A power grid oscillation source detection method and system based on a spatial Bayesian model

Abstract

The application discloses a power grid oscillation source detection method and system based on a space Bayesian model, and the method comprises the following steps: collecting and preprocessing electrical characteristic time sequence signals of each node in a power grid; extracting node time-frequency characteristic of the preprocessed electrical characteristic time sequence signals; constructing a power grid node space Bayesian prior model according to a topological line connection relationship between nodes in the power grid to obtain a prior probability of a node having an oscillation risk; constructing a comprehensive likelihood function of the node as an oscillation source according to the node time-frequency characteristic; and performing Bayesian fusion calculation on a posterior probability distribution of each node as an oscillation source by using the prior probability of the node having the oscillation risk and the comprehensive likelihood function of the node as the oscillation source. The application aims to give the most possible oscillation source position based on the space Bayesian model, realize probability statistics of observation nodes in the whole network and each region, and provide quantitative power grid oscillation source detection uncertainty information for dispatchers.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of oscillation source detection in power systems, and particularly relates to a power grid oscillation source detection method and system based on a spatial Bayesian model. BACKGROUND

[0002] With the deployment of synchronized phasor measurement units (PMUs), waveform measurement units (WMUs), and wide-area measurement systems (WAMS), more and more nodes in power systems can collect high-precision voltage and current measurement data in real time. Based on these data, a relatively mature oscillation monitoring technology system has been established, mainly including: 1) oscillation detection, i.e., using spectral analysis, time-frequency analysis, wavelet analysis, modal analysis, etc. to identify whether there is sustained oscillation or low-damping mode; 2) oscillation feature extraction, i.e., extracting oscillation frequency, damping ratio, amplitude, duration, etc.; 3) alarm and record: once the oscillation is detected, an alarm is generated according to the preset threshold, and the relevant data is recorded and archived. In the field of oscillation monitoring, common time-frequency analysis methods include but are not limited to: short-time Fourier transform (STFT), wavelet transform (WT), wavelet packet analysis, S transform and its variants, filter bank and band-pass filtering combined with Hilbert transform. These methods can show the energy distribution of the signal in the time-frequency plane, thus depicting the occurrence time, duration, and frequency change of the oscillation. When multiple nodes are analyzed simultaneously, the oscillation energy change process of each node in a certain frequency band, the dynamic change of the phase difference between nodes over time, and the rough estimation of the energy diffusion or propagation pattern between nodes can be obtained. Around the oscillation source positioning, existing technologies can be roughly divided into the following categories: 1) methods based on model and modal analysis, i.e., based on the linearized model of the system and eigenvalue analysis, the characteristic vectors and participation factors of each mode are calculated to find the most contributing bus or node to a certain oscillation mode; there are also some methods that combine damping sensitivity analysis to evaluate the impact of different nodes or devices on damping. 2) methods based on energy balance or power flow, i.e., using generalized energy function or power imbalance theory to calculate energy injection points and absorption points during oscillation; by analyzing the direction of energy flow, the area where the oscillation source is located is determined. 3) measurement-driven methods based on phase difference and amplitude ratio, i.e., using the voltage / current phase measured by PMU to construct "phase gradient" or "phase cone" criteria, and using the relative size and decay rate of oscillation amplitude at different nodes for empirical judgment. 4) methods based on machine learning or deep learning, i.e., using a large amount of simulation or historical disturbance data to train a classification model (e.g., "a node is a source / non-source"); using convolutional neural networks, recurrent networks, or graph neural networks to fit complex mapping relationships. In addition to power systems, spatial statistics and spatial Bayesian models have been widely used in epidemiology, environmental science, and traffic safety. Typical representatives are: ICAR (intrinsic conditional autoregressive) model: based on the adjacency relationship of regions, the difference between adjacent regions is constrained to form a spatially smooth random field; BYM (Besag-York-Mollié) model: on the basis of ICAR, unstructured random effects are introduced to mix structured spatial effects and independent noise, which is used to model the disease incidence rate, accident risk, etc. in each region.Its typical features are: using "adjacency matrix W" to describe the spatial correlation; regarding "risk" as a spatially varying random variable; and integrating covariates (such as social and economic indicators) and historical data to output the risk distribution of each region. Therefore, how to introduce the spatial Bayesian model into power grid oscillation source assessment for the first time is still a key technical problem to be solved. SUMMARY

[0003] The technical problem solved by the present application: In view of the above problems of the prior art, the present application provides a power grid oscillation source detection method and system based on a spatial Bayesian model, which aims to detect power grid oscillation sources based on a spatial Bayesian model, give the most likely oscillation source location, realize probability statistics of observation nodes in the whole network and each region, and provide dispatchers with quantitative uncertainty information of power grid oscillation source detection.

