model, so that the relay protection decision model, with the goal of completing fault isolation as quickly as possible, outputs a series of action sequences for several protection devices. Based on the action sequences, an action command for each protection device is generated and sent to the corresponding protection device. Therefore, this invention, with the goal of "completing fault isolation as quickly as possible," calculates and outputs a complete set of coordinated action sequences for protection devices based on the global fault propagation trend and the state data of the protection devices. This method minimizes fault clearing time and reduces the impact on system stability, overcoming the problem that existing relay protection methods have poor robustness and fault isolation failure when facing complex disturbances. Attached Figure Description

[0016] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a relay protection method based on multimodal data fusion according to a certain embodiment of this application; Figure 2 This is a schematic diagram of the structure of a relay protection device based on multimodal data fusion according to a certain embodiment of this application; Figure 3 This is a schematic diagram of the structure of a terminal device provided in a certain embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0020] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0021] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0022] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0023] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0024] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0025] See Figure 1 To address the problems in the prior art, an embodiment of the present invention provides a relay protection method based on multi-modal data fusion, comprising: S1. Acquire electrical data of each node, traveling wave data monitored by each traveling wave sensor, and status data of several protection devices when a power system fault occurs. In a preferred embodiment of the present invention, the electrical data includes: voltage, current, and voltage phase angle; traveling wave data refers to the extremely high-frequency voltage and current transient waves propagating along the transmission line during a fault, as monitored by a traveling wave sensor. The protection device refers to a circuit breaker, and the status data is the circuit breaker's open / closed state.

[0026] S2. Based on the electrical data, construct an electrical quantity view, and based on the traveling wave data, construct a traveling wave residual view; Preferably, constructing the traveling wave residual view based on the traveling wave data includes: Acquire historical traveling wave data of each traveling wave sensor before the fault occurs; divide the traveling wave data of each traveling wave sensor into several target frames and the short-time energy of each target frame according to a preset time window, and divide the historical traveling wave data into several historical frames and the historical short-time energy of each historical frame; calculate the energy residual of several target frames of each traveling wave sensor based on the short-time energy of each target frame and the historical short-time energy of the corresponding historical frame; determine the mapping coordinates of each traveling wave sensor in the electrical quantity view based on the actual coordinates of each traveling wave sensor and each node; In a preferred embodiment of the present invention, the voltage amplitude, phase angle, and current amplitude are quantized into an electrical quantity view of a three-dimensional spatial tensor: In the formula, View of electrical quantities; The coordinates of the power grid nodes are used; the spatial resolution is set to 0.1 kV / pixel for voltage (its value range is...). The phase angle is 1° / pixel (its range is...). The current is 0.1 kV / pixel (its value range is...). ) For the corresponding node The magnitude of the voltage; For the corresponding node The value of the phase angle of the voltage. For the corresponding node The amplitude of the current.

[0027] Furthermore, the traveling wave data collected by each traveling wave sensor are synchronized in time. For each synchronized traveling wave data point, the signal energy of each time window is calculated to characterize the wave intensity at that time point. In the formula, For time Short-term energy; For time windows; For time coefficient; For traveling wave sensors in time Short-term energy.

[0028] To highlight the dynamic changes caused by faults, the energy of the current frame is compared with the energy of historical frames. Specifically, the absolute difference between the energy of the current frame and the energy of the nth previous frame is calculated and normalized using a sliding statistic: ; In the formula, The energy of the current frame; For the first The historical energy of frames; For historical frame counts; It is a minimal constant to prevent initial zero; For traveling wave sensors in time Short-term energy. For normalized residuals.

