Equipment fault early warning method and system

Abstract

The invention discloses an equipment fault early warning method and system, relates to the technical field of power equipment fault warning, and solves the problem that an existing method is difficult to capture cross-equipment early fault features. According to the embodiment of the invention, by monitoring the change of the electrical variable representing the coupling relationship between the devices, the early signal transmitted by the fault through the cross-device coupling feature can be captured when the body device feature is not obviously abnormal, and early warning is carried out; by combining a map model and a causal analysis technology, an initial fault source can be positioned from the anomalies of a plurality of associated devices, fault propagation path analysis is provided, and accurate maintenance is guided.

Description

Technical Field

[0001] This invention relates to the field of power equipment fault alarm technology, and in particular to a method and system for early warning of equipment faults. Background Technology

[0002] During the operation of power plants, key equipment such as generators, main transformers, and high-voltage transformers are the core carriers for ensuring a stable power supply. Their operating status directly affects the safety, reliability, and economy of the entire power system. With the rapid development of the power industry, power plant equipment is evolving towards larger capacity, higher parameters, and greater complexity. Consequently, the losses caused by equipment failures are also increasing, thus placing higher demands on early warning systems for equipment failures.

[0003] The core idea of ​​existing equipment early warning systems focuses on the time-series characteristics of electrical quantities of individual devices, achieving anomaly detection by constructing a dynamic baseline for that device. However, critical equipment in actual power plants does not operate in isolation. Generators, main transformers, and power plant transformers are tightly coupled through the power grid, forming an interconnected and dynamically interactive energy and information transmission system. In this system, there are complex electromagnetic couplings, mechanical connections, and control loop relationships between devices. When an early, latent fault occurs in a device, it not only causes changes in its own electrical parameters but also triggers specific, subtle changes in the responsive electrical characteristics of other coupled devices through the aforementioned relationships.

[0004] Existing technologies analyze electrical quantity data for individual devices only, severing the inherent physical and electrical connections between devices. This fails to capture the cross-device, causal-related fault propagation characteristics, resulting in the loss of valuable early signals of fault propagation within the system. When multiple devices simultaneously exhibit minor anomalies, traditional single-point monitoring models struggle to distinguish between the fault-originating device and the affected devices, failing to provide maintenance personnel with accurate fault location guidance. This leads to insufficiently targeted maintenance work, increasing costs and potentially causing further escalation of the fault due to delayed intervention of the originating fault, thus impacting the safe and stable operation of the power plant.

[0005] Therefore, a method and system for early warning of equipment failure is needed. Summary of the Invention

[0006] To address the problem that existing technologies struggle to capture early fault characteristics across devices, this invention provides a device fault early warning method and system that can better capture early fault characteristics across devices. The specific technical solution is as follows: In a first aspect, embodiments of this application provide a method for early warning of equipment failure, including: Acquire time-series electrical variable data of each power device in a group of associated devices in a power system; wherein a direct electrical coupling relationship exists between the first and second devices in the group, and the first and second devices are any two power devices in the group; based on the time-series electrical variable data, construct a cross-device coupling feature vector for the group; wherein the first element of the cross-device coupling feature vector is used to characterize the physical coupling channel between the first and second devices; input the cross-device coupling feature vector into a normal coupling relationship graph model to calculate a coupling anomaly index; wherein the normal coupling relationship graph model is used to learn the normal coupling relationship pattern between the first and second devices under fault-free operating conditions; when the coupling anomaly index exceeds a warning threshold, identify the source device of the anomaly and the fault propagation path based on the cross-device coupling feature vector and the normal coupling relationship graph model; generate alarm information based on the coupling anomaly index, the source device of the anomaly, and the fault propagation path.

[0007] Preferably, the associated device group includes two or more power devices; the cross-device coupling feature vector also includes a second element, which is used to characterize the association features of all power devices in the associated device group.

[0008] Preferably, the first element includes a transfer function feature, a causal strength feature, and a symmetry deviation feature; the construction of the cross-device coupling feature vector of the associated device group based on the electrical variable time series data includes: calculating the transfer function feature based on the first electrical variable time series data of the first device and the second electrical variable time series data of the second device; wherein the first electrical variable time series data and the second electrical variable time series data have a frequency response relationship; calculating the transfer entropy as the causal strength feature based on the third electrical variable time series data of the first device and the fourth electrical variable time series data of the second device; wherein the third electrical variable time series data and the fourth electrical variable time series data have an information transmission relationship; when the first device and the second device are directly connected by three-phase electricity, calculating the symmetry deviation feature based on the three-phase electrical variable time series data of the first device and the three-phase electrical variable time series data of the second device; wherein the symmetry deviation feature is used to indicate the imbalance relationship of the three-phase symmetrical components caused by a fault.

[0009] Preferably, the normal coupling relationship graph model is a spatiotemporal graph variational autoencoder, which includes an encoder and a decoder. The encoder is used to map the heterogeneous graph constructed based on the cross-device coupling feature vector into a first latent variable. The heterogeneous graph includes nodes, edges, a first node feature matrix, and a first edge feature matrix. The node represents the power equipment in the associated device group, the edge represents the physical coupling channel corresponding to the first element, the first node feature matrix includes the state features of the power equipment in the associated device group, and the first edge feature matrix includes the first element. The first latent variable is used to characterize the coupling relationship pattern corresponding to the cross-device coupling feature vector. The decoder is used to reconstruct the second node feature matrix and the second edge feature matrix based on the distribution parameters of the first latent variable.