[0004] To solve the above technical problems, the technical scheme adopted by the present application is:

[0005] A power grid oscillation source detection method based on a spatial Bayesian model, comprising the following steps:

[0006] S1, data acquisition and preprocessing of electrical characteristic time series signals of each node in the power grid;

[0007] S2, node time-frequency feature extraction on the preprocessed electrical characteristic time series signals;

[0008] S3, constructing a spatial Bayesian prior model of the power grid nodes according to the topological line connection relationship between the nodes to obtain the prior probability of node i having an oscillation risk ; constructing a comprehensive likelihood function of the node as an oscillation source according to the node time-frequency features;

[0009] S4, using the prior probability of node i having an oscillation risk and the comprehensive likelihood function of the node as an oscillation source to perform Bayesian fusion calculation on the posterior probability distribution of each node as an oscillation source.

[0010] Optionally, the preprocessing in step S1 includes time alignment, filtering and denoising, removing trend components, and normalizing the amplitude, wherein the removing trend components is subtracting the average value of the electrical characteristic time series signals in a specified size time window from the electrical characteristic time series signals.

[0011] Optionally, when performing node time-frequency feature extraction in step S2, the extracted node time-frequency features include energy spectrum, response time difference and phase difference, wherein the response time difference is the time difference between the time when the energy spectrum of the node and the nodes affected by it first exceeds a preset threshold E th at the target frequency.

[0012] Optionally, the step S3 of constructing the power grid node space Bayesian prior model according to the topological line connection relationship between the power grid nodes comprises:

[0013] S3.1, generating an adjacency matrix according to the topological line connection relationship between the power grid nodes:

[0014] ;

[0015] wherein, is an ith row and jth column element in the adjacency matrix ;

[0016] S3.2, defining a spatial structured effect for each node i to reflect the spatial smoothness of the risk of the node i with adjacent nodes, the smoothness constraint of the spatial structured effect is:

[0017] ;

[0018] wherein, is the smoothness constraint of the spatial structured effect , and denotes the spatial structured effect, and are spatial structured effects of nodes i and j respectively; and the smoothness constraint of the spatial structured effect of the node i corresponds to a Gaussian prior:

[0019] ;

[0020] wherein, is the Gaussian prior of the spatial structured effect , and is a variance of the spatial structured effect; defining an unstructured effect for each node i to depict the independent difference of the node itself and modeling by using an independent Gaussian distribution of the following formula:

[0021] ;

[0022] wherein, denotes an independent Gaussian distribution with a mean of 0 and a variance of ;

[0023] S3.3, for each node i, determining a definition of the overall prior risk value of the node i according to the spatial structured effect and the unstructured effect of the node i:

[0024] ;

[0025] where, is the overall prior risk value of node i, is a scaling factor, is the weight coefficient of structured and unstructured effects, and has ;

[0026] S3.4, estimate the spatial structured effect of each node i using the historical oscillation event data of the power grid , unstructured effect , weight coefficient , variance of spatial structured effect , and variance of independent Gaussian distribution , get the prior probability of each node i having oscillation risk .

[0027] Optionally, the function expression of constructing the comprehensive likelihood function of node as an oscillation source according to the node time-frequency characteristics in step S3 is:

[0028] ;

[0029] ;

[0030] ;

[0031] ;

[0032] where, is the comprehensive likelihood function of node i, , and are the likelihood function components of energy spectrum, response time difference and phase difference respectively, , and are the weight coefficients of energy spectrum, response time difference and phase difference respectively; is the observed energy of node i at the main frequency of oscillation, is the expected energy level of the source node, is the variance of energy spectrum, is the variance of response time difference, is the first time when the energy spectrum of node i and the affected node j at the target frequency exceeds the preset threshold E th , is the variance of phase difference, is the phase difference between node i and the affected node j, is the phase difference between node i and the affected node j, is the expected value of the phase difference between node i and the affected node j.

[0033] Optionally, the function expression of the Bayesian fusion calculation of each node as an oscillation source in step S4 is:

[0034] ;

[0035] wherein, is the posterior probability of node i being an oscillation source given data, data is the node time-frequency features of each node in the power grid and the oscillation starting time; is the prior probability of node i having an oscillation risk, is the prior probability of node j having an oscillation risk, is the comprehensive likelihood function of node i, is the comprehensive likelihood function of node j, and N is the number of nodes in the power grid.

