Power grid fault positioning and intelligent relay protection selective control method and system

Abstract

The invention provides a power grid fault positioning and intelligent relay protection selective control method and system, and relates to the technical field of intelligent power grids, and the method comprises the steps: collecting the multi-point electrical data of a power grid, adjusting a sampling window and a filtering threshold value according to the operation state of the power grid, and carrying out the data processing; after eliminating the end effect, performing empirical mode decomposition to obtain a fault feature component; inputting the characteristic component into a fault diagnosis model, calculating a fault initial probability based on a Bayesian network, generating multiple groups of fault prediction data in combination with a time sequence evolution rule, and taking a position corresponding to a minimum information entropy value as a fault positioning result; determining a fault area according to a positioning result, establishing a local communication network of relay protection devices in the area, obtaining the operation state of each device, calculating a protection sensitivity coefficient, determining a protection range and a tripping time sequence, generating and sending a control instruction, and realizing selective isolation of the fault area.

Description

Technical Field

[0001] The present invention relates to the field of smart grid technology, and in particular to a method and system for selectively controlling grid fault location and smart relay protection. Background Art

[0002] When a fault occurs in a traditional power grid, the relay protection device mainly relies on the local electrical quantity measurement to perform fault judgment and tripping operations. As the scale of the power grid continues to expand, distributed power sources and new loads are connected in large numbers, and the fault characteristics of the power grid are becoming more diverse and complex. At the same time, the requirements of the power grid for power supply reliability and operation stability are constantly increasing, requiring the relay protection system to have higher fault identification accuracy and isolation selectivity.

[0003] In the prior art, relay protection devices usually use fixed protection settings and tripping sequences, lacking the ability to perceive the operating status of the power grid in real time and to adjust it adaptively. Due to the lack of effective information interaction and coordination mechanisms between the relay protection devices, problems such as overlapping protection ranges, protection blind spots, false operations, and refusal to operate are prone to occur. Especially under complex fault conditions, it is difficult for a single protection device to accurately identify the nature and location of the fault, which often leads to an excessively large protection action range and unnecessary expansion of the power outage range. In addition, traditional fault location methods are mainly based on steady-state quantity analysis, and insufficient use is made of transient characteristic information, resulting in low positioning accuracy in complex scenarios such as multiple faults and fault transfers.

[0004] In summary, there is an urgent need for a power grid fault location and intelligent relay protection selective control method, which can monitor the power grid operation status in real time and extract features to establish an accurate fault location model; quickly build a local communication network based on the fault location results to achieve information sharing of relay protection devices in the fault area; dynamically adjust the protection range and coordination sequence based on the real-time operation status of each relay protection device to improve the selective control capability of the relay protection system. The present invention can solve the problems in the prior art. Summary of the invention

[0005] The embodiments of the present invention provide a method and system for power grid fault location and intelligent relay protection selective control, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] Provided is a power grid fault location and intelligent relay protection selective control method, comprising:

[0008] Collect electrical data of multiple measurement units in the power grid, dynamically detect the operation status of the power grid, adjust the sampling window time length according to the operation status of the power grid, calculate the adaptive filtering threshold, filter the electrical data according to the adaptive filtering threshold to obtain filtered data, perform boundary extension on the filtered data to eliminate endpoint effects, perform empirical mode decomposition operation, and decompose to obtain multi-layer fault characteristic component data;

[0009] The multi-layer fault characteristic component data is input into the fault diagnosis model, a causal relationship chain between power grid equipment is established based on the Bayesian network, and the initial probability of the fault is calculated; according to the initial probability of the fault, combined with the preset time series evolution rules, multiple groups of fault prediction data are generated; the information entropy values ​​of the multiple groups of fault prediction data are calculated, and the position corresponding to the fault prediction data with the minimum information entropy value is selected as the fault location result;

[0010] Determine the fault area that needs protection based on the fault location result, establish a local communication network in the fault area, and connect all relay protection devices in the fault area; obtain the real-time operating status of each relay protection device through the local communication network, calculate the protection sensitivity coefficient of each relay protection device, determine the protection range and tripping sequence of each relay protection device based on the protection sensitivity coefficient, and generate a relay protection control instruction; send the relay protection control instruction to each relay protection device through the local communication network, control each relay protection device to perform a tripping operation, and realize selective isolation of the fault area.

[0011] In an optional embodiment,

[0012] Collect electrical data of multiple measurement units in the power grid, dynamically detect the operation status of the power grid, adjust the sampling window time length according to the operation status of the power grid, calculate the adaptive filtering threshold, filter the electrical data according to the adaptive filtering threshold to obtain filtered data, perform boundary extension on the filtered data to eliminate endpoint effects, perform empirical mode decomposition operation, and decompose to obtain multi-layer fault feature component data including:

[0013] The electrical data includes voltage data, current data and phase angle data, which are respectively normalized to obtain reference electrical data;

[0014] Setting an initial sampling window length, calculating the variance value of the reference electrical data within the sampling window, obtaining a voltage variance value, a current variance value, and a phase angle variance value, and multiplying the voltage variance value, the current variance value, and the phase angle variance value by corresponding preset weight coefficients respectively and summing them to obtain a comprehensive state index;

[0015] Detecting the change trend of the comprehensive state indicator over time, calculating the first-order change rate and the second-order change rate of the comprehensive state indicator, substituting the first-order change rate and the second-order change rate into a preset window adjustment function, dynamically adjusting the time length of the sampling window, and obtaining an adaptive sampling window length;

[0016] Extracting the deviation degree of the voltage variance value, the current variance value and the phase angle variance value relative to their respective preset reference states, substituting the deviation degree into a preset exponential decay function, and generating an adaptive threshold value;

[0017] Based on the adaptive threshold, a characteristic mutation interval in the reference electrical data is identified, and based on the characteristic mutation interval, data segments at both ends of the reference electrical data, each accounting for 10% of the total signal length, are intercepted as data segments to be processed; local maximum points and local minimum points are detected in the data segments to be processed to obtain a local extreme point sequence, and the time difference and amplitude difference between adjacent local extreme points in the local extreme point sequence are respectively calculated to obtain a characteristic time series and a characteristic amplitude series;

[0018] Based on the characteristic time series and the characteristic amplitude series, construct a periodic variation function and an amplitude variation function of the extended signal, and combine the periodic variation function and the amplitude variation function to generate an extended processing function;

[0019] At the left end of the reference electrical data, a first group of extended data is generated to the left according to the extended processing function; at the right end of the reference electrical data, a second group of extended data is generated to the right according to the extended processing function; the first group of extended data, the reference electrical data and the second group of extended data are smoothly connected to obtain boundary extended processing data;

[0020] An empirical mode decomposition operation based on cubic spline interpolation is performed on the boundary extension processing data to obtain multiple intrinsic mode components, component characteristic parameters of the multiple intrinsic mode components are calculated, and the multiple intrinsic mode components are screened according to the component characteristic parameters to obtain multi-layer fault characteristic component data.