[0010] Preferably, the cross-device coupling feature vector is input into the normal coupling relationship graph model to calculate the coupling anomaly index, including: constructing the heterogeneous graph based on the cross-device coupling features; calculating the node reconstruction error based on the first node feature matrix and the second node feature matrix; calculating the latent variable anomaly score based on the distribution parameters of the first latent variable and the distribution parameters of the second latent variable corresponding to the normal coupling relationship pattern; and performing a weighted summation based on the node reconstruction error and the latent variable anomaly score to obtain the coupling anomaly index.

[0011] Preferably, the identification of the source device and fault propagation path based on the cross-device coupling feature vector and the normal coupling relationship graph model includes: selecting candidate abnormal edges based on the attention coefficient volatility of the edge obtained by the encoder during the mapping of the first latent variable, and the edge reconstruction error; calculating the cumulative abnormal contribution of each node from the nodes associated with the candidate abnormal edge through a multi-round iterative contribution backpropagation algorithm; determining the device corresponding to the node with the highest cumulative abnormal contribution as the source device of the abnormality; and generating the fault propagation path based on the source device of the abnormality and the candidate abnormal edge.

[0012] Preferably, the warning threshold includes a first threshold and a second threshold, wherein the second threshold is higher than the first threshold; the alarm information includes a first alarm information and a second alarm information; generating alarm information based on the coupling anomaly indicator, the anomaly source device, and the fault propagation path includes: outputting a first alarm information when the coupling anomaly indicator is greater than the first threshold and less than the second threshold, the first alarm information including a diagnostic report of the anomaly source device and the fault propagation path; outputting a second alarm information when the coupling anomaly indicator is greater than or equal to the second threshold, and simultaneously pushing an audible and visual alarm; wherein the second alarm information includes the diagnostic report and device protection suggestions, the device protection suggestions being generated based on the type of the anomaly source device and the coupling relationship anomaly characteristics revealed by the fault propagation path.

[0013] Secondly, embodiments of this application provide a device fault early warning system, applied to the method described in the first aspect, the system comprising: The acquisition module is used to acquire the electrical variable time-series data of each power device in the associated device group in the power system; wherein, there is a direct electrical coupling relationship between the first device and the second device in the associated device group, and the first device and the second device are any two power devices in the associated device group; The feature engineering module is used to construct a cross-device coupling feature vector for the associated device group based on the electrical variable time series data; wherein, the first element of the cross-device coupling feature vector is used to characterize the physical coupling channel between the first device and the second device; The anomaly detection module is used to input the cross-device coupling feature vector into the normal coupling relationship graph model to calculate the coupling anomaly index; wherein, the normal coupling relationship graph model is used to learn the normal coupling relationship pattern between the first device and the second device under fault-free operating conditions; The identification and positioning module is used to identify the source device of the anomaly and the fault propagation path based on the cross-device coupling feature vector and the normal coupling relationship graph model when the coupling anomaly index is greater than the warning threshold. The fault early warning module is used to generate alarm information based on the coupled abnormal indicator, the abnormal source device, and the fault propagation path.

[0014] Thirdly, embodiments of this application provide a computing device, including: a memory for storing a program; and a processor for loading the program to execute the method as described in the first aspect.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to perform the method described in the first aspect.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: by monitoring the changes in the coupling relationship between devices, it is possible to capture early signals of faults transmitted through cross-device coupling before the characteristics of the main device are obviously abnormal, and to provide early warning; by combining the graph model and causal analysis technology, it is possible to locate the source of the fault from the anomalies of multiple related devices and provide fault propagation path analysis to guide precise maintenance. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0018] Figure 1 A flowchart illustrating a device fault early warning method provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of an equipment fault early warning system provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0021] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0022] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0023] To address the problem that traditional methods struggle to capture early fault characteristics across devices, this invention provides a device fault early warning method and system that can better capture early fault characteristics across devices.

[0024] Please see Figure 1 , Figure 1This is a flowchart illustrating a device fault early warning method provided in an embodiment of this application. The method is applied to a computing device. Figure 1 As shown, the method includes: Step 101: The computing device obtains the time-series data of electrical variables of each power device in the associated equipment group in the power system.

[0025] The computing device can be a server or a terminal, specifically a server or terminal within a power system. It can connect to the equipment in the power system via wired or wireless means to directly or indirectly obtain the operating data of each device in the power system.

[0026] In this application, the embodiments focus on fault analysis and early warning based on the direct electrical coupling relationship between devices. Therefore, the power system can be a system where there is a strong correlation between devices such as power plants, substation systems, and distribution systems, and where faults are easy to propagate.

[0027] There is a direct electrical coupling relationship between the first device and the second device in the associated device group, and the first device and the second device are any two electrical devices in the associated device group.

[0028] Specifically, a group of devices with direct electrical coupling refers to a group of devices that are directly connected by metallic conductors, with no other devices in between that have significant impedance or energy conversion functions, and whose electrical state variables have an instantaneous and deterministic mutual constraint relationship through circuit laws.

[0029] The computing device can determine the associated equipment group based on the electrical main wiring diagram and physical connection relationship of the power system.