[0029] Furthermore, traveling wave sensors are typically mounted in specific locations, and their detection is directional. This is to map the residual information to the electrical quantity... Figure 1 In the spatial coordinate system, spherical projection is required. The spatial azimuth and elevation angles of each traveling wave sensor are projected onto the two-dimensional image plane, yielding its two-dimensional pixel coordinates. : ; In the formula, ; These are the actual physical coordinates of the traveling wave sensor; The preset image width for the traveling wave view; The image height of the preset traveling wave view; This is the sensor's maximum elevation angle; This is the difference between the sensor's maximum and minimum elevation angles; These are the two-dimensional pixel coordinates after projection.

[0030] For each sensor, the calculated normalized residuals are... , with projection point Centered on a Gaussian weight kernel, superimposed on it. On the covered image area: ; in, Traveling wave view Pixel value at that location, To represent an assignment operation, update the image pixel values; Let be a two-dimensional Gaussian function, with weights centered at (u, v). Gaussian weighting avoids the abrupt jumps between discrete points, better reflecting the continuity of physical phenomena. Its formula can be expressed as: in, Standard deviation, This is for exponential operations.

[0031] S3. Perform view fusion on the electrical quantity view and the traveling wave residual view to generate a fault propagation trend map characterizing the fault propagation trend of the power system; Preferably, the step of fusing the electrical quantity view and the traveling wave residual view to generate a fault propagation trend map characterizing the fault propagation trend of the power system includes: The electrical quantity view and the traveling wave residual view are aligned and fused in the spatial dimension to generate spatial alignment features, and the electrical quantity view and the traveling wave residual view are aligned and fused in the temporal dimension to generate temporal alignment features. The spatial alignment features and the temporal alignment features are then concatenated to generate spatiotemporal fusion features. The spatiotemporal fusion features are then input into a preset spatiotemporal convolutional network, so that the spatiotemporal convolutional network can identify the fault distribution features at each moment in the spatial dimension and identify the evolution trend features of the fault over time in the temporal dimension, and generate a fault propagation trend map based on the fault distribution features and evolution trend features.

[0032] In a preferred embodiment of the present invention, the electrical quantity view is projected onto the traveling wave residual view coordinate system by bilinear interpolation, and cubic spline interpolation is performed on the traveling wave data based on the sampling time: ; ; In the formula, Spatial alignment feature; This is the transformation matrix from the electrical quantity view to the traveling wave converter; Characteristics of a traveling wave converter; For time alignment features; For bilinear interpolation calculations; This is a cubic spline interpolation; For traveling wave data; For traveling wave timestamps; for Timestamp.

[0033] This series of processing steps transforms the original high-speed sampled data into a tensor representation rich in fault features, aligned spatial coordinates, and with regular dimensions, providing ideal input for subsequent multimodal fusion and entropy maximization learning. This method not only effectively suppresses noise interference but also highlights transient changes caused by faults through residual operations, significantly improving the detection capability for difficult situations such as early faults and low-current grounding.

[0034] Furthermore, the features are spliced ​​and fused, then input into the fault prediction head to generate a fault location and dynamic evolution map: As a multimodal fusion feature, its function is to fuse temporal and spatial features; For multimodal fusion computing. It is a spatiotemporal convolutional network. This is to create a fault location and dynamic evolution map after feature splicing.

[0035] Understandably, spatiotemporal convolutional networks employ convolutional architectures such as ConvLSTM and 3D CNN. Spatial convolution captures the spatial distribution of faults at any given moment, identifying fault points and outlining their impact range. Temporal convolution analyzes the fused features across multiple consecutive time steps, capturing the propagation and evolution trends of faults over time. It determines whether the fault arc is moving along the line and whether it will endanger adjacent areas. This results in the output of pixel-level fault localization and dynamic evolution maps. In essence, the output of convolutional architectures like ConvLSTM and 3D CNN is a multi-channel image of the same size as the input image. Channel one outputs a fault probability map. The value of each pixel (0-1) represents the probability of a fault occurring at that location. This is pixel-level localization, accurate to the meter level, and even capable of distinguishing different sections of the line. Channel two outputs an evolution trend map. The value of each pixel represents the risk or speed of fault propagation in that direction. This might be represented by a vector or heatmap. These are then fused to generate a fault propagation trend map.