[0021] In an optional embodiment,

[0022] Performing an empirical mode decomposition operation based on cubic spline interpolation on the boundary extension processing data to obtain multiple intrinsic mode components, calculating component characteristic parameters of the multiple intrinsic mode components, and screening the multiple intrinsic mode components according to the component characteristic parameters to obtain multi-layer fault characteristic component data includes:

[0023] Taking the boundary extension processing data as the data to be processed, start envelope screening:

[0024] Searching for all extreme points in the data to be processed, fitting the extreme points using cubic spline interpolation to generate an upper envelope and a lower envelope; calculating the arithmetic mean of the upper envelope and the lower envelope to generate an average line, subtracting the average line from the boundary extension processing data to generate a candidate component;

[0025] Searching for all zero-crossing points and local extreme value points in the candidate components, obtaining the number of zero-crossing points and the number of local extreme value points by counting, and calculating the absolute value of the difference between the number of zero-crossing points and the number of local extreme value points; performing envelope analysis on the candidate components, and calculating the symmetry of the upper envelope and the lower envelope of the candidate components about the time axis;

[0026] When the absolute value is less than or equal to 1 and the symmetry is greater than a preset symmetry threshold, the candidate component is determined to meet the conditions of the intrinsic mode function; when the absolute value is greater than 1 or the symmetry is less than or equal to the preset symmetry threshold, the candidate component is used as new data to be processed, and the envelope screening is repeatedly performed until the conditions of the intrinsic mode function are met, completing a round of envelope screening, and the candidate component is recorded as the first intrinsic mode function;

[0027] Subtracting the first intrinsic mode function from the boundary extension processing data to generate a first residual signal; searching for all local extreme points in the first residual signal and counting the number of the local extreme points; when the number of local extreme points is greater than a preset threshold, taking the first residual signal as new data to be processed, repeatedly performing envelope screening until a new candidate component that meets the conditions of the intrinsic mode function is obtained, and recording the new candidate component as a new intrinsic mode function; subtracting the new intrinsic mode function from the new data to be processed to generate a new residual signal, until the number of local extreme points is less than or equal to the preset threshold, ending the decomposition process, recording the new residual signal as the last intrinsic mode function, and obtaining multiple intrinsic mode functions;

[0028] Calculate the energy entropy values ​​of multiple intrinsic mode functions respectively to generate an energy entropy value sequence; calculate the mutual correlation coefficients of multiple intrinsic mode functions and boundary extension processing data respectively to generate a mutual correlation coefficient sequence; multiply each energy entropy value in the energy entropy value sequence by a first preset weight coefficient, multiply each mutual correlation coefficient in the mutual correlation coefficient sequence by a second preset weight coefficient, sum the multiplied results, and generate a feature importance index corresponding to each intrinsic mode function;

[0029] According to the numerical value of the characteristic importance index, a preset number of intrinsic mode functions are selected in descending order to generate a sequence of intrinsic mode functions to be analyzed; the time domain characteristic parameters, frequency domain characteristic parameters and time-frequency characteristic parameters of each intrinsic mode function in the sequence of intrinsic mode functions to be analyzed are extracted to generate multi-layer fault characteristic component data.

[0030] In an optional embodiment,

[0031] Inputting the multi-layer fault characteristic component data into the fault diagnosis model, establishing a causal relationship chain between power grid equipment based on the Bayesian network, and calculating the initial probability of the fault; generating multiple groups of fault prediction data according to the initial probability of the fault in combination with the preset time series evolution rule; calculating the information entropy values ​​of the multiple groups of fault prediction data, and selecting the position corresponding to the fault prediction data with the minimum information entropy value as the fault location result includes:

[0032] Constructing a three-layer Bayesian network, inputting the time domain characteristic parameters in the multi-layer fault characteristic component data into the first layer nodes, inputting the frequency domain characteristic parameters and time-frequency characteristic parameters in the multi-layer fault characteristic component data into the second layer nodes, and setting the power grid equipment nodes as the third layer nodes;

[0033] Obtaining the feature mean of the second-layer node, calculating the Euclidean distance between the time-domain feature parameter and the feature mean, multiplying the Euclidean distance by a preset attenuation coefficient, taking the negative exponent, and then multiplying the Euclidean distance by a preset adjustment coefficient to obtain a first conditional probability between the first-layer node and the second-layer node;

[0034] Extract the feature-fault correspondence in the historical fault case data, establish a second conditional probability matrix between the second-layer nodes and the third-layer nodes, and record the probability that the feature of each second-layer node triggers the fault of each power grid device; calculate the posterior probability of the second-layer node under the preset time domain feature parameters, multiply the posterior probability by the second conditional probability matrix, and obtain the initial probability of failure of each power grid device;

[0035] Based on the physical connection relationship between the power grid devices, determine the fault propagation probability between the power grid devices and establish a time series evolution matrix; set the initial fault probability as an initial state vector, multiply the initial state vector by the time series evolution matrix to obtain a state vector at the next moment, repeat the execution, and generate multiple sets of fault prediction data at different moments;

[0036] Extract the failure probability of each power grid device in each group of fault prediction data, calculate the product of each fault probability and the corresponding natural logarithm, sum and take the negative value, and obtain the information entropy value of each group of fault prediction data; select the fault prediction data corresponding to the smallest information entropy value, determine the position of the power grid device with the largest failure probability in the corresponding fault prediction data, and determine the fault location result.

[0037] In an optional embodiment,

[0038] The initial probability of the fault is set as an initial state vector, the initial state vector is multiplied by the time series evolution matrix to obtain a state vector at the next moment, and the process is repeated to generate multiple groups of fault prediction data at different moments; including:

[0039] Collecting the operating condition data of the power grid equipment, extracting the load change cycle, temperature parameters, and humidity parameters from the operating condition data, and setting the evolution time threshold and time step according to the operating condition data;

[0040] Based on the load change cycle, the load change rate is calculated, a load disturbance function is established, and the load change rate is input into the load disturbance function to obtain a load disturbance coefficient; the temperature parameter and the humidity parameter are input into a preset environmental model to calculate the environmental disturbance coefficient;

[0041] Start state evolution calculation, set the initial fault probability as the initial state vector, multiply the initial state vector by the time series evolution matrix to obtain a first state vector; multiply the first state vector by the load disturbance coefficient and the environmental disturbance coefficient respectively to obtain a second state vector; multiply the second state vector by the time series evolution matrix to obtain a third state vector, and execute state evolution calculation repeatedly until the evolution time threshold is reached to generate a state vector sequence;

[0042] Setting the number of Monte Carlo iterations, repeatedly executing the state evolution calculation at each time step, and obtaining multiple groups of state vector sequences;

[0043] Calculate the mean and variance of multiple groups of state vector sequences at each of the time steps, set confidence parameters, calculate the upper limit and lower limit of the confidence interval based on the mean and the variance, and retain the state vector sequences within the confidence interval; organize the retained state vector sequences according to the time step to generate multiple groups of fault prediction data at different times.