[0030] For example, the first device in the associated equipment group is a generator and the second device is a main transformer; or, the first device is a generator and the second device is an excitation transformer; or, the first device is a high-voltage station service transformer and the second device is a station service busbar; or, the first device is a main transformer and the second device is a high-voltage busbar.

[0031] Specifically, the computing device can obtain the time series of the core electrical variables that best reflect the state of different pairs of devices in the critical equipment group and participate in the coupling process. Specifically, the core electrical variables refer to the measurable basic electrical variables that directly participate in energy exchange and electromagnetic coupling between devices, as well as the electrical quantities derived directly from these basic electrical variables that are strongly correlated with the coupling process in the physical equations. The selection of core electrical variables does not depend on a specific equipment model, but rather on the coupling relationship reflected in the coupling physical model.

[0032] Specifically, the core electrical variables include basic electrical variables, derived electrical variables, and dedicated control electrical variables. Basic electrical variables include current and voltage; derived electrical variables are calculated in real time from the basic variables and are used to characterize specific forms or states of coupled energy; dedicated control electrical variables are used to characterize the internal control state of the equipment, and the control output directly affects the coupling relationship with other equipment.

[0033] For example, derived electrical variables include active power, reactive power, and magnetic flux density. When the coupling channel to be analyzed clearly points to a certain energy form or physical field, the computing device can select the corresponding derived electrical variables.

[0034] For example, dedicated control electrical variables include the generator's excitation current and excitation voltage, as well as the coupling coefficients of different tap positions on the transformer. These variables are specific to a particular device, and changes in the active output parameters or controllable parameters of that specific device can trigger changes in the coupling relationships.

[0035] In this embodiment, the primary purpose of selecting core electrical variables is not to evaluate individual devices, but to characterize the coupling channels between devices in pairs and groups. More specifically, the selection of core electrical variables needs to follow the principle of physical coupling channel mapping, including two electrical variables that are directly physically coupled, where an instantaneous change in variable A will directly and primarily cause a predictable change in variable B; and two electrical variables that represent the same physical quantity on the coupling path, where variables A and B are upstream and downstream measurements of the same physical quantity on the energy flow path.

[0036] For example, the computing device simultaneously collects the generator excitation voltage and the main transformer reactive power in order to analyze the specific coupling relationship of the effect of excitation control on reactive power coupling.

[0037] For example, taking the first device as a generator and the second device as a main transformer, the computing device can collect the generator's three-phase stator current, three-phase electronic voltage, excitation current, and excitation voltage; at the same time, it can collect the main transformer's high-voltage side three-phase current, three-phase voltage, low-voltage side three-phase current, and neutral point current.

[0038] After periodically acquiring the time-series data of each electrical variable within a time window, the computing device can perform time synchronization alignment, filtering, and quality verification on these data to obtain a set of multivariate synchronized data streams.

[0039] Step 102: The computing device constructs the cross-device coupling feature vector of the associated device group based on the time series data of the electrical variable.

[0040] The first element of the cross-device coupling feature vector characterizes the physical coupling channel between the first device and the second device. A physical coupling channel refers to the physical medium or structure that connects two devices and transmits energy or signals; it is the path through which fault characteristics and changes in electrical parameters are mutually transmitted between the devices.

[0041] Specifically, the first element may include multiple features between the first device and the second device, and the first elements of multiple device pairs in the associated device group constitute the cross-device coupling feature vector.

[0042] Preferably, the associated device group includes two or more power devices; the cross-device coupling feature vector also includes a second element, which is used to characterize the association features of all power devices in the associated device group.

[0043] Preferably, the first element includes a transfer function feature, a causal strength feature, and a symmetry deviation feature; the computing device can calculate the transfer function feature based on the first electrical variable time-series data of the first device and the second electrical variable time-series data of the second device; wherein, there is a frequency response relationship between the first electrical variable time-series data and the second electrical variable time-series data; based on the third electrical variable time-series data of the first device and the fourth electrical variable time-series data of the second device, the transfer entropy is calculated as the causal strength feature; wherein, the third electrical variable time-series data and the fourth electrical variable time-series data have an information transmission relationship; when the first device and the second device are directly connected by three-phase electricity, the symmetry deviation feature is calculated based on the three-phase electrical variable time-series data of the first device and the three-phase electrical variable time-series data of the second device; wherein, the symmetry deviation feature is used to indicate the imbalance relationship of the three-phase symmetrical components caused by a fault.

[0044] In this context, a frequency response relationship between two electrical variables refers to a linear or approximately linear dynamic coupling between them. A change in one electrical variable, A, will cause an observable change in the other, B, exhibiting specific frequency response characteristics. Frequency response characteristics mean that electrical variable B will respond differently to different frequency components of electrical variable A.

[0045] Among them, the information transmission relationship between two electrical variable sequences means that there is a statistically identifiable causal influence between the two sequences, including linear or nonlinear influences.

[0046] The transfer function characteristic is used to quantify the frequency response of two electrical variables within a specific frequency band, reflecting the frequency characteristics of the coupling path. Its calculation process is as follows: 1. Extract the N-point discrete sequence of the first electrical variable time series data within the current data window. N-point discrete sequence of second electrical variable time series data .