[0036] Optionally, after step S4, it further includes visualizing the posterior probability distribution of each node as an oscillation source, and the visualization result includes part or all of the node ranking table, the topology probability graph, the regional probability graph and the multi-source identification graph, the node ranking table includes a sorted list of the posterior probability of each node i being an oscillation source, the topology probability graph is an image showing the posterior probability of each node i being an oscillation source in the power grid node topology graph, the regional probability graph is an image of the posterior probability of each node i being an oscillation source in a given region, and the multi-source identification graph is an image of one or more nodes i marked as having a posterior probability of being an oscillation source exceeding a preset threshold.

[0037] The application further provides a power grid oscillation source detection system based on a spatial Bayesian model, which comprises a microprocessor and a memory connected to each other, and the microprocessor is programmed or configured to execute the power grid oscillation source detection method based on the spatial Bayesian model.

[0038] The application further provides a computer readable storage medium, which stores a computer program or instructions programmed or configured to execute the power grid oscillation source detection method based on the spatial Bayesian model through a processor.

[0039] The application further provides a computer program product, which comprises a computer program or instructions programmed or configured to execute the power grid oscillation source detection method based on the spatial Bayesian model through a processor.

[0040] Compared with the prior art, the present application mainly has the following beneficial effects: the present application takes the spatial Bayesian model as the core, comprehensively utilizes the power grid topology structure, historical oscillation statistical information and real-time time-frequency characteristics, and outputs the power grid oscillation source probability of "oscillation source posterior probability distribution" under the unified Bayesian inference framework, which has the following advantages: 1) the present application can solve the problem of "only single-point positioning and lack of uncertainty description". Through the construction of the spatial Bayesian model + node likelihood function, the posterior probability of all nodes in the network is output, the oscillation source probability distribution map is formed, and the quantitative expression and ordering of the uncertainty of the source position are realized. 2) the present application can solve the problem of "ignoring spatial correlation and failing to fully utilize the power grid topology information". By introducing the adjacency matrix W and the ICAR / BYM spatial Bayesian model, the risk of adjacent nodes is modeled in the form of a spatial random field, the risk is smoothed and propagated in space, and thus the stability and robustness of the source positioning result are improved. 3) the present application can solve the problem of "unable to utilize long-term historical data to form a robust priori". The activity level, fault type and working condition information of each node in the historical oscillation event can be mapped into model parameters and covariates to construct an updateable node oscillation risk priori, so that the system can learn from historical experience and be updated in real time. 4) the present application can solve the problem of "weak multi-source oscillation recognition ability". Through the multi-peak structure of the Bayesian posterior distribution in the node space, the multi-source scene is naturally supported: multiple nodes can have a high posterior probability at the same time, and the system can identify and display multiple suspected source areas and their relative likelihoods. 5) the present application can improve the positioning robustness in the weak oscillation, short-time oscillation and high-frequency oscillation scenarios. By reasonably describing the relationship between the oscillation signal and the noise in the likelihood function and combining the spatial priori for constraint, even if the signal of a single event is weak, a relatively reliable source probability evaluation can still be obtained through "priori + neighborhood information + cumulative history". 6) the present application can establish a unified "priori-likelihood-posterior" Bayesian inference logic closed loop. The spatial Bayesian model is taken as the priori, the node oscillation response characteristics are taken to construct the likelihood function, and the Bayesian formula is taken to perform fusion inference, so that the relationship between the power grid topology, historical events and real-time data is uniformly described in mathematics, and a standardized probability output is provided for subsequent risk assessment and pre-control strategy optimization. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 It is a basic flowchart of the embodiment method of the present application.

[0042] Figure 2 It is a module division schematic diagram of the embodiment method of the present application. DETAILED DESCRIPTION

[0043] The core idea of the present application is: on the basis of multi-node synchronous measurement, modeling the oscillation risk of the nodes of the power grid as a "spatial random field" constrained by the topological structure, forming an a priori risk distribution through historical oscillation data, combining the time-frequency response characteristics of each node in a specific oscillation event to construct a likelihood function, and finally obtaining the posterior probability distribution of each node as an oscillation source by using the Bayes formula. In order to enable personnel in the technical field to better understand the technical solutions of the present application, the technical solutions of the present application will be further described in detail below in combination with the drawings in the embodiments of the present application.