[0044] In an optional embodiment,

[0045] Determine the fault area to be protected according to the fault location result, establish a local communication network in the fault area, and connect all relay protection devices in the fault area; obtain the real-time operation status of each relay protection device through the local communication network, calculate the protection sensitivity coefficient of each relay protection device, determine the protection range and tripping sequence of each relay protection device according to the protection sensitivity coefficient, and generate a relay protection control instruction; send the relay protection control instruction to each relay protection device through the local communication network, control each relay protection device to perform a tripping operation, and realize the selective isolation of the fault area, including:

[0046] Receive the fault location result, and determine the fault area to be protected according to the fault location result, wherein the fault area includes the line segment where the fault point is located and adjacent equipment electrically connected to the corresponding line segment;

[0047] Extract the relay protection devices in the fault area as communication nodes, measure the physical distance between the communication nodes, detect the communication bandwidth and transmission delay between the nodes, add the physical distance, the inverse of the communication bandwidth and the transmission delay according to the preset weight, calculate the communication link cost, traverse all the node connections in the fault area, sort them from small to large according to the communication link cost, and form a candidate link set;

[0048] Traversing each communication link in the candidate link set, determining whether the nodes at both ends of the current link are in the same connected component, if not, determining the current link as a valid communication link, and merging the connected components where the nodes at both ends are located, repeating the process until all nodes are connected and there is no loop, combining all valid communication links, determining a minimum spanning tree, and establishing a local communication network based on the minimum spanning tree;

[0049] The local communication network is started, and the fault current detected by each relay protection device is collected according to a preset sampling period, and a time window is set according to a preset window length, and the ratio of the fault current to the set current is accumulated and calculated within the time window to obtain the protection sensitivity coefficient of each device;

[0050] Determine the protection sensitivity coefficient of each device, and when the protection sensitivity coefficient exceeds a first preset threshold, expand the protection range of the corresponding device according to a first adjustment coefficient; when it is lower than a second preset threshold, reduce the protection range of the corresponding device according to a second adjustment coefficient; record the adjusted protection range of each device;

[0051] Construct a protection coordination optimization model, set the minimum time interval constraint of protection action, the coordination time difference constraint of upper and lower protection, and the upper and lower limit constraints of action time, and use the particle swarm algorithm to iteratively solve and obtain the optimal tripping sequence that meets the constraints;

[0052] In combination with the protection range of each relay protection device and the optimal tripping sequence, the action time, action sequence and protection setting value of each relay protection device are determined, control strategy parameters are generated, and the control strategy parameters are combined with the corresponding relay protection device identification information to form a relay protection control instruction;

[0053] The relay protection control instruction is encapsulated, and the corresponding information is written into the instruction, and the control instruction is sent to each relay protection device through the local communication network.

[0054] In an optional embodiment,

[0055] The particle swarm algorithm comprises:

[0056] Extracting the electrical connection relationship of the relay protection devices in the fault area, establishing an adjacent level matrix between the devices based on the minimum spanning tree formed by the effective communication link, and initializing each particle in the particle swarm so that each particle contains the tripping time of all the relay protection devices in the fault area;

[0057] Detecting the current number of iterations, calculating the iteration progress ratio, and dynamically updating the magnitude of the inertia coefficient according to a preset inertia coefficient interval, wherein the iteration progress ratio is inversely proportional to the magnitude of the inertia coefficient;

[0058] Calculate the electrical distance between adjacent relay protection devices, convert the electrical distance into a time penalty factor, and add a penalty term to the particles that violate the protection coordination time difference constraint;

[0059] Count the velocity distribution characteristics of each particle in the historical search process, construct a probability density function for velocity update, and generate new velocity components by random sampling based on the probability density function;

[0060] Combined with the adjusted protection range of each relay protection device, the time cost and constraint violation degree of each particle in the current particle swarm are calculated, and the particles are stratified by the non-dominated sorting method to obtain multiple non-dominated layers. The particles in the optimal non-dominated layer are selected to update the individual optimal position and the global optimal position.

[0061] Until the preset number of iterations is reached, the optimal position of the particles is determined and the optimal tripping timing is obtained.

[0062] According to a second aspect of the embodiments of the present invention,

[0063] Provided is a power grid fault location and intelligent relay protection selective control system, including:

[0064] The first unit is used to collect electrical data of multiple measurement units in the power grid, dynamically detect the operation status of the power grid, adjust the sampling window time length according to the operation status of the power grid, calculate the adaptive filtering threshold, filter the electrical data according to the adaptive filtering threshold to obtain filtered data, perform boundary extension on the filtered data to eliminate endpoint effects, perform empirical mode decomposition operation, and decompose to obtain multi-layer fault characteristic component data;

[0065] The second unit is used to input the multi-layer fault characteristic component data into the fault diagnosis model, establish a causal relationship chain between power grid equipment based on the Bayesian network, and calculate the initial probability of the fault; according to the initial probability of the fault, combined with the preset time series evolution rules, generate multiple groups of fault prediction data; calculate the information entropy values ​​of the multiple groups of fault prediction data, and select the position corresponding to the fault prediction data with the minimum information entropy value as the fault location result;

[0066] The third unit is used to determine the fault area that needs to be protected according to the fault location result, establish a local communication network in the fault area, and connect all relay protection devices in the fault area; obtain the real-time operation status of each relay protection device through the local communication network, calculate the protection sensitivity coefficient of each relay protection device, determine the protection range and tripping sequence of each relay protection device according to the protection sensitivity coefficient, and generate a relay protection control instruction; send the relay protection control instruction to each relay protection device through the local communication network, control each relay protection device to perform a tripping operation, and realize selective isolation of the fault area.

[0067] According to a third aspect of the embodiments of the present invention,

[0068] An electronic device is provided, comprising:

[0069] processor;

[0070] a memory for storing processor-executable instructions;

[0071] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0072] A fourth aspect of the embodiments of the present invention is:

[0073] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0074] In an embodiment of the present invention, fault feature information is effectively extracted through signal processing methods such as adaptive filtering, boundary extension and empirical mode decomposition, and the accurate positioning of power grid faults is achieved by combining Bayesian networks and information entropy theory; the real-time status of the relay protection device is obtained based on the local communication network, and the protection range and tripping sequence are dynamically adjusted according to the protection sensitivity coefficient to avoid unnecessary tripping and improve the selectivity and reliability of relay protection; the relay protection control instructions are quickly transmitted through the local communication network to shorten the fault isolation time, reduce the scope and loss of power outages, and ensure the safe and stable operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 A schematic diagram of a flow chart of a method for selectively controlling power grid fault location and intelligent relay protection according to an embodiment of the present invention;

[0076] Figure 2 It is a schematic diagram of the structure of a power grid fault location and intelligent relay protection selective control system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0077] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0078] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0079] Figure 1 FIG. 1 is a flow chart of a method for selectively controlling a power grid fault location and intelligent relay protection according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0080] S101. Collect electrical data of multiple measurement units in the power grid, dynamically detect the operation status of the power grid, adjust the sampling window time length according to the operation status of the power grid, calculate the adaptive filtering threshold, filter the electrical data according to the adaptive filtering threshold to obtain filtered data, perform boundary extension on the filtered data to eliminate endpoint effects, perform empirical mode decomposition operation, and decompose to obtain multi-layer fault characteristic component data;

[0081] In this embodiment, the grid operation status is dynamically detected and the sampling window time length is adjusted to ensure the timeliness and accuracy of the grid data sampling. At the same time, the data is filtered based on the adaptive filtering threshold calculated in real time to effectively remove noise and interference and improve data quality; the boundary extension technology is used to eliminate the endpoint effect in traditional signal processing, avoid errors introduced by improper signal endpoint processing, and ensure the continuity and accuracy of data processing; by performing empirical mode decomposition, the electrical data is decomposed into multiple intrinsic mode functions, from which fault feature component data of different frequency bands are extracted, providing more refined multi-level feature information for subsequent fault analysis.