[0047] 2. Subtract the linear trend component from each sequence to eliminate the influence of the slowly changing DC offset on the frequency domain analysis; then apply the Hanning window function to the detrended sequences. This is to reduce spectral leakage. The calculation includes: ; ; ; .

[0048] 3. Perform a Discrete Fourier Transform on the windowed sequence. The calculation formula includes: ; Where k is the frequency index, and the corresponding actual frequency is... , is the sampling frequency; j is the imaginary unit.

[0049] 4. Calculate the autopower spectral density estimate of the discrete sequence of the first electrical variable. And the cross-power spectral density estimation between the discrete sequences of the first and second electrical variables. The calculation formula includes: in, express The complex conjugate, express The complex conjugate of ; U is the energy correction factor of the Hanning window function, with a value of 0.375.

[0050] 5. Calculate the estimated value of the transfer function H[k], which characterizes the strength and transmission characteristics of the causal relationship between two electrical variables in the frequency domain. The calculation formula includes: Then, to assess the reliability of the H[k] estimate, the amplitude squared coherence function is calculated: .

[0051] The amplitude-squared coherence function can determine the degree of linear coupling between two electrical variables at a certain frequency. Under healthy system conditions, the transfer function of a specific coupling path... The amplitude profile and phase curve in the characteristic frequency band are stable. When the equipment is in an abnormal state, the H[k] curve will change. The calculation equipment can select the power frequency, main characteristic harmonics, and system oscillation mode as key frequency points, and take these key frequency points in the H[k] curve. Corresponding amplitude Phase and coherence function values As eigenvalues, they are incorporated into the coupling feature vector as features of the transfer function.

[0052] Among these features, causal strength characteristics are used to quantify the strength and direction of the causal influence between two electrical variables. Specifically, the computing device can use transfer entropy as a measure of causal strength.

[0053] Specifically, to simplify probability estimation, the computing device can convert the time-series data of the third and fourth electrical variables into symbol sequences S and T. Using an equal-probability binning method, the time-series data of the third and fourth electrical variables are each divided into m intervals, ensuring that the number of data points in each interval is approximately equal. Each value in the sequence is replaced with the index of its corresponding interval.

[0054] Calculate the transfer entropy from source sequence S to target sequence T. ; Where n is the current time series, Let T be the sign value of the target sequence T at time n+1; , is the k-order historical symbol vector of the target sequence T at time n, where k is the embedding dimension of the target sequence T; , is the source sequence S in The l-th order history symbol vector at time step S, where l is the embedding dimension of the source sequence S; The joint probability distribution is obtained by counting and normalizing the occurrence counts of all possible state combinations within the entire data window. It is a conditional probability distribution. Transitive entropy. The larger the value, the stronger the causal influence of S on T.

[0055] Specifically, the embedding dimension is used to characterize how many consecutive times the state of a sequence at a given moment is collectively determined by its values ​​at a given time. The specific meaning of propagation entropy is the number of consecutive times the source sequence S has passed through. The state at time n will have a correlation with the state of the target sequence T at time n.

[0056] Under normal healthy conditions, the causal strength from the third electrical variable time series data to the fourth electrical variable time series data is within a typical range. If an early signal caused by a fault occurs, this causal strength may be abnormally enhanced or weakened. Preferably, the computing device can also simultaneously calculate the reverse propagation entropy. The ratio or difference between the forward and reverse propagation entropies can serve as a more robust feature of causal strength and can also be used to determine whether the dominant direction of the influence has changed.

[0057] Among them, the symmetry deviation feature is used to monitor the transmission relationship of three-phase imbalance between equipment and to capture changes in coupling characteristics caused by internal asymmetric faults in the equipment.

[0058] Specifically, the computing device can acquire the three-phase electrical variable time-series data of the first and second devices, and calculate the negative sequence component using the symmetrical component method. It is understood that the negative sequence component is 0 when the three-phase power system is operating stably, and the degree of three-phase imbalance can be quantified through the negative sequence component.

[0059] Then, the computing device can use the negative sequence component of the current from the first device as input. The negative sequence component of the current from the second device is used as the output. Calculate the multiple correlation coefficient between the two within the data window. First, calculate the estimated value of their cross-correlation function at zero delay. and their respective autocorrelation estimates , The calculation formula includes: ; Where N is the number of data points in the negative-order component. The conjugate operator for complex numbers.

[0060] Then, calculate the negative-order transfer complex gain. . It is a complex number whose amplitude reflects the attenuation / amplification ratio of the negative sequence current transmitted from the generator to the transformer, and its phase ∠ This reflects the phase shift determined by the transformer connection group and system impedance. When the equipment is healthy and the connections are fixed, It should remain basically constant.

[0061] When the negative sequence current of the first device is very small, the second device retains a negative sequence current level that would not normally exist. The computing device can calculate this negative sequence residual. To detect the asymmetry generated by the transformer itself, the calculation formula is: ; in, This is the rated current of the transformer, used for per-unit standardization.

[0062] The first minor asymmetric fault inside the equipment will lead to... Increase, but It should remain unchanged; if Abnormal changes in amplitude or phase are more likely to indicate changes in the state of transformer windings or connection points. This cross-device symmetric correlation characteristic is extremely sensitive to early, subtle internal faults. Monitoring the transfer complex gain of the negative sequence component between two devices can help identify this. and residue It can identify early-stage equipment malfunctions.