[0044] As shown in Figure 1 , the power grid oscillation source detection method based on the spatial Bayesian model in the present embodiment includes the following steps:

[0045] S1, data acquisition and preprocessing of the electrical characteristic time series signals of each node in the power grid;

[0046] S2, node time-frequency feature extraction on the preprocessed electrical characteristic time series signals;

[0047] S3, constructing a spatial Bayesian prior model of the nodes of the power grid according to the topological line connection relationship between the nodes of the power grid to obtain the a priori probability that node i has an oscillation risk ; constructing a comprehensive likelihood function of the node as an oscillation source according to the node time-frequency features;

[0048] S4, using the a priori probability that node i has an oscillation risk and the comprehensive likelihood function of the node as an oscillation source to perform Bayesian fusion calculation on the posterior probability distribution of each node as an oscillation source.

[0049] As shown in Figure 2As shown, the power grid oscillation source detection method based on the spatial Bayesian model in the embodiment is specifically divided into a data acquisition module (for performing data acquisition of step S1), a data processing module (for performing preprocessing of step S1), a time-frequency feature extraction module (for performing step S2), a spatial prior modeling module (for constructing a power grid node spatial Bayesian prior model of step S3), a likelihood calculation module (for constructing a comprehensive likelihood function of step S3), a Bayesian fusion module (for performing step S4), and a model updating module. After an event ends, the model updating module updates the spatial prior model according to new event records, so that the system has long-term self-learning ability. The modules work as follows: the data acquisition module is connected to the power grid synchronous measurement system, and acquires data periodically or when an event is triggered; the data preprocessing module cleans and uniformly formats the data; the time-frequency feature extraction module can be implemented based on CPU or GPU, and is suitable for real-time or quasi-real-time calculation requirements; the spatial prior modeling module can be run offline, and regularly updates parameters by using historical oscillation events; the likelihood calculation module is run online after each oscillation event occurs; the Bayesian fusion module usually has small calculation amount, and can be completed within seconds; the source probability evaluation module displays the probability results to dispatchers or sends them to other security control modules; and the model updating module updates the spatial prior model according to new event records after an event ends, so that the system has long-term self-learning ability.

[0050] In step S1 of the embodiment, when the electrical characteristic time sequence signals of each node in the power grid are collected, the data acquisition module is connected to the synchronous measurement device in the power grid, including a PMU (synchronous phasor measurement unit) and a WMU (waveform measurement unit), and the voltage and current time sequence data of each node in the power grid are taken as the electrical characteristic time sequence signals by the synchronous phasor measurement unit PMU and the waveform measurement unit WMU. The collected original electrical characteristic time sequence signals can be denoted as:

[0051] ;

[0052] wherein, i is the node number, t is the time, N is the total number of monitoring nodes.

[0053] The preprocessing in step S1 of the embodiment includes time alignment, filtering and noise removal, removing trend components, and normalizing the amplitude. The removing trend components is subtracting the average value (DC component) of the electrical characteristic time series signal in a specified size time window from the electrical characteristic time series signal. The electrical characteristic time series signal can be a voltage or current signal, etc. Specifically, in order to ensure the accuracy of subsequent analysis, the original signal is processed by the data preprocessing module as follows: time alignment: using GPS or other precise clock sources, aligning the data of each node to a unified time reference; filtering and noise removal: using band-pass filtering, wavelet denoising, etc., to suppress power frequency deviation, DC bias and high frequency noise; trend removal and normalization: removing trend components, normalizing the amplitude, and making the oscillation characteristics of different nodes comparable. The signal obtained after preprocessing is denoted as , which is the input of subsequent time-frequency analysis.

[0054] In step S2 of the embodiment, the extracted node time-frequency characteristics include energy spectrum, response time difference and phase difference. The response time difference is the time difference between the time when the energy spectrum of the node and the nodes affected by it first exceeds the preset threshold E th at the target frequency.

[0055] In the embodiment, the time-frequency feature extraction module performs time-frequency transformation on each node signal to obtain its energy distribution and phase information in the time-frequency plane. The present application does not limit the specific time-frequency analysis method, which can be short-time Fourier transform (STFT), wavelet transform (WT), S transform, etc. Taking S transform as an example, the time-frequency representation of the node i can be denoted as:

[0056] ;

[0057] wherein, is the complex-valued time-frequency coefficient of the node i at frequency f and time t .

[0058] The oscillation energy spectrum and the oscillation phase trajectory are defined as:

[0059]

[0060] ;

[0061] In specific calculations, the present application can select a group of frequencies of interest according to the oscillation frequency band (such as low frequency, subsynchronous or high frequency), and extract the features at these frequencies.

[0062] To reflect the relative response between nodes, the phase difference and energy difference between nodes are further calculated in the embodiment, for example:

[0063] ;

[0064] ;

[0065] wherein is the current oscillation main frequency or frequency band center. By analyzing and over time, the possible propagation direction information of the source node can be obtained.