[0082] S102. Input the multi-layer fault characteristic component data into the fault diagnosis model, establish a causal relationship chain between power grid equipment based on the Bayesian network, and calculate the initial probability of the fault; generate multiple groups of fault prediction data based on the initial probability of the fault and in combination with the preset time series evolution rules; calculate the information entropy values ​​of the multiple groups of fault prediction data, and select the position corresponding to the fault prediction data with the minimum information entropy value as the fault location result;

[0083] In this embodiment, a causal relationship chain between power grid devices is established through a Bayesian network, which can quantify the dependencies and interactions between various devices or components, and provide a scientific reasoning framework for fault diagnosis; by using preset time series evolution rules and combining the initial probability of faults, multiple groups of fault prediction data are generated, which can dynamically predict possible changes in the fault development process and improve the response speed and accuracy to changes in the power grid operation state; by calculating the information entropy values ​​of multiple groups of fault prediction data, the position corresponding to the fault prediction data with the minimum information entropy value is selected as the fault location result, and the data with a lower information entropy value represents that it has less uncertainty and can provide more accurate positioning results, thereby improving the accuracy of fault location; it can quickly and accurately locate the fault source when a fault occurs in the power grid, reduce manual intervention, and improve the automation level and fault recovery efficiency of the system.

[0084] S103. Determine the fault area that needs protection based on the fault location result, establish a local communication network in the fault area, and connect all relay protection devices in the fault area; obtain the real-time operation status of each relay protection device through the local communication network, calculate the protection sensitivity coefficient of each relay protection device, determine the protection range and tripping sequence of each relay protection device according to the protection sensitivity coefficient, and generate a relay protection control instruction; send the relay protection control instruction to each relay protection device through the local communication network, control each relay protection device to perform a tripping operation, and realize selective isolation of the fault area.

[0085] In this embodiment, a local communication network is established in the fault area to connect all relay protection devices to ensure that the status information of each relay protection device can be obtained in real time. The localized communication network helps to improve the speed and reliability of data transmission and reduce the fault response time. According to the real-time operating status of each relay protection device, the protection sensitivity coefficient is calculated, and the protection range and tripping sequence of each device are adjusted accordingly. The dynamic adjustment based on the sensitivity coefficient can ensure that each device is protected at the right time and in the right way to avoid premature or late tripping and reduce the impact on other parts of the power grid. By accurately calculating and generating relay protection control instructions, the instructions are sent to each relay protection device with the support of the local communication network to accurately control the tripping operation, realize selective isolation of the fault area, that is, only cut off the fault area to avoid affecting other normal operating parts of the power grid, and improve the reliability and stability of the power grid. Through precise protection sensitivity adjustment and tripping sequence optimization, the high efficiency and accuracy of fault isolation are achieved, the response speed and processing capacity of the power grid in the event of a fault are significantly improved, and the downtime of the power system and the impact on users are reduced.

[0086] In an optional implementation, electrical data of multiple measurement units in a power grid are collected, the power grid operation state is dynamically detected, the sampling window time length is adjusted according to the power grid operation state, an adaptive filtering threshold is calculated, the electrical data is filtered according to the adaptive filtering threshold to obtain filtered data, the filtered data is subjected to boundary extension to eliminate endpoint effects, and an empirical mode decomposition operation is performed to decompose and obtain multi-layer fault feature component data including:

[0087] The electrical data includes voltage data, current data and phase angle data, which are respectively normalized to obtain reference electrical data;

[0088] Setting an initial sampling window length, calculating the variance value of the reference electrical data within the sampling window, obtaining a voltage variance value, a current variance value, and a phase angle variance value, and multiplying the voltage variance value, the current variance value, and the phase angle variance value by corresponding preset weight coefficients respectively and summing them to obtain a comprehensive state index;

[0089] Detecting the change trend of the comprehensive state indicator over time, calculating the first-order change rate and the second-order change rate of the comprehensive state indicator, substituting the first-order change rate and the second-order change rate into a preset window adjustment function, dynamically adjusting the time length of the sampling window, and obtaining an adaptive sampling window length;

[0090] Extracting the deviation degree of the voltage variance value, the current variance value and the phase angle variance value relative to their respective preset reference states, substituting the deviation degree into a preset exponential decay function, and generating an adaptive threshold value;

[0091] Based on the adaptive threshold, a characteristic mutation interval in the reference electrical data is identified, and based on the characteristic mutation interval, data segments at both ends of the reference electrical data, each accounting for 10% of the total signal length, are intercepted as data segments to be processed; local maximum points and local minimum points are detected in the data segments to be processed to obtain a local extreme point sequence, and the time difference and amplitude difference between adjacent local extreme points in the local extreme point sequence are respectively calculated to obtain a characteristic time series and a characteristic amplitude series;

[0092] Based on the characteristic time series and the characteristic amplitude series, construct a periodic variation function and an amplitude variation function of the extended signal, and combine the periodic variation function and the amplitude variation function to generate an extended processing function;

[0093] At the left end of the reference electrical data, a first group of extended data is generated to the left according to the extended processing function; at the right end of the reference electrical data, a second group of extended data is generated to the right according to the extended processing function; the first group of extended data, the reference electrical data and the second group of extended data are smoothly connected to obtain boundary extended processing data;

[0094] An empirical mode decomposition operation based on cubic spline interpolation is performed on the boundary extension processing data to obtain multiple intrinsic mode components, component characteristic parameters of the multiple intrinsic mode components are calculated, and the multiple intrinsic mode components are screened according to the component characteristic parameters to obtain multi-layer fault characteristic component data.

[0095] Collect electrical data from multiple measurement units in the power grid, such as voltage, current and phase angle data. These data come from different locations in the power grid and can provide comprehensive information on the power grid operation status.

[0096] First, the collected voltage, current and phase angle data are normalized. For example, the minimum-maximum normalization method can be used to scale the data range to between 0 and 1. This step can eliminate the differences between data of different dimensions and facilitate subsequent processing. Assume that the collected voltage data is 10kV, 11kV, 12kV, the current data is 100A, 110A, 120A, and the phase angle data is 30°, 35°, 40°. After normalization, the voltage data becomes 0, 0.5, 1, the current data becomes 0, 0.5, 1, and the phase angle data becomes 0, 0.5, 1. These normalized data are called reference electrical data.

[0097] Set an initial sampling window length, for example, 1 second. Calculate the variance value of the benchmark electrical data within the sampling window to obtain the voltage variance, current variance, and phase angle variance. For example, the voltage variance is 0.2, the current variance is 0.25, and the phase angle variance is 0.1. Multiply these variance values ​​by the corresponding preset weight coefficients, for example, the voltage weight is 0.4, the current weight is 0.3, and the phase angle weight is 0.3. Then add the weighted variance values ​​to obtain the comprehensive state index. For example, the comprehensive state index is 0.4×0.2+0.3×0.25+0.3×0.1=0.185.