[0063] Preferably, the correlation characteristics represented by the second element include the partial coherence function or conditional transfer entropy of the active power of the corresponding device, which can be used to determine whether the power fluctuation has formed an abnormal resonance or abnormal energy distribution in the local network composed of the associated device group.

[0064] Then, the computing device can combine the above features of each pair of first and second devices in a predetermined order into a first element, and then combine the first elements of all device pairs, or the first elements of all device pairs into a second element, to construct the cross-device coupling feature vector.

[0065] Step 103: The computing device inputs the cross-device coupling feature vector into the normal coupling relationship graph model to calculate the coupling anomaly index.

[0066] The normal coupling relationship graph model is used to learn the normal coupling relationship pattern between the first device and the second device under fault-free operating conditions. The computing device can first use historical power system operation data under fault-free operating conditions to train the initial graph model, allowing it to learn the coupling relationship pattern that electrical variables should have between normal devices under fault-free operating conditions.

[0067] Preferably, the normal coupling relationship graph model is a spatiotemporal graph variational autoencoder, which includes an encoder and a decoder. The computing device can first convert the corresponding cross-device coupling feature vectors in the historical operating data of the power system into heterogeneous graphs, with each heterogeneous graph corresponding to a cross-device coupling feature vector within a time window.

[0068] The heterogeneous graph includes nodes, edges, a first node feature matrix, and a first edge feature matrix. A node represents a power device in the associated device group; an edge is a directed edge representing the physical coupling channel corresponding to the first element, pointing from the first device to the second device; the first node feature matrix includes the state characteristics of the power device corresponding to the node, including univariate statistics such as the moving average and standard deviation of the effective voltage value, the total harmonic distortion rate of the current, and the percentage of negative sequence current; the first edge feature matrix includes the first element, with each row corresponding to one edge, and the elements in each row being the feature values ​​of a first element. Node features will serve as attributes of the corresponding node, and edge features will follow the same principle.

[0069] Preferably, when the cross-device coupling feature vector includes a second element, the second element will serve as a structural constraint in the graph model.

[0070] This structured representation allows the model to explicitly distinguish between node features that represent the state of a device itself and edge features that represent the interaction between devices, providing a correct and efficient input format for learning the complex joint distribution between the two.

[0071] The encoder is used to map the heterogeneous graph constructed based on the cross-device coupling feature vector into a low-dimensional first latent variable, which is used to characterize the coupling relationship pattern corresponding to the cross-device coupling feature vector; the decoder is used to reconstruct the second node feature matrix and the second edge feature matrix based on the distribution parameters of the first latent variable.

[0072] Specifically, the spatiotemporal graph variational autoencoder can not only model the dependencies of graph structure data between devices, but also its dynamic evolution patterns over time.

[0073] Specifically, the spatiotemporal graph variational autoencoder includes a spatial graph encoder and a temporal encoder. The spatial graph encoder uses a multi-layer graph attention network to aggregate neighbor information and learn a context-aware representation of each node.

[0074] The representation of node i at layer l Its update formula is: ; ; in, Let i be the set of neighbors of node i. This is a vector concatenation operation. For the features of node i in the (l-1)th layer, It equals the original feature vector of node i. Let be the feature vector of the edge connecting nodes i and j. Let L be the learnable weight matrix of the l-th layer; Let be the normalized attention coefficient of the l-th layer, representing the importance of node j to node i; Let be the transpose of the learnable weight vectors of the attention mechanism. , This is the activation function.

[0075] After L-layer message passing, we obtain the final representation of each node. Then, through a global pooling layer, the spatial representation vector of the entire graph is obtained. .

[0076] Among them, attention coefficient The physical meaning of this can be interpreted as the dynamic weight of the coupling strength between devices. Under healthy conditions, the attention coefficient of a given edge should be relatively stable. If the attention coefficient of a given edge... Dramatic fluctuations occur, even if their characteristics Even a small change could indicate a shift in the dominant coupling path, and this attention coefficient can serve as a diagnostic signal.

[0077] The timing encoder can be a gated loop unit. To capture the dynamic evolution of coupling relationships, the computing device can input a sequence of spatial representation vectors for multiple consecutive time windows into the timing encoder. Then, the hidden state of the last time step output by the timing encoder is used. Timing encoding serves as the timing context information for the entire time series.

[0078] Then, the computing device can represent the space. and timing coding After concatenation, the resulting vector is input into two parallel fully connected layers, which output latent variables respectively. mean vector Sum of logarithmic variance vector .

[0079] Then, reparameterized sampling is used to obtain the latent variables: ,in, For noise that is normally distributed, This is for element-wise multiplication.

[0080] The decoder aims to reconstruct the original node feature matrix and edge feature matrix, which forces... It must contain enough information to reconstruct the entire coupling graph.

[0081] Specifically, the decoder also includes a gated recurrent unit layer and a spatial graph attention network layer, with a structure roughly symmetrical to the encoder but in the opposite direction. First, the gated recurrent unit layer... Based on the reconstruction state from the previous time step, a decoding temporal context is generated. Then, a graph generation network generates the reconstruction features of each node and each edge in parallel according to this temporal context.

[0082] Then, the decoder can calculate the mean squared error of the original features and the reconstructed features as the reconstruction error, and use this reconstruction error to construct a loss function to train the graph model. After training, the parameters of the encoder and decoder are fixed, forming a normally coupled graph model.