[0036] S4. Construct the target state sequence of the power system based on the fault propagation trend map and the state data; Preferably, constructing the target state sequence of the power system based on the fault propagation trend map and the state data includes: Based on the fault propagation trend map and the state data, an initial state space is constructed; using a preset temporal latent space environment model, the initial state space is abstracted and its dimensionality reduced to generate a target state sequence.

[0037] In a preferred embodiment of the present invention, a highly abstract, low-dimensional hidden state is generated through a temporal hidden space environment model based on a probabilistic graphical model (PGM). This hidden state encapsulates the current operating status, fault information, and evolution trend of the power grid, providing an efficient and information-rich decision-making basis for subsequent reinforcement learning agents.

[0038] Specifically, through temporal spatial modeling, the environment model is defined as a probabilistic graphical model: In the formula, The state transition probability; This is the current hidden state; For the current action; These are the mean and covariance functions, respectively; This is the state transition function; This is a multimodal fusion feature; The empirical distribution after the hidden state; It is a convolutional neural network for encoders; The input fault location and dynamic evolution map is obtained by splicing the fused features.

[0039] It should be noted that this embodiment trains the temporal latent space environment model by constructing an objective function that maximizes the lower bound of evidence: ; In the formula, For reconstruction losses; Let KL divergence be a metric. Evidence-based lower bound; The reconstruction term measures the model's ability to reconstruct the original fused features from the hidden state. The regularization term constrains the learned posterior distribution. Do not deviate from the prior distribution defined by the state transition. Too far; Hyperparameters used to control the balance between latent space compactness and reconstruction accuracy.

[0040] S5. Input the target state sequence into the preset relay protection decision model so that the relay protection decision model can output a number of action sequences of the protection devices with the goal of completing fault isolation as quickly as possible. Preferably, the construction of the relay protection decision model includes: Obtain the simulation operation model of the power system; construct a reward function and an initial policy network with the goal of completing fault isolation as quickly as possible; construct an agent and use the agent to repeatedly optimize the initial policy network until a relay protection decision model is generated. The optimization operation includes: acquiring a strategy network to be optimized; initially, the strategy network to be optimized is an initial strategy network; randomly setting fault points and the states of protection devices in the simulation operation model to construct several state sequence samples, so that the agent generates an action sequence consisting of the actions of several protection devices based on each state sequence sample and the strategy network to be optimized, and interacts with the simulation operation model based on the action sequence; generating corresponding reward function values ​​based on the reward function and the response state sequences of the simulation operation model after the interaction, and constructing several experience sequences based on the state sequence samples, action sequences, reward function values, and response state sequences; optimizing the strategy network to be optimized based on the experience sequences to generate a strategy network to be evaluated; when several reward function values ​​tend to converge, the strategy network to be evaluated is used as the relay protection decision model; otherwise, the strategy network to be evaluated is used as the strategy network to be optimized in the next round of optimization operation.