[0098] Continuously monitor the changing trend of the comprehensive status indicators over time, and calculate the first-order rate of change (for example, by dividing the difference of the indicators at adjacent moments by the time interval) and the second-order rate of change (for example, by dividing the difference of the first-order rate of change at adjacent moments by the time interval). Substitute the calculated first-order rate of change and second-order rate of change into a preset window adjustment function (for example, a function that adjusts the window length according to the size of the rate of change) to dynamically adjust the length of the sampling window. For example, if both the first-order rate of change and the second-order rate of change are large, indicating that the power grid state changes dramatically, the sampling window length is shortened; conversely, if the rates of change are very small, the sampling window length is extended. Assuming that the calculated first-order rate of change is 0.1 and the second-order rate of change is 0.05, the sampling window length is adjusted to 0.8 seconds according to the window adjustment function.

[0099] Extract the degree of deviation of voltage variance, current variance and phase angle variance relative to their respective preset reference states (e.g., average variance during normal operation). Substitute these deviation degrees into a preset exponential decay function (e.g., a function that increases exponentially as the deviation increases) to generate an adaptive filtering threshold. For example, if the voltage deviation is 0.1, the current deviation is 0.15, and the phase angle deviation is 0.05, according to the exponential decay function, the generated voltage threshold is 0.02, the current threshold is 0.03, and the phase angle threshold is 0.01.

[0100] Based on the adaptive threshold, the characteristic mutation interval in the reference electrical data is identified, such as the interval where the voltage, current or phase angle exceeds the threshold. Based on the characteristic mutation interval, the data segments at both ends of the reference electrical data, which account for 10% of the total signal length, are intercepted as the data segments to be processed. The local maximum point and the local minimum point are detected in the data segment to be processed to obtain the local extreme point sequence. The time difference and amplitude difference between adjacent local extreme points in the local extreme point sequence are calculated respectively to obtain the characteristic time series and characteristic amplitude series.

[0101] Based on the characteristic time series and the characteristic amplitude series, the periodic variation function and the amplitude variation function of the extended signal are constructed. The periodic variation function and the amplitude variation function are combined to generate an extended processing function. At the left end of the reference electrical data, the first group of extended data is generated to the left according to the extended processing function; at the right end of the reference electrical data, the second group of extended data is generated to the right according to the extended processing function. The first group of extended data, the reference electrical data, and the second group of extended data are smoothly connected to obtain the boundary extended processing data.

[0102] Perform an empirical mode decomposition operation based on cubic spline interpolation on the boundary extension processing data to obtain multiple intrinsic mode components. Calculate component characteristic parameters of multiple intrinsic mode components, such as energy entropy, average frequency, etc. Filter multiple intrinsic mode components according to the component characteristic parameters, such as selecting components with larger energy entropy, to obtain multi-layer fault characteristic component data.

[0103] In this embodiment, the sampling window length and the filtering threshold can be adaptively adjusted to effectively filter out noise and interference and extract more accurate fault feature information; the endpoint effect is eliminated through boundary extension processing, the loss of useful information is avoided, and the sensitivity of fault detection is improved; the empirical mode decomposition technology is adopted to quickly decompose multi-layer fault feature component data and improve the efficiency of fault detection.

[0104] In an optional implementation, performing an empirical mode decomposition operation based on cubic spline interpolation on the boundary extension processing data to obtain multiple intrinsic mode components, calculating component characteristic parameters of the multiple intrinsic mode components, and screening the multiple intrinsic mode components according to the component characteristic parameters to obtain multi-layer fault characteristic component data includes:

[0105] Taking the boundary extension processing data as the data to be processed, start envelope screening:

[0106] Searching for all extreme points in the data to be processed, fitting the extreme points using cubic spline interpolation to generate an upper envelope and a lower envelope; calculating the arithmetic mean of the upper envelope and the lower envelope to generate an average line, subtracting the average line from the boundary extension processing data to generate a candidate component;

[0107] Searching for all zero-crossing points and local extreme value points in the candidate components, obtaining the number of zero-crossing points and the number of local extreme value points by counting, and calculating the absolute value of the difference between the number of zero-crossing points and the number of local extreme value points; performing envelope analysis on the candidate components, and calculating the symmetry of the upper envelope and the lower envelope of the candidate components about the time axis;

[0108] When the absolute value is less than or equal to 1 and the symmetry is greater than a preset symmetry threshold, the candidate component is determined to meet the conditions of the intrinsic mode function; when the absolute value is greater than 1 or the symmetry is less than or equal to the preset symmetry threshold, the candidate component is used as new data to be processed, and the envelope screening is repeatedly performed until the conditions of the intrinsic mode function are met, completing a round of envelope screening, and the candidate component is recorded as the first intrinsic mode function;

[0109] Subtracting the first intrinsic mode function from the boundary extension processing data to generate a first residual signal; searching for all local extreme points in the first residual signal and counting the number of the local extreme points; when the number of local extreme points is greater than a preset threshold, taking the first residual signal as new data to be processed, repeatedly performing envelope screening until a new candidate component that meets the conditions of the intrinsic mode function is obtained, and recording the new candidate component as a new intrinsic mode function; subtracting the new intrinsic mode function from the new data to be processed to generate a new residual signal, until the number of local extreme points is less than or equal to the preset threshold, ending the decomposition process, recording the new residual signal as the last intrinsic mode function, and obtaining multiple intrinsic mode functions;

[0110] Calculate the energy entropy values ​​of multiple intrinsic mode functions respectively to generate an energy entropy value sequence; calculate the mutual correlation coefficients of multiple intrinsic mode functions and boundary extension processing data respectively to generate a mutual correlation coefficient sequence; multiply each energy entropy value in the energy entropy value sequence by a first preset weight coefficient, multiply each mutual correlation coefficient in the mutual correlation coefficient sequence by a second preset weight coefficient, sum the multiplied results, and generate a feature importance index corresponding to each intrinsic mode function;

[0111] According to the numerical value of the characteristic importance index, a preset number of intrinsic mode functions are selected in descending order to generate a sequence of intrinsic mode functions to be analyzed; the time domain characteristic parameters, frequency domain characteristic parameters and time-frequency characteristic parameters of each intrinsic mode function in the sequence of intrinsic mode functions to be analyzed are extracted to generate multi-layer fault characteristic component data.

[0112] In an optional specific implementation, before performing envelope screening on the electrical data after boundary extension processing, a symmetric extension method is used to eliminate the data endpoint effect. A data segment with a length of 20% of the original data is extended at both ends of the original data. Taking an actual case as an example, for the original data containing 1000 sampling points, 200 data points are extended at both ends to form boundary extension processing data with a total length of 1400 sampling points.

[0113] When performing envelope screening on the boundary extension processing data, first determine all local maximum points and local minimum points through extreme point search. Use the cubic spline interpolation method to fit these extreme points to generate the upper envelope and lower envelope. For the data of the above 1400 sampling points, usually about 50-80 extreme points can be searched. The values ​​of the upper envelope and the lower envelope at each sampling time are averaged to form an average line. Subtract the average line from the boundary extension processing data to obtain the candidate component.