[0083] Preferably, after the normal coupling relationship graph model is put into use, the computing device can construct the heterogeneous graph based on the cross-device coupling characteristics, and calculate the first node feature matrix, the second node feature matrix, the first edge feature matrix, the second edge feature matrix, as well as the first latent variable and its distribution parameters; then, based on the first node feature matrix and the second node feature matrix, the node reconstruction error is calculated; based on the distribution parameters of the first latent variable and the distribution parameters of the second latent variable corresponding to the normal coupling relationship pattern, the latent variable anomaly score is calculated; and the node reconstruction error and the latent variable anomaly score are weighted and summed to obtain the coupling anomaly index.

[0084] Among them, node reconstruction error Anomalies used to measure the state characteristics of a single device are calculated using the following formulas: ; in, Let be the actual value of the d-th dimension state feature of the i-th node in the real-time graph. The d-th dimension state feature value of the i-th node reconstructed from the model; Let d be the standard deviation of the d-th dimension node features over the entire training set; The total number of nodes. This is a node feature dimension. When this value is too large, it indicates that the operating state of a certain device has deviated from the normal mode.

[0085] Among them, latent variable anomaly scoring The formula used to measure anomalies in the coupling patterns of the entire system includes: ; in, The posterior mean vector output by the encoder for the current sequence is a representation of the center point of the current coupling relationship pattern in the latent space; This is the mean vector of the latent variable distribution of the health data calculated during the training phase. It is the inverse of the covariance matrix of the latent variable distribution of health data calculated during the training phase.

[0086] Then, the two indicators mentioned above can be merged into a comprehensive coupled anomaly indicator. The calculation formula includes: ; in, To integrate the weighting coefficients, and These are the mean and standard deviation of the health data reconstruction error calculated during the training phase, respectively. and These are the mean and standard deviation of the latent variable anomaly scores calculated during the training phase, respectively.

[0087] By adding an anomaly score—a latent variable that quantifies the deviation of the entire interaction network from the healthy baseline—to the data reconstruction error of a single device, a system-level perspective can be provided to observe and monitor device malfunctions from a global viewpoint.

[0088] Step 104: When the coupling anomaly index exceeds the warning threshold, the computing device identifies the source device of the anomaly and the fault propagation path based on the cross-device coupling feature vector and the normal coupling relationship graph model.

[0089] The early warning threshold can be obtained by statistically analyzing the coupled abnormal index values ​​of various faults in the early, middle, and late stages from historical data. Specifically, the early warning threshold can include multi-level early warning thresholds, and the computing device can set different early warning actions and protection actions for different levels of early warning.

[0090] Once an early warning is issued, the computing device needs to locate the root cause of the anomaly and the path of fault propagation, providing decision-makers with effective handling suggestions to address the anomaly effectively and as early as possible.

[0091] Preferably, the computing device can filter out candidate anomalous edges based on the attention coefficient volatility of the edge obtained by the encoder during the mapping of the first latent variable, as well as the edge reconstruction error; starting from the node associated with the candidate anomalous edge, calculate the cumulative anomalous contribution of each node through a multi-round iterative contribution backpropagation algorithm; determine the device corresponding to the node with the highest cumulative anomalous contribution as the source device of the anomalous edge; and generate the fault propagation path based on the source device of the anomalous edge and the candidate anomalous edge.

[0092] The computing device can calculate the ratio of the standard deviation to the mean of the attention coefficient of each edge as the volatility of the attention coefficient. If the volatility of the attention coefficient is higher than the historical baseline value by more than the anomaly threshold, and the edge feature reconstruction error of the corresponding edge ranks in the top K of the global edge reconstruction errors, the corresponding edge is determined to be a candidate anomaly edge.

[0093] Specifically, K can be the percentile of the total number of edges, such as the median or quartile.

[0094] Then, for each node connected by a candidate anomalous edge, its initial anomalous contribution is set to the node feature reconstruction error of that node; then, multiple rounds of iterative propagation are performed. In the k-th iteration, for each candidate anomalous edge, the contribution is propagated from the affected target node to the source node that exerted the influence, based on the anomalousness and direction of the edge's features. The cumulative anomalous contribution is then calculated. The formula for calculation is: ; in, Let i be the cumulative anomaly contribution of node i after the k-th iteration. For the set of candidate abnormal edges, This is the propagation attenuation factor, representing the attenuation of the contribution during the propagation process; The anomaly weight of the edge is the product of the reconstruction error of the edge feature and the attention volatility.

[0095] The meaning of this backpropagation process is that if a node's own state is abnormal, and it strongly influences other nodes that also exhibit abnormalities through abnormally significant edges, then the node's contribution will continuously accumulate in the iteration.

[0096] After a preset number of iterations, the computing device can determine the cumulative anomaly contribution. The device corresponding to the highest one or more nodes is the source device of the anomaly.

[0097] Then, starting from the source node, one or more of the most likely fault propagation paths can be generated along the paths that propagate outward from its contribution and belong to the candidate anomaly edges.

[0098] Step 105: The computing device generates alarm information based on the coupling anomaly indicator, the anomaly source device, and the fault propagation path.

[0099] The computing device can transform the relevant conclusions about abnormal coupling relationships obtained from the aforementioned steps into executable hierarchical early warning operations and specific relay protection system collaborative strategies.