[0041] In a preferred embodiment of the present invention, a Maximum Entropy Reinforcement Learning (MaxEnt RL) agent is trained using the compact, information-rich low-dimensional hidden states generated by the temporal hidden space environment model. This agent learns a stochastic policy whose objective is not only to maximize the cumulative reward (i.e., successfully isolating the fault) but also to maximize the policy's entropy (i.e., maintaining exploratory nature). Ultimately, this policy can coordinate multiple protection devices (such as circuit breakers and relays) to issue the optimal sequence of cooperative actions (tripping, delay, blocking) based on the real-time state of the power grid, thereby achieving fast, accurate, and reliable fault isolation. The specific steps are as follows: First, we define the objective of maximum entropy reinforcement learning. Unlike traditional deep reinforcement learning, which only maximizes the cumulative reward, the objective function of MaxEnt RL adds policy entropy. Item: ; In the formula, The goal of maximum entropy reinforcement learning is also the collaborative decision-making strategy for relay protection; It is a strategy Induced state-action access distribution; As a balancing parameter, it controls the relative importance between exploration (entropy) and exploitation (reward). The hidden state of the power grid at time t is generated by the time-series hidden space model and includes the fault evolution trend, fault type, and location. The coordinated protection action vector at time t includes circuit breaker trip command, reclosing control signal area blocking / acceleration signal; For in action and hidden state The protection effect reward function evaluates the quality of the action. A positive reward represents the correct isolation of faults, maintenance of power supply, and protection of equipment; a negative reward represents false action, failure to act, or delayed action. Let entropy be the policy entropy. Higher entropy encourages the exploration of new protection strategies, thus encouraging the agent to explore more, while lower entropy prioritizes the use of known effective strategies. In the current policy... In all the power grid operation scenarios and historical decisions generated as a result, all possible hidden states of the power grid are considered. Combined with coordinated protective actions The probability-weighted average.

[0042] Next, the policy network (Actor) and value function network (Critic) are designed and implemented. Policy Network Hidden state As input, the output is a Gaussian distribution and its log-standard deviation, thus defining a stochastic policy over a continuous action space. Value function network Then assess the state Next action The quality of these networks is important. Both networks are typically multilayer perceptrons (MLPs). The policy network parameters are then optimized using stochastic gradient ascent. The goal of updating the strategy is to maximize the sum of expected return and entropy: ; In the formula, As a policy network optimization objective, it measures the current cooperative protection strategy. The score represents the overall advantages and disadvantages of the relay protection strategy. It also represents the comprehensive performance score of the relay protection strategy; the higher the score, the better the strategy is in terms of long-term protection effectiveness and ability to cope with uncertainties. This serves as an experience replay buffer, storing historical protection decision-making experience: grid status, protection actions, rewards, and the next state; The action value function is used to evaluate the state of the power grid. Perform protective actions The long-term expected return; To select protective actions The probability; where the exploration mechanism for protective actions is , , These are the mean vector and the standard vector output by the policy network, respectively. To ensure the exploratory nature of protection actions, random noise is used. For all possible protective actions to be performed According to the strategy The given probability distributions are weighted averaged. This refers to the experience replay buffer. Extracted hidden states of the power grid The expected value.

[0043] Simultaneously, the value function network parameters are optimized by minimizing the temporal difference (TD) error. Overestimation is mitigated by employing two Q-networks and minimizing their values, and updates are performed by minimizing the following Bellman error: Among them, the target value Calculated through the target network: In the formula, To protect the error index of the effectiveness prediction system, the goal is to minimize this error through training, so that the agent can more accurately predict the value of protective actions. This is a discount factor, a coefficient that weighs the importance of current rewards against future rewards. The higher the value, the more emphasis is placed on long-term system stability, taking into account the delayed consequences of actions; the lower the value, the more emphasis is placed on immediate protection effects, prioritizing the resolution of current problems. This represents the minimum value of the double-Q network, signifying a conservative value assessment strategy to prevent overly optimistic estimates of the protection effect. The parameters of the target Q and policy network are used to slowly track the current network through soft updates in order to improve training stability. For experience replay buffer Randomly sample historical data (including the current power grid status, protective actions taken, immediate rewards obtained, and new status) and average the expected value of these cases. For instant rewards; To the target policy network The action of the next moment in the middle sampling The expected value is calculated. Finally, the trained policy network is deployed and applied. In real-time operation, the environment model continuously encodes multimodal observations into hidden states. The policy network is based on Real-time output action distribution .

[0044] S6. Based on the action sequence, generate an action command for each of the protection devices and send the action command to the corresponding protection device.