[0066] In addition, the oscillation "starting time" of each node is extracted, that is, the node i is defined as the time when the energy on the frequency first exceeds a certain threshold E th : Thus, the response time difference between nodes can be obtained:

[0067] ;

[0068] The above energy spectrum, phase difference and response time difference features jointly constitute the input of the likelihood function in step S3.

[0069] The spatial prior modeling module constructs the spatial Bayesian model of "node oscillation risk" through the power grid topology information. In step S3 of the embodiment, the spatial Bayesian prior model of the power grid node is constructed according to the topological line connection relationship between the power grid nodes, including:

[0070] S3.1, generating an adjacency matrix according to the topological line connection relationship between the power grid nodes:

[0071] ;

[0072] wherein, is the i-th row and j-th element in the adjacency matrix ;

[0073] S3.2, ICAR (Intrinsic Conditional Auto Regression) is used in the embodiment to describe the spatial correlation, and two random effects are introduced: the spatial structured effect is defined for each node i to reflect the smoothness of the risk of node i in space with the adjacent nodes, and the smoothness constraint of the spatial structured effect is:

[0074] ;

[0075] wherein, is the spatial structured effectthe smoothness constraint of , denotes the spatial structure effect, and are the spatial structure effects of nodes i and j, respectively; and the spatial structure effect of node i corresponds to a Gaussian-type prior:

[0076] ;

[0077] where, is the Gaussian-type prior of the spatial structure effect , is the variance of the spatial structure effect; and the unstructured effect is defined for each node i to characterize the independent difference of the node itself and is modeled by an independent Gaussian distribution with:

[0078] ;

[0079] where, denotes an independent Gaussian distribution with mean 0 and variance ;

[0080] S3.3, for each node i, the definition of the overall prior risk value of node i is determined according to its spatial structure effect and unstructured effect :

[0081] ;

[0082] where, is the overall prior risk value of node i, is a scaling factor, is a weight coefficient of the structured and unstructured effects, and has ;

[0083] S3.4, the spatial structure effect , the unstructured effect , the weight coefficient , the variance of the spatial structure effect , and the variance of the independent Gaussian distribution of each node i are estimated using the historical oscillation event data of the power grid, to obtain the prior probability that each node i has an oscillation risk. The prior can be automatically updated by the historical oscillation events.

[0084] Assuming node i is the source of this oscillation, it should exhibit specific characteristics in terms of energy, phase, and response timing. Therefore, likelihood function components based on oscillation energy, response time, and phase difference are constructed. In this embodiment, the likelihood function components of the energy spectrum are:

[0085] ;

[0086] in, Let i be the observed energy at the dominant oscillation frequency. The expected energy level of the source node (which can be obtained statistically from historical source events or adaptively estimated). The variance of the energy spectrum;

[0087] In this embodiment, the likelihood function component of the response time difference is:

[0088] ;

[0089] in, The variance of the response time difference, divide and sum The energy spectrum of node i and its affected node j at the target frequency exceeds the preset threshold E for the first time. th At the moment when multiple nodes are significantly later than node i When oscillation begins, it will increase. The value of .

[0090] In oscillation propagation, the phase of nodes near the source typically exhibits a certain gradient characteristic. A phase error metric can be defined for a specific moment or time window; therefore, in this embodiment, the likelihood function components of the phase difference are:

[0091] ;

[0092] in, The variance of the phase difference. Let be the phase difference between node i and the node j affected by it. Let be the expected value of the phase difference between node i and the affected node j, that is, the expected pattern of the phase difference between node i and node j when node i is the source, obtained from historical events or simulations.

[0093] In step S3, the expression for the comprehensive likelihood function of the nodes as oscillation sources, constructed based on the time-frequency characteristics of the nodes, is as follows:

[0094] ;

[0095] in, Let be the comprehensive likelihood function for node i. , and are the likelihood function components of the energy spectrum, response time difference and phase difference, respectively, , and are the weight coefficients of the energy spectrum, response time difference and phase difference, respectively. The weight coefficients can be determined in combination with historical data or experience.