[0100] Preferably, the warning threshold includes a first threshold and a second threshold, wherein the second threshold is higher than the first threshold; the alarm information includes a first alarm information and a second alarm information; when the coupling anomaly index is greater than the first threshold and less than the second threshold, the computing device can output the first alarm information, which includes a diagnostic report of the anomaly source device and the fault propagation path; when the coupling anomaly index is greater than or equal to the second threshold, the second alarm information is output, and an audible and visual alarm is pushed simultaneously; wherein the second alarm information includes the diagnostic report and device protection suggestions, which are generated based on the type of the anomaly source device and the abnormal coupling characteristics revealed by the fault propagation path.

[0101] Among them, the equipment protection recommendation is a temporary adjustment suggestion for the settings of one or more protection devices.

[0102] Specifically, the computing device can determine the set of protection devices directly affected by abnormal coupling relationships based on the fault propagation path; then, based on the abnormal edge with the highest contribution and its specific abnormal characteristics, it can generate adjustment strategies; finally, it can package the adjustment strategies for all protection devices in the set of protection devices into a protection coordination strategy package, check the logical conflicts within the strategy package, and arbitrate according to the preset priority rules to generate a final consistent device protection recommendation.

[0103] In this embodiment, by monitoring changes in electrical variables that characterize the coupling relationship between devices, the earliest signal transmitted by the fault through the system coupling characteristics can be captured before the characteristics of the main device are obviously abnormal, thus providing an early warning time. Combined with graph models and causal analysis techniques, the initial source of the fault can be located from the anomalies of multiple related devices, and fault propagation path analysis can be provided to guide precise maintenance.

[0104] The method provided in the embodiments of this application has been described above. The system provided in the embodiments of this application will be described below.

[0105] Please see Figure 2 , Figure 2 This is a schematic diagram of the structure of a device fault early warning system provided in an embodiment of this application, as shown below. Figure 2 As shown, the system 20 includes: The acquisition module 201 is used to acquire the electrical variable time series data of each power device in the associated equipment group in the power system; wherein, there is a direct electrical coupling relationship between the first device and the second device in the associated equipment group, and the first device and the second device are any two power devices in the associated equipment group; The feature engineering module 202 is used to construct a cross-device coupling feature vector for the associated device group based on the electrical variable time series data; wherein, the first element of the cross-device coupling feature vector is used to characterize the features of the physical coupling channel between the first device and the second device; The anomaly detection module 203 is used to input the cross-device coupling feature vector into the normal coupling relationship graph model to calculate the coupling anomaly index; wherein, the normal coupling relationship graph model is used to learn the normal coupling relationship pattern between the first device and the second device under fault-free operating conditions; The identification and positioning module 204 is used to identify the abnormal source device and the fault propagation path based on the cross-device coupling feature vector and the normal coupling relationship map model when the coupling anomaly index is greater than the warning threshold. The fault warning module 205 is used to generate alarm information based on the coupled abnormal indicator, the abnormal source device and the fault propagation path.

[0106] The equipment fault early warning system provided in this application embodiment can be understood by referring to the relevant content in the foregoing method embodiment section, and will not be repeated here.

[0107] like Figure 3 As shown, Figure 3This is a schematic diagram of a possible logical structure of a computing device provided in an embodiment of this application. The computing device 30 includes a processor 301, a communication interface 302, a memory 303, and a bus 304. The processor 301, the communication interface 302, and the memory 303 are interconnected via the bus 304. In an embodiment of this application, the processor 301 is used to control and manage the operation of the computing device 30. For example, the processor 301 is used to execute... Figure 1 The steps in the embodiments and / or other processes used in the techniques described herein. Communication interface 302 is used to support communication by computing device 30. Memory 303 is used to store program code and data of computing device 30.

[0108] The processor 301 can be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. The processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a digital signal processor and a microprocessor, etc. The bus 304 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0109] In another embodiment of this application, a computer-readable storage medium is also provided, the computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the above-described... Figure 1 The method described in the embodiments.

[0110] Those skilled in the art will recognize that the units of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

[0111] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0112] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0113] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0114] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0115] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A device failure early warning method, characterized by, The method comprises: obtaining electrical variable time series data of each power device in an associated device group in a power system; wherein there is a direct electrical coupling relationship between a first device and a second device in the associated device group, and the first device and the second device are any two power devices in the associated device group; based on the electrical variable time series data, constructing a cross-device coupling feature vector of the associated device group; wherein a first element of the cross-device coupling feature vector is used to represent the characteristics of the physical coupling channel between the first device and the second device; inputting the cross-device coupling feature vector into a normal coupling relationship graph model to calculate a coupling anomaly index; wherein the normal coupling relationship graph model is used to learn the normal coupling relationship mode of the first device and the second device under fault-free working conditions; in the case where the coupling anomaly index is greater than a warning threshold, identifying an abnormal source device and a fault propagation path based on the cross-device coupling feature vector and the normal coupling relationship graph model; generating an alarm information based on the coupling anomaly index, the abnormal source device and the fault propagation path.

2. The method of claim 1, wherein, The associated device group comprises 2 or more power devices; the cross-device coupling feature vector further comprises a second element, and the second element is used to represent the associated characteristics of all power devices in the associated device group.