[0114] Perform zero-crossing and extreme point analysis on the candidate components. Search for all zero-crossing points in the candidate components, and search for all local extreme points at the same time. Taking actual data as an example, a qualified candidate component usually contains 30-50 zero-crossing points and 31-51 extreme points. Calculate the absolute value of the difference between the number of zero-crossing points and the number of extreme points, which should usually be less than or equal to 1. At the same time, calculate the symmetry of the upper and lower envelopes of the candidate component, which is specifically measured by calculating the degree of proximity between the sum of the corresponding point values ​​of the upper envelope and the lower envelope about the time axis and zero. The symmetry range is 0 to 1. When the symmetry is greater than 0.85, it can be considered to have good symmetry.

[0115] When the candidate component meets the two conditions that the absolute value of the difference in the number of zero-crossing extreme points is less than or equal to 1 and the symmetry is greater than 0.85, it is determined to be an intrinsic mode function. If the conditions are not met, the candidate component is used as the new data to be processed and the envelope screening is repeated. Usually, an eigenmode function that meets the conditions can be obtained after 3-5 iterations.

[0116] Subtract the obtained intrinsic mode function from the boundary extension processing data to obtain the residual signal. Count the local extreme points in the residual signal. When the number of extreme points exceeds 10, the residual signal is used as the new data to be processed to continue envelope screening. In this way, multiple intrinsic mode functions can be decomposed in sequence, usually 6-10 intrinsic mode functions can be obtained. When the number of extreme points of the residual signal does not exceed 10, the residual signal is used as the last intrinsic mode function.

[0117] The energy entropy values ​​and the mutual correlation coefficients with the original data of the multiple intrinsic mode functions were calculated respectively. The energy entropy value was multiplied by the weight coefficient 0.6, and the mutual correlation coefficient was multiplied by the weight coefficient 0.4. The multiplication results were summed to obtain the feature importance index. According to the feature importance index, the first four intrinsic mode functions were selected to form the sequence to be analyzed.

[0118] For each component in the intrinsic mode function sequence to be analyzed, the time domain characteristic parameters such as maximum amplitude, root mean square value, zero crossing density, etc. are extracted; the frequency domain characteristic parameters such as main frequency component, bandwidth, energy concentration, etc. are extracted; the time-frequency characteristic parameters such as wavelet energy and singular value decomposition eigenvalue are extracted; these characteristic parameters are combined to form multi-layer fault characteristic component data for subsequent fault characteristic analysis.

[0119] In this embodiment, the intrinsic mode functions in the signal are gradually extracted through boundary extension processing and envelope screening methods, which ensures that the signal is effectively decomposed into multiple components with different frequency characteristics, can accurately reflect the instantaneous frequency changes of the signal, and is particularly suitable for the processing of nonlinear and non-stationary signals; in each round of envelope screening, through statistical analysis of zero-crossing points and local extreme points, and calculation of symmetry, it is ensured that the candidate components meet the conditions of the intrinsic mode function, avoid misjudgment, and improve the accuracy of the decomposition results; by calculating the energy entropy value of each intrinsic mode function and the mutual correlation coefficient with the original signal, combined with the preset weight coefficient, the characteristic importance index of each modal function can be generated, which helps to identify which modal components are more important for signal analysis and fault detection; after selecting important modal functions, extracting multi-level feature parameters such as time domain, frequency domain and time-frequency domain can provide richer and more accurate data support for subsequent fault diagnosis.

[0120] In an optional implementation, the multi-layer fault characteristic component data is input into a fault diagnosis model, a causal relationship chain between power grid equipment is established based on a Bayesian network, and an initial fault probability is calculated; based on the initial fault probability, multiple groups of fault prediction data are generated in combination with a preset time series evolution rule; the information entropy values ​​of the multiple groups of fault prediction data are calculated, and the position corresponding to the fault prediction data with the minimum information entropy value is selected as the fault location result, including:

[0121] Constructing a three-layer Bayesian network, inputting the time domain characteristic parameters in the multi-layer fault characteristic component data into the first layer nodes, inputting the frequency domain characteristic parameters and time-frequency characteristic parameters in the multi-layer fault characteristic component data into the second layer nodes, and setting the power grid equipment nodes as the third layer nodes;

[0122] Obtaining the feature mean of the second-layer node, calculating the Euclidean distance between the time-domain feature parameter and the feature mean, multiplying the Euclidean distance by a preset attenuation coefficient, taking the negative exponent, and then multiplying the Euclidean distance by a preset adjustment coefficient to obtain a first conditional probability between the first-layer node and the second-layer node;

[0123] Extract the feature-fault correspondence in the historical fault case data, establish a second conditional probability matrix between the second-layer nodes and the third-layer nodes, and record the probability that the feature of each second-layer node triggers the fault of each power grid device; calculate the posterior probability of the second-layer node under the preset time domain feature parameters, multiply the posterior probability by the second conditional probability matrix, and obtain the initial probability of failure of each power grid device;

[0124] Based on the physical connection relationship between the power grid devices, determine the fault propagation probability between the power grid devices and establish a time series evolution matrix; set the initial fault probability as an initial state vector, multiply the initial state vector by the time series evolution matrix to obtain a state vector at the next moment, repeat the execution, and generate multiple sets of fault prediction data at different moments;

[0125] Extract the failure probability of each power grid device in each group of fault prediction data, calculate the product of each fault probability and the corresponding natural logarithm, sum and take the negative value, and obtain the information entropy value of each group of fault prediction data; select the fault prediction data corresponding to the smallest information entropy value, determine the position of the power grid device with the largest failure probability in the corresponding fault prediction data, and determine the fault location result.

[0126] In an optional specific implementation, first, a three-layer Bayesian network model is constructed. The first layer nodes of the model correspond to the time domain characteristic parameters collected from the power grid monitoring system, such as the current amplitude change rate, voltage fluctuation amplitude, etc. The second layer nodes correspond to the frequency domain characteristic parameters and time-frequency characteristic parameters, such as harmonic content, wavelet energy, etc., which are also collected from the power grid monitoring system. The third layer nodes represent various devices in the power grid, such as transformers, transmission lines, circuit breakers, etc.

[0127] Next, calculate the conditional probability of the first-layer node to the second-layer node. Specifically, first calculate the distance between each time-domain characteristic parameter and the average value of all characteristic parameters of the same type. For example, calculate the distance between the rate of change of the current amplitude collected at a certain moment and the average value of all current amplitude change rates in the historical data. Then, multiply this distance by a preset attenuation coefficient, take its negative exponent, and finally multiply it by a preset adjustment coefficient to obtain the conditional probability of the first-layer node to the second-layer node. This attenuation coefficient and adjustment coefficient can be adjusted according to actual conditions. For example, the attenuation coefficient can be set to 0.5 and the adjustment coefficient can be set to 1.

[0128] Then, a conditional probability matrix from the second-layer nodes to the third-layer nodes is established. This matrix records the probability of each second-layer node's feature triggering each power grid equipment failure. These probability values ​​can be obtained by analyzing historical failure case data. For example, by analyzing historical data, it can be found that when the harmonic content exceeds a certain threshold, the probability of a transformer failure is 80%, the probability of a transmission line failure is 15%, and the probability of a circuit breaker failure is 5%.

[0129] Next, the initial failure probability of each power grid device is calculated. After obtaining the conditional probability matrix from the first layer to the second layer and the conditional probability matrix from the second layer to the third layer, the posterior probability of the second layer node is calculated in combination with the time domain feature parameters collected at the current moment. Then the posterior probability of the second layer node is multiplied by the conditional probability matrix from the second layer to the third layer to obtain the initial failure probability of each power grid device.