[0096] The Bayesian fusion module fuses the prior probability P(i) obtained in step S3 with the likelihood function L(i) to calculate the posterior probability of each node as an oscillation source. In step S4 of the embodiment, the function expression of the posterior probability distribution of each node as an oscillation source is calculated by Bayesian fusion:

[0097] ;

[0098] wherein, is the posterior probability of node i as an oscillation source under the given data condition, and data is the node time-frequency characteristics and the oscillation starting time of each node in the power grid; is the prior probability of node i having an oscillation risk, is the prior probability of node j having an oscillation risk, is the comprehensive likelihood function of node i, is the comprehensive likelihood function of node j, and N is the number of nodes in the power grid. Based on the above Bayesian fusion reasoning method, when the historical risk of a certain node is high (the prior is large), and the response mode of the current event also conforms to the source characteristics (the likelihood is large), the posterior probability of the node will be significantly improved; if the historical risk of a certain node is high, but the current response is obviously inconsistent with the source characteristics, then the likelihood term will “correct” its probability, avoiding blind reliance on the prior; if the signal of the current event is weak and the noise is large, then the prior will have a higher proportion in the posterior, and the system can still output a smoother source probability distribution, improving the robustness.

[0099] The oscillation source probability evaluation module processes and displays the posterior probability distribution. For example, Figure 1As shown, as an optional embodiment, the embodiment step S4 further includes visualizing the posterior probability distribution of each node as an oscillation source, and the visualization results include part or all of a node ranking table, a topology probability graph, a region probability graph, and a multi-source identification graph. The node ranking table includes a ranked list of the posterior probability of each node i as an oscillation source. The topology probability graph is an image showing the posterior probability of each node i as an oscillation source in a power grid node topology graph. The region probability graph is an image of the posterior probability of each node i as an oscillation source in a given region. The multi-source identification graph is an image of one or more nodes i marked as having a posterior probability of being an oscillation source exceeding a preset threshold. The embodiment not only gives the most likely oscillation source location, but also gives the probability ranking of each observation node in the entire network and the region probability graph, thereby providing quantitative uncertainty information for dispatch personnel. For example, in the embodiment, the node ranking table, the topology probability graph, the region probability graph, and the multi-source identification graph are generated in the following manner: 1) Node ranking table: arrange all nodes in descending order of posterior probability, and list the top several as the focus of investigation; 2) Topology probability graph: on the power grid topology graph, use the size or grayscale depth of the node circle to represent the probability size, forming an “oscillation source probability heat map”; 3) Region probability graph: attribute nodes to different regions (such as plants, sub-areas, and provinces), sum or weighted average the node probabilities in each region, thereby generating oscillation source probabilities at the regional level; 4) Multi-source identification graph: when the posterior distribution has multiple obvious peaks, multiple high-probability nodes or regions can be marked at the same time, indicating the possible existence of multi-source oscillation. Through the above output, not only the most likely source location is given, but also complete probability distribution information is provided, providing more comprehensive and quantitative decision-making basis for dispatch personnel.

[0100] In summary, the prior art mostly makes deterministic judgments on single oscillation events, either relying on idealized system models and modal analysis or following single energy, phase or threshold rules, which cannot systematically utilize historical information and power grid topology and cannot quantify uncertainty, and thus are prone to misjudgment or unstable results in complex scenarios such as multi-source oscillation, weak oscillation and high-frequency oscillation. The power grid oscillation source detection method based on a spatial Bayesian model in the embodiment first constructs a spatial Bayesian prior model based on the power grid topology, converts long-term statistics and structural information such as “a node has a high incidence of long-term oscillation and is strongly coupled with surrounding nodes” into a calculable node risk prior; then, a “node as a source” likelihood function is constructed through multi-node time-frequency characteristics, reflecting the matching degree of the actual response of each node in the event and the physical behavior of the source node. Since the prior and the likelihood are fused in a unified Bayesian framework, the system can balance between historical information and real-time information, and even if the signal of a single event is weak or the noise is large, a stable source probability evaluation can be obtained through the prior and the neighborhood constraint. The power grid oscillation source detection method based on a spatial Bayesian model in the embodiment has the following technical effects: 1) The traditional single oscillation source “deterministic positioning” problem is upgraded to a new paradigm of “oscillation source probability evaluation” of all nodes in the grid. The embodiment outputs the probability distribution of each node as an oscillation source, as well as node ranking and regional probability map, rather than only giving a judgment result of a source node. 2) The oscillation risk of the grid node is first regarded as a “spatial random field”, and the spatial correlation between nodes is constructed based on the power grid topology structure. By introducing structured spatial effects and unstructured random effects, a spatial Bayesian prior distribution of the node risk is formed, and the prior can be continuously updated using historical oscillation events. 3) On the basis of multi-node synchronous measurement, the oscillation energy, phase change, oscillation starting time and other time-frequency characteristics of each node are used to construct a node likelihood function of “the reasonable degree of the current observation when a node is the oscillation source of this time”. 4) The spatial Bayesian model is used to give the node risk prior, and the node time-frequency characteristics are used to construct the likelihood function, and the posterior probability of each node as an oscillation source is obtained through Bayesian updating. This framework unifies the power grid topology information, historical oscillation statistical information and real-time oscillation response information. 5) The probability framework of the method in the embodiment does not forcibly assume “a unique oscillation source”, allows the posterior distribution to present multiple high-probability nodes or regions, and naturally supports multi-source oscillation identification. At the same time, with the help of spatial prior and neighborhood information constraint, even if the signal of a single event is weak or the noise is large, a relatively stable source probability estimate can be obtained, and the adaptability to complex scenarios such as weak oscillation, short-time oscillation and high-frequency power electronic oscillation is improved. Further, the power grid oscillation source detection method based on a spatial Bayesian model in the embodiment outputs the posterior probability distribution of each node as an oscillation source in the grid, rather than a hard decision of a single source node.The probabilistic output directly corresponds to the actual decision needs in scheduling and operation: on the one hand, the dispatcher can sort and respond to nodes or regions according to the probability; on the other hand, when there are multi-source oscillations or regional oscillations, the embodiment does not forcibly "choose one point", but presents the potential source region in the form of multiple high-probability peaks, so as to be more in line with the actual situation of the oscillation mechanism in the complex power grid. In summary, through the complete technical scheme of spatial Bayesian prior, time-frequency likelihood modeling and Bayesian fusion, the embodiment overcomes the problems of insufficient utilization of spatial correlation, inability to quantify uncertainty, difficulty in judging in a multi-source scenario, and poor robustness of weak oscillations in the prior art, and realizes more reliable, more interpretable, and self-learning evolution of the power grid oscillation source probability evaluation capability.