[0045] In summary, this embodiment provides a relay protection method based on multimodal data fusion. It constructs an electrical quantity view based on electrical data and a traveling wave residual view based on traveling wave data. The electrical quantity view and the traveling wave residual view are fused to generate a fault propagation trend map characterizing the fault propagation trend of the power system. Therefore, by fusing two types of data with different characteristics—electrical quantities and traveling waves—this invention can more comprehensively and accurately assess fault conditions under complex operating conditions, capture fault propagation trends, and generate a fault propagation trend map. Based on the fault propagation trend map and the state data, a target state sequence of the power system is constructed. This target state sequence is input into a preset relay protection decision model, enabling the model to output action sequences of several protection devices with the goal of completing fault isolation as quickly as possible. Based on the action sequences, action commands for each protection device are generated and sent to the corresponding protection device. Therefore, this invention, with the goal of "completing fault isolation as quickly as possible," calculates and outputs a complete set of coordinated action sequences for protection devices based on the global fault propagation trend and the state data of the protection devices. This method minimizes fault clearing time and reduces the impact on system stability, overcoming the problem that existing relay protection methods have poor robustness and fail to isolate faults when facing complex disturbances.

[0046] See Figure 2 This invention provides a relay protection device based on multimodal data fusion, comprising: The data acquisition module is used to acquire electrical data of each node, traveling wave data monitored by each traveling wave sensor, and status data of several protection devices when a power system fault occurs. A view construction module is used to construct an electrical quantity view based on the electrical data, and to construct a traveling wave residual view based on the traveling wave data; The view fusion module is used to fuse the electrical quantity view and the traveling wave residual view to generate a fault propagation trend map characterizing the fault propagation trend of the power system. A sequence construction module is used to construct the target state sequence of the power system based on the fault propagation trend map and the state data. The relay decision module is used to input the target state sequence into a preset relay protection decision model so that the relay protection decision model can output a series of action sequences of the protection devices with the goal of completing fault isolation as quickly as possible. The relay protection module is used to generate an action command for each of the protection devices according to the action sequence, and send the action command to the corresponding protection device.

[0047] Furthermore, the view construction module constructs a traveling wave residual view based on the traveling wave data, including: Acquire historical traveling wave data of each traveling wave sensor prior to the occurrence of the fault; According to the preset time window, the traveling wave data of each traveling wave sensor is divided into several target frames, and the short-time energy of each target frame is calculated. The historical traveling wave data is also divided into several historical frames, and the historical short-time energy of each historical frame is calculated. Based on the short-time energy of each target frame and the historical short-time energy of the corresponding historical frame, calculate the energy residual of several target frames for each traveling wave sensor. Based on the actual coordinates of each traveling wave sensor and each node, determine the mapped coordinates of each traveling wave sensor in the electrical quantity view; A traveling wave residual view is constructed based on the mapping coordinates and the energy residuals of several target frames of each traveling wave sensor.

[0048] Furthermore, the view fusion module performs view fusion on the electrical quantity view and the traveling wave residual view to generate a fault propagation trend map characterizing the fault propagation trend of the power system, including: Spatially and temporally align the electrical quantity view and the traveling wave residual view to generate aligned target electrical quantity view and target traveling wave residual view; The target electrical quantity view and the target traveling wave residual view are fused to generate a fused feature view; The fused feature view is input into a preset spatiotemporal convolutional network so that the spatiotemporal convolutional network can identify the fault distribution features at each moment in the spatial dimension and identify the evolution trend features of the fault over time in the temporal dimension, and generate a fault propagation trend map based on the fault distribution features and evolution trend features.

[0049] It is understood that the above-described device embodiments correspond to the method embodiments of the present invention, and can realize the relay protection method based on multimodal data fusion provided by any of the above-described method embodiments of the present invention.

[0050] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0051] See Figure 3 One embodiment of this application also provides a terminal device, including: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement a relay protection method based on multimodal data fusion as described above.