[0130] Based on the physical connection relationship between power grid devices, a time evolution matrix is ​​established, which represents the probability of fault propagation between adjacent devices. For example, if the transformer and the transmission line are directly connected, the probability of the transformer fault propagating to the transmission line may be 90%. The calculated initial probability of failure of the power grid device is used as the initial state vector, which is multiplied by the time evolution matrix to obtain the state vector at the next moment, that is, the failure probability of each device at the next moment. Repeating this step can generate multiple sets of fault prediction data at different moments.

[0131] Finally, the optimal fault prediction data is selected and the fault is located according to the information entropy value. For each set of fault prediction data, its information entropy value is calculated. The information entropy value is calculated by adding the product of the failure probability of each device and its natural logarithm, and then taking the negative value. The fault prediction data with the smallest information entropy value is selected, and the power grid device with the highest failure probability is found in the set of data, and its location is determined as the fault location result. For example, if in the prediction data at a certain moment, the failure probability of the transformer is 95%, and the failure probability of other devices is less than 5%, then the transformer can be judged as a faulty device.

[0132] In this embodiment, comprehensive consideration is given to multi-layer fault feature information and the causal relationship between devices, which can more accurately determine the fault location and avoid misjudgment. Through the Bayesian network and time evolution model, the failure probability of each device at different times can be quickly calculated, so as to quickly locate the fault. Based on a large amount of historical data and power grid topology information, the impact of noise and abnormal data can be effectively reduced, thereby improving the reliability of fault location.

[0133] In an optional implementation, the initial probability of failure is set as an initial state vector, the initial state vector is multiplied by the time series evolution matrix to obtain a state vector at the next moment, and the execution is repeated to generate multiple sets of fault prediction data at different moments; including:

[0134] Collecting the operating condition data of the power grid equipment, extracting the load change cycle, temperature parameters, and humidity parameters from the operating condition data, and setting the evolution time threshold and time step according to the operating condition data;

[0135] Based on the load change cycle, the load change rate is calculated, a load disturbance function is established, and the load change rate is input into the load disturbance function to obtain a load disturbance coefficient; the temperature parameter and the humidity parameter are input into a preset environmental model to calculate the environmental disturbance coefficient;

[0136] Start state evolution calculation, set the initial fault probability as the initial state vector, multiply the initial state vector by the time series evolution matrix to obtain a first state vector; multiply the first state vector by the load disturbance coefficient and the environmental disturbance coefficient respectively to obtain a second state vector; multiply the second state vector by the time series evolution matrix to obtain a third state vector, and execute state evolution calculation repeatedly until the evolution time threshold is reached to generate a state vector sequence;

[0137] Setting the number of Monte Carlo iterations, repeatedly executing the state evolution calculation at each time step, and obtaining multiple groups of state vector sequences;

[0138] Calculate the mean and variance of multiple groups of state vector sequences at each of the time steps, set confidence parameters, calculate the upper limit and lower limit of the confidence interval based on the mean and the variance, and retain the state vector sequences within the confidence interval; organize the retained state vector sequences according to the time step to generate multiple groups of fault prediction data at different times.

[0139] In an optional implementation, first, the operating condition data of the power grid equipment is collected, including load data, temperature data, humidity data, etc. For example, the load data, ambient temperature and humidity data of the transformer can be collected every 15 minutes, and these data are stored in the database.

[0140] Extract the load change cycle, temperature parameters, and humidity parameters from the collected operating condition data. For example, by analyzing the load data of the transformer for one week, it can be determined that its load change cycle is one day, with the peak value occurring at 6 pm and the valley value occurring at 4 am. The temperature parameter can be the average temperature within a day, such as 25 degrees Celsius. The humidity parameter can be the average humidity within a day, such as 70%. Set the evolution time threshold based on the operating condition data and historical fault data. For example, to predict the probability of failure in the next week, the evolution time threshold is set to 7 days. The time step is set to 1 hour.

[0141] Then, based on the extracted load change cycle, the load change rate is calculated. For example, the ratio of the load change per hour to the load of the previous hour can be calculated as the load change rate. A load disturbance function is established, which can map the load change rate to the load disturbance coefficient. For example, a linear function can be used to convert the load change rate into a coefficient between 0.9 and 1.1, which represents the impact of the load change on the probability of equipment failure. The calculated load change rate is input into the load disturbance function to obtain the load disturbance coefficient. For example, if the load change rate in a certain hour is 5%, the corresponding load disturbance coefficient is 1.05. The collected temperature parameters and humidity parameters are input into the preset environmental model to calculate the environmental disturbance coefficient. For example, an environmental model can be established based on historical data to map the temperature and humidity to a coefficient between 0.95 and 1.05, which represents the impact of environmental factors on the probability of equipment failure. If the average temperature of the day is 25 degrees Celsius and the average humidity is 70%, the environmental disturbance coefficient can be calculated to be 1.02 according to the preset environmental model.

[0142] Next, perform state evolution calculations. First, set the initial probability of failure of the device to the initial state vector. For example, based on the historical failure data of the device, the failure probability of the transformer at the initial moment can be set to 0.01. Multiply the initial state vector by the pre-trained time series evolution matrix to obtain the first state vector. The time series evolution matrix describes the change pattern of the equipment failure probability over time. Multiply the first state vector by the load disturbance coefficient and the environmental disturbance coefficient respectively to obtain the second state vector. Multiply the second state vector by the time series evolution matrix to obtain the third state vector. Repeat the state evolution calculation until the set evolution duration threshold is reached to generate a state vector sequence.

[0143] Set the number of Monte Carlo iterations, for example, 1000. Repeat the state evolution calculation at each time step to obtain multiple sets of state vector sequences.

[0144] Calculate the mean and variance of multiple sets of state vector sequences at each time step. Set the confidence parameter, such as 95%. Calculate the upper and lower limits of the confidence interval based on the calculated mean and variance. Keep the state vector sequences within the confidence interval. Organize the retained state vector sequences according to the time step to generate multiple sets of fault prediction data at different times.

[0145] In this embodiment, various factors affecting the probability of equipment failure, such as load changes and environmental factors, are taken into account, and Monte Carlo simulation and confidence interval screening are used to improve the accuracy of fault prediction; by considering multiple factors and using confidence interval screening, the randomness and uncertainty of the prediction results can be effectively reduced, and the reliability of the prediction results is enhanced; by using matrix operations and Monte Carlo simulation, multiple groups of fault prediction data at different times can be efficiently generated, thereby improving the efficiency of fault prediction.