[0101] In addition, the embodiment also provides a power grid oscillation source detection system based on a spatial Bayesian model, which comprises a microprocessor and a memory connected with each other, the microprocessor is programmed or configured to execute the power grid oscillation source detection method based on the spatial Bayesian model. The application also provides a computer readable storage medium, which stores a computer program or instructions, the computer program or instructions are programmed or configured to execute the power grid oscillation source detection method based on the spatial Bayesian model by a processor. The application also provides a computer program product, which comprises a computer program or instructions, the computer program or instructions are programmed or configured to execute the power grid oscillation source detection method based on the spatial Bayesian model by a processor.

[0102] Those skilled in the art should understand that the technical scheme provided by the application can be in the form of a method, a system, or a computer program product. Therefore, the application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the application can be in the form of a computer program product implemented on one or more computer readable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer usable program codes. The application is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices produce a machine that implements the functions described in the flowcharts and / or block diagrams. Figure 1 one flow or multiple flows and / or blocks Figure 1apparatuses that implement the functions specified in one or more flowcharts and / or blocks. These computer program instructions can also be loaded onto a computer or other programmable data processing devices to cause a series of operational steps to be performed on the computer or other programmable devices to produce a computer implemented process such that the instructions that are executed on the computer or other programmable devices provide steps for implementing the functions specified in the flowcharts and / or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 one or more flowcharts and / or blocks Figure 1 one or more flowcharts and / or blocks Figure 1 one or more flowcharts and / or blocks

[0103] The above description is merely preferred embodiments of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical solutions falling within the concept of the present application shall be considered as falling within the protection scope of the present application. It should be noted that, for those of ordinary skill in the art, some improvements and refinements without departing from the principles of the present application shall also be considered as falling within the protection scope of the present application.