[0052] The processor controls the overall operation of the terminal device to complete all or part of the steps of the aforementioned relay protection method based on multimodal data fusion. The memory stores various types of data to support the operation of the terminal device. This data may include, for example, instructions for any application or method operating on the terminal device, as well as application-related data. The memory can be implemented using 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 storage, flash memory, magnetic disk, or optical disk.

[0053] In an exemplary embodiment, the terminal device may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform a relay protection method based on multimodal data fusion as described in any of the foregoing embodiments, and achieve the same technical effects as the methods described above.

[0054] In another exemplary embodiment, a computer-readable storage medium including a computer program is also provided. When executed by a processor, the computer program implements the steps of a relay protection method based on multimodal data fusion as described in any of the foregoing embodiments. For example, the computer-readable storage medium may be the aforementioned memory including the computer program, which may be executed by a processor of a terminal device to complete the relay protection method based on multimodal data fusion as described in any of the foregoing embodiments and achieve the same technical effects as the aforementioned method.

[0055] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A relay protection method based on multi-modal data fusion, characterized in that, include: Acquire electrical data of each node, traveling wave data monitored by each traveling wave sensor, and status data of several protection devices when a power system fault occurs; Based on the electrical data, construct an electrical quantity view, and based on the traveling wave data, construct a traveling wave residual view; The electrical quantity view and the traveling wave residual view are fused to generate a fault propagation trend map characterizing the fault propagation trend of the power system. Based on the fault propagation trend map and the state data, a target state sequence of the power system is constructed; The target state sequence is input into a preset relay protection decision model so that the relay protection decision model outputs a sequence of actions of the protection devices with the goal of completing fault isolation as quickly as possible. Based on the action sequence, an action command for each of the protection devices is generated, and the action command is sent to the corresponding protection device.

2. The relay protection method based on multi-modal data fusion as described in claim 1, characterized in that, The step of constructing a traveling wave residual view based on the traveling wave data includes: Acquire historical traveling wave data of each traveling wave sensor prior to the occurrence of the fault; According to the preset time window, the traveling wave data of each traveling wave sensor is divided into several target frames, and the short-time energy of each target frame is calculated. The historical traveling wave data is also divided into several historical frames, and the historical short-time energy of each historical frame is calculated. Based on the short-time energy of each target frame and the historical short-time energy of the corresponding historical frame, calculate the energy residual of several target frames for each traveling wave sensor. Based on the actual coordinates of each traveling wave sensor and each node, determine the mapped coordinates of each traveling wave sensor in the electrical quantity view; A traveling wave residual view is constructed based on the mapping coordinates and the energy residuals of several target frames of each traveling wave sensor.

3. The relay protection method based on multi-modal data fusion as described in claim 2, characterized in that, The step of fusing the electrical quantity view and the traveling wave residual view to generate a fault propagation trend map characterizing the fault propagation trend of the power system includes: The electrical quantity view and the traveling wave residual view are aligned and fused in the spatial dimension to generate spatial alignment features, and the electrical quantity view and the traveling wave residual view are aligned and fused in the time dimension to generate time alignment features. The spatial alignment features and temporal alignment features are concatenated to generate a spatiotemporal fusion feature; The spatiotemporal fusion features are input into a preset spatiotemporal convolutional network so that the spatiotemporal convolutional network can identify the fault distribution features at each moment in the spatial dimension and identify the evolution trend features of the fault over time in the temporal dimension, and generate a fault propagation trend map based on the fault distribution features and evolution trend features.

4. The relay protection method based on multi-modal data fusion as described in claim 3, characterized in that, The step of constructing the target state sequence of the power system based on the fault propagation trend map and the state data includes: Based on the fault propagation trend map and the state data, an initial state space is constructed; Using a pre-defined temporal latent space environment model, the initial state space is abstracted and its dimensionality reduced to generate a target state sequence.