[0146] In an optional implementation, a fault area to be protected is determined according to the fault location result, a local communication network is established in the fault area, and the local communication network connects all relay protection devices in the fault area; the real-time operating status of each relay protection device is obtained through the local communication network, and the protection sensitivity coefficient of each relay protection device is calculated, and the protection range and tripping sequence of each relay protection device are determined according to the protection sensitivity coefficient, and a relay protection control instruction is generated; the relay protection control instruction is sent to each relay protection device through the local communication network, and each relay protection device is controlled to perform a tripping operation, so as to achieve selective isolation of the fault area, including:

[0147] Receive the fault location result, and determine the fault area to be protected according to the fault location result, wherein the fault area includes the line segment where the fault point is located and adjacent equipment electrically connected to the corresponding line segment;

[0148] Extract the relay protection devices in the fault area as communication nodes, measure the physical distance between the communication nodes, detect the communication bandwidth and transmission delay between the nodes, add the physical distance, the inverse of the communication bandwidth and the transmission delay according to the preset weight, calculate the communication link cost, traverse all the node connections in the fault area, sort them from small to large according to the communication link cost, and form a candidate link set;

[0149] Traversing each communication link in the candidate link set, determining whether the nodes at both ends of the current link are in the same connected component, if not, determining the current link as a valid communication link, and merging the connected components where the nodes at both ends are located, repeating the process until all nodes are connected and there is no loop, combining all valid communication links, determining a minimum spanning tree, and establishing a local communication network based on the minimum spanning tree;

[0150] The local communication network is started, and the fault current detected by each relay protection device is collected according to a preset sampling period, and a time window is set according to a preset window length, and the ratio of the fault current to the set current is accumulated and calculated within the time window to obtain the protection sensitivity coefficient of each device;

[0151] Determine the protection sensitivity coefficient of each device, and when the protection sensitivity coefficient exceeds a first preset threshold, expand the protection range of the corresponding device according to a first adjustment coefficient; when it is lower than a second preset threshold, reduce the protection range of the corresponding device according to a second adjustment coefficient; record the adjusted protection range of each device;

[0152] Construct a protection coordination optimization model, set the minimum time interval constraint of protection action, the coordination time difference constraint of upper and lower protection, and the upper and lower limit constraints of action time, and use the particle swarm algorithm to iteratively solve and obtain the optimal tripping sequence that meets the constraints;

[0153] In combination with the protection range of each relay protection device and the optimal tripping sequence, the action time, action sequence and protection setting value of each relay protection device are determined, control strategy parameters are generated, and the control strategy parameters are combined with the corresponding relay protection device identification information to form a relay protection control instruction;

[0154] The relay protection control instruction is encapsulated, and the corresponding information is written into the instruction, and the control instruction is sent to each relay protection device through the local communication network.

[0155] In an optional specific implementation, first, the fault location result provided by the fault location system is received. The result may include information such as the name of the line segment where the fault occurred, the fault type (such as short circuit, grounding, etc.), and the fault distance. For example, the location result shows that a single-phase ground short circuit fault occurred between the 110kV Shengtian Tower No. 23 and the 24th Tower. Based on this location result, the fault area that needs to be protected is determined, including the Shengtian Tower No. 23-24 section where the fault point is located, and the adjacent equipment electrically connected to the line segment, such as the circuit breakers and disconnectors on the 23rd and 24th towers, and the transformer connected to the Shengtian Tower.

[0156] Next, extract all relay protection devices in the fault area as communication nodes. These devices may include line protection, transformer protection, bus protection, etc. For example, there are 5 relay protection devices in the fault area, namely, line protection A on tower 23, line protection B on tower 24, transformer protection C, bus protection D, and bus protection E. Measure the physical distance between each communication node. Assume that the distance between A and B is 2 kilometers, the distance between A and C is 1 kilometer, the distance between B and D is 1.5 kilometers, the distance between C and D is 0.5 kilometers, the distance between D and E is 1 kilometer, and the distance between other nodes is similar. At the same time, detect the communication bandwidth and transmission delay between nodes. For example, the bandwidth between A and B is 100Mbps and the delay is 1ms; the bandwidth between A and C is 50Mbps and the delay is 2ms, and so on. Add the physical distance, the inverse of the communication bandwidth, and the transmission delay according to the preset weights to calculate the communication link cost. For example, if the weights are set to 0.5, 0.3, and 0.2, the link cost between A and B is 0.5×2+0.3×(1 / 100)+0.2×1=1.203. All node connections in the fault area are traversed and sorted from small to large according to the communication link cost to form a candidate link set.

[0157] Then, traverse each communication link in the candidate link set. Determine whether the nodes at both ends of the current link are in the same connected component. If not, determine the current link as a valid communication link, and merge the connected components where the nodes at both ends are located. Repeat until all nodes are connected and there are no loops. For example, first select C and D with the lowest cost, and connect C and D; then select A and C, and add A to the connected component of C and D; then select A and B, and add B; finally select D and E to form a connected component containing all nodes, and there is no loop. Combine all valid communication links, determine the minimum spanning tree, and establish a local communication network based on this minimum spanning tree.

[0158] After the local communication network is established, start the network and begin data collection. Collect the fault current value detected by each relay protection device according to the preset sampling period (for example, 1 millisecond). Set the time window according to the preset window length (for example, 10 milliseconds). Accumulate and calculate the ratio of the fault current to the set current within the time window to obtain the protection sensitivity coefficient of each device. For example, if the fault current detected by device A is 1000A and the set current is 500A, its sensitivity coefficient is (1000 / 500)×10ms=20.

[0159] Next, determine the protection sensitivity coefficient of each device. When the protection sensitivity coefficient exceeds the first preset threshold (e.g., 15), expand the protection range of the corresponding device according to the first adjustment coefficient (e.g., 1.2). When it is lower than the second preset threshold (e.g., 5), reduce the protection range of the corresponding device according to the second adjustment coefficient (e.g., 0.8). Record the adjusted protection range of each device. For example, if the sensitivity coefficient of device A is 20, which exceeds the threshold of 15, its protection range will be expanded to 1.2 times the original value.

[0160] Then, a protection coordination optimization model is constructed. The minimum time interval constraint of the protection action (e.g. 10ms), the coordination time difference constraint of the upper and lower protections (e.g. 30ms), and the upper and lower limit constraints of the action time (e.g. 20ms to 100ms) are set. Through the particle swarm algorithm, the optimal tripping sequence that meets the constraints is obtained through iterative solution. For example, the solution result is: device A trips after 20ms, device B trips after 50ms, and device C trips after 80ms.

[0161] Combined with the protection range and optimal tripping sequence of each relay protection device, the action time, action sequence and protection setting of each relay protection device are determined to generate control strategy parameters. The control strategy parameters are combined with the corresponding relay protection device identification information to form a relay protection control instruction. For example, the instruction "Device A trips after 20ms, and the setting is XXX" is generated.

[0162] Finally, the relay protection control instructions are encapsulated, and the corresponding information is written into the instructions, and the control instructions are sent to each relay protection device through the local communication network.

[0163] In this embodiment, the real-time status of each protection device is quickly acquired through a local communication network, and the protection range and tripping sequence are dynamically adjusted based on the sensitivity coefficient to achieve rapid isolation of the fault. By optimizing the protection coordination, false operation and refusal to operate are avoided, and the reliability of the protection action is improved, thereby ensuring the safe and stable operation of the power system and enhancing the reliability of power supply. By selectively isolating the fault area, the power outage range is limited to the smallest area, reducing the impact on users.

[0164] In an optional implementation, the particle swarm algorithm includes:

[0165] Extracting the electrical connection relationship of the relay protection devices in the fault area, establishing an adjacent level matrix between the devices based on the minimum spanning tree formed by the effective communication link, and initializing each particle in the particle swarm so that each particle contains the tripping time of all the relay protection devices in the fault area;

[0166] Detecting the