3. The method of claim 1, wherein, The first element comprises a transfer function feature, a causal strength feature and a symmetry deviation feature; the cross-device coupling feature vector of the associated device group is constructed based on the electrical variable time series data, comprising: based on the first electrical variable time series data of the first device and the second electrical variable time series data of the second device, the transfer function feature is calculated; wherein the first electrical variable time series data and the second electrical variable time series data have a frequency response relationship; based on the third electrical variable time series data of the first device and the fourth electrical variable time series data of the second device, the transfer entropy is calculated as the causal strength feature; wherein the third electrical variable time series data and the fourth electrical variable time series data have an information transmission relationship; in the case where the first device and the second device are directly connected by three-phase electricity, the symmetry deviation feature is calculated based on the three-phase electrical variable time series data of the first device and the three-phase electrical variable time series data of the second device; wherein the symmetry deviation feature is used to indicate the unbalanced relationship of three-phase symmetry components caused by faults.

4. The method according to any one of claims 1 to 3, characterized in that, The normal coupling relationship graph model is a spatio-temporal graph variational autoencoder, and the normal coupling relationship graph model comprises an encoder and a decoder; the encoder is used to map a heterogeneous graph constructed based on the cross-device coupling feature vector into a first latent variable, the heterogeneous graph comprising nodes, edges, a first node feature matrix and a first edge feature matrix, the nodes representing power devices in the associated device group, the edges representing physical coupling channels corresponding to the first elements, the first node feature matrix comprising state features of power devices in the associated device group, and the first edge feature matrix comprising the first elements; The first latent variable is used to represent a coupling relationship mode corresponding to the cross-device coupling feature vector; The decoder is used to reconstruct a second node feature matrix and a second edge feature matrix based on a distribution parameter of the first latent variable.

5. The method of claim 4, wherein, The inputting of the cross-device coupling feature vector into the normal coupling relationship graph model includes: constructing the heterogeneous graph based on the cross-device coupling feature; calculating a node reconstruction error based on the first node feature matrix and the second node feature matrix; calculating a latent variable anomaly score based on a distribution parameter of the first latent variable and a distribution parameter of a second latent variable corresponding to the normal coupling relationship mode; performing weighted summation based on the node reconstruction error and the latent variable anomaly score to obtain the coupling anomaly index.

6. The method of claim 4, wherein, The identification of the abnormal source device and the fault propagation path based on the cross-device coupling feature vector and the normal coupling relationship graph model includes: filtering candidate abnormal edges based on an edge attention coefficient fluctuation rate obtained by the encoder in the process of mapping the first latent variable and an edge reconstruction error; calculating a cumulative abnormal contribution degree of each node through a multi-round iteration contribution degree back propagation algorithm from the nodes associated with the candidate abnormal edges; determining a device corresponding to a node with the highest cumulative abnormal contribution degree as the abnormal source device; generating the fault propagation path based on the abnormal source device and the candidate abnormal edges.

7. The method according to any one of claims 1-3, characterized in that, The pre-warning threshold includes a first threshold and a second threshold, and the second threshold is higher than the first threshold; the alarm information includes first alarm information and second alarm information; The generation of the alarm information based on the coupling anomaly index, the abnormal source device and the fault propagation path includes: in a case where the coupling anomaly index is greater than the first threshold and less than the second threshold, outputting first alarm information, the first alarm information containing a diagnostic report of the abnormal source device and the fault propagation path; in a case where the coupling anomaly index is greater than or equal to the second threshold, outputting second alarm information and synchronously pushing an audible and visual alarm; wherein the second alarm information includes the diagnostic report and a device protection suggestion, and the device protection suggestion is generated based on a type of the abnormal source device and coupling relationship anomaly characteristics revealed by the fault propagation path.

8. A device failure early warning system, characterized by, The system is applied to the method of any one of claims 1-7, and the system includes: an acquisition module configured to acquire electrical variable time series data of each power device in a group of associated devices in a power system; wherein a direct electrical coupling relationship exists between a first device and a second device in the group of associated devices, and the first device and the second device are any two power devices in the group of associated devices; a feature engineering module configured to construct a cross-device coupling feature vector of the group of associated devices based on the electrical variable time series data; wherein a first element of the cross-device coupling feature vector is used to represent a feature of a physical coupling channel between the first device and the second device. An anomaly detection module is configured to input the cross-device coupling feature vector into a normal coupling relationship graph model to calculate a coupling anomaly index, wherein the normal coupling relationship graph model is configured to learn a normal coupling relationship mode of the first device and the second device under a fault-free working condition. An identification positioning module is configured to identify an abnormal source device and a fault propagation path based on the cross-device coupling feature vector and the normal coupling relationship graph model when the coupling anomaly index is greater than a pre-warning threshold. A fault pre-warning module is configured to generate an alarm information based on the coupling anomaly index, the abnormal source device and the fault propagation path.

9. A computing device, comprising: The computer readable storage medium comprises a program stored therein, and when the program is executed, the computer readable storage medium controls a device where the computer readable storage medium is located to execute the method in any one of claims 1-7. The computer readable storage medium comprises a program stored therein, and when the program is executed, the computer readable storage medium controls a device where the computer readable storage medium is located to execute the method in any one of claims 1-7. ​ 10. A computer-readable storage medium, characterized in that, ​

Images