Claims

1. A method for power grid oscillation source detection based on spatial Bayesian model, characterized in that, The method comprises the following steps: S1, collecting and preprocessing the electrical characteristic time series signals of each node in the power grid; S2, extracting the node time-frequency characteristics from the preprocessed electrical characteristic time series signals; S3, constructing a power grid node space Bayesian prior model according to a topological line connection relationship between grid nodes to obtain a prior probability that node i has an oscillation risk ; constructing a comprehensive likelihood function of a node as an oscillation source according to node time-frequency characteristics; S4, using the prior probability that node i has the risk of oscillation and the integrated likelihood function of the node as the oscillation source, the posterior probability distribution of each node as the oscillation source is calculated by Bayesian fusion. The step S3 of constructing the power grid node space Bayesian prior model according to the topological line connection relationship between the nodes comprises: S3.1, generating an adjacency matrix according to the topological line connection relationship between the nodes: ； wherein is the adjacency matrix is the element in the ith row and jth column of S3.2 Define a spatial structuring effect for each node i to reflect that the risk of node i is spatially smoothed with respect to neighboring nodes. The smoothing constraint for the spatial structuring effect of node i is: ​ ； where, the smoothness constraint for the spatial structuring effect, the smoothness constraint for the spatial structuring effect, the spatial structuring effect, and respectively the spatial structuring effect of nodes i and j; and the spatial structuring effect of node i the smoothness constraint for the spatial structuring effect corresponds to a Gaussian-type prior: ； where, for the spatially structured effects Gaussian prior, for the spatially structured effects; and for each node i define an unstructured effect to capture the independent variation of the nodes themselves and model with an independent Gaussian distribution given by ； wherein, represents an independent Gaussian distribution with mean 0 and variance of 1. S3.3 For each node i, determine its overall prior risk value, according to its spatially structured effects and unstructured effects Definition of the overall prior risk value of node i: ； wherein, is the overall prior risk value for node i, is a scaling factor, is a weight coefficient for structured and unstructured effects, and has ; S3.4, estimating the spatially structured effect for each node i using historical oscillation event data of the power grid , unstructured effect , weight coefficient , variance of spatially structured effect , and variance of independent Gaussian distribution , obtaining the prior probability that each node i has an oscillation risk ; The function expression of the step S3 of constructing the comprehensive likelihood function of the node as an oscillation source according to the node time-frequency characteristics is: ； ； ； ； in, Let be the comprehensive likelihood function for node i. , and These are the likelihood function components for the energy spectrum, response time difference, and phase difference, respectively. , and These are the weighting coefficients for the energy spectrum, response time difference, and phase difference, respectively. Let i be the observed energy at the dominant oscillation frequency. The expected energy level of the source node. Let Variance be the variance of the energy spectrum. The variance of the response time difference, divide and sum The energy spectrum of node i and its affected node j at the target frequency exceeds the preset threshold E for the first time. th At that moment, The variance of the phase difference. Let be the phase difference between node i and the node j affected by it. Let be the expected value of the phase difference between node i and node j affected by it.

2. The method for power grid oscillation source detection based on spatial Bayesian model according to claim 1, characterized in that, The preprocessing in the step S1 comprises time alignment, filtering and denoising, removing the trend component, and normalizing the amplitude, wherein the removing the trend component is subtracting the average value of the electrical characteristic time series signals in a specified size time window from the electrical characteristic time series signals.

3. The method for power grid oscillation source detection based on spatial Bayesian model according to claim 1, characterized in that, The node time-frequency features extracted in step S2 include energy spectrum, response time difference and phase difference, wherein the response time difference is the time difference between the time when the energy spectrum of the node and the nodes affected by the node first exceeds a preset threshold E th at the target frequency.

4. The method for power grid oscillation source detection based on spatial Bayesian model according to claim 1, characterized in that, The function expression of the step S4 of performing Bayesian fusion to calculate the posterior probability distribution of each node as an oscillation source is: ； wherein, is the posterior probability that node i is the source of oscillation under given data, data is the node time-frequency features of each node in the power grid and the oscillation start time; is the prior probability that node i has the risk of oscillation, is the prior probability that node j has the risk of oscillation, is the comprehensive likelihood function of node i, is the comprehensive likelihood function of node j, and N is the number of nodes in the power grid.

5. The method for power grid oscillation source detection based on spatial Bayesian model according to claim 1, characterized in that, After the step S4, the method further comprises visualizing the posterior probability distribution of each node as an oscillation source, and the visualization result comprises part or all of a node ranking table, a topological probability graph, a regional probability graph, and a multi-source identification graph, wherein the node ranking table comprises a ranking list of the posterior probability of each node i as an oscillation source, the topological probability graph is an image showing the posterior probability of each node i as an oscillation source in a power grid node topology graph, the regional probability graph is an image of the posterior probability of each node i as an oscillation source in a given region, and the multi-source identification graph is an image of one or more nodes i whose posterior probability as an oscillation source exceeds a preset threshold.

6. A power grid oscillation source detection system based on spatial Bayesian model, comprising a microprocessor and a memory connected to each other, characterized in that, The microprocessor is programmed or configured to perform the power grid oscillation source detection method based on the spatial Bayesian model according to any one of claims 1-5.

7. A computer-readable storage medium having stored therein a computer program or instructions, characterized in that, The computer program or instructions are programmed or configured to perform the power grid oscillation source detection method based on the spatial Bayesian model according to any one of claims 1-5 by the processor.

8. A computer program product comprising computer programs or instructions, characterized in that, The computer program or instructions are programmed or configured to perform the power grid oscillation source detection method based on the spatial Bayesian model according to any one of claims 1-5 by the processor.

Images