5. A relay protection method based on multi-modal data fusion as described in claim 4, characterized in that, The construction of the relay protection decision model includes: Obtain the simulation operation model of the power system; With the goal of completing fault isolation as quickly as possible, a reward function and an initial policy network are constructed. Construct an intelligent agent and use the intelligent agent to repeatedly optimize the initial policy network until a relay protection decision model is generated; The optimization operation includes: Obtain the policy network to be optimized; wherein, initially, the policy network to be optimized is the initial policy network; Fault points and the states of protection devices are randomly set in the simulation operation model to construct several state sequence samples. The agent generates an action sequence consisting of the actions of several protection devices based on each state sequence sample and the strategy network to be optimized, and interacts with the simulation operation model based on the action sequence. Based on the reward function and the response state sequence of the simulation running model after the interaction, the corresponding reward function value is generated, and several empirical sequences are constructed based on the state sequence sample, action sequence, reward function value and response state sequence. Based on the empirical sequence, the policy network to be optimized is optimized to generate a policy network to be evaluated. When several reward function values ​​tend to converge, the strategy network to be evaluated is used as the relay protection decision model; otherwise, the strategy network to be evaluated is used as the strategy network to be optimized in the next round of optimization operation.

6. A relay protection device based on multi-modal data fusion, characterized in that, include: The data acquisition module is used to acquire electrical data of each node, traveling wave data monitored by each traveling wave sensor, and status data of several protection devices when a power system fault occurs. A view construction module is used to construct an electrical quantity view based on the electrical data, and to construct a traveling wave residual view based on the traveling wave data; The view fusion module is used to fuse the electrical quantity view and the traveling wave residual view to generate a fault propagation trend map characterizing the fault propagation trend of the power system. A sequence construction module is used to construct the target state sequence of the power system based on the fault propagation trend map and the state data. The relay decision module is used to input the target state sequence into a preset relay protection decision model so that the relay protection decision model can output a series of action sequences of the protection devices with the goal of completing fault isolation as quickly as possible. The relay protection module is used to generate an action command for each of the protection devices according to the action sequence, and send the action command to the corresponding protection device.

7. A relay protection device based on multi-modal data fusion as described in claim 6, characterized in that, The view construction module constructs a traveling wave residual view based on the traveling wave data, including: Acquire historical traveling wave data of each traveling wave sensor prior to the occurrence of the fault; According to the preset time window, the traveling wave data of each traveling wave sensor is divided into several target frames, and the short-time energy of each target frame is calculated. The historical traveling wave data is also divided into several historical frames, and the historical short-time energy of each historical frame is calculated. Based on the short-time energy of each target frame and the historical short-time energy of the corresponding historical frame, calculate the energy residual of several target frames for each traveling wave sensor. Based on the actual coordinates of each traveling wave sensor and each node, determine the mapped coordinates of each traveling wave sensor in the electrical quantity view; A traveling wave residual view is constructed based on the mapping coordinates and the energy residuals of several target frames of each traveling wave sensor.

8. A relay protection device based on multi-modal data fusion as described in claim 7, characterized in that, The view fusion module fuses the electrical quantity view and the traveling wave residual view to generate a fault propagation trend map characterizing the fault propagation trend of the power system, including: The electrical quantity view and the traveling wave residual view are aligned and fused in the spatial dimension to generate spatial alignment features, and the electrical quantity view and the traveling wave residual view are aligned and fused in the time dimension to generate time alignment features. The spatial alignment features and temporal alignment features are concatenated to generate a spatiotemporal fusion feature; The spatiotemporal fusion features are input into a preset spatiotemporal convolutional network so that the spatiotemporal convolutional network can identify the fault distribution features at each moment in the spatial dimension and identify the evolution trend features of the fault over time in the temporal dimension, and generate a fault propagation trend map based on the fault distribution features and evolution trend features.

9. A terminal device, characterized in that, include: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement a relay protection method based on multimodal data fusion as described in any one of claims 1-5.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a relay protection method based on multimodal data fusion as described in any one of claims 1-5.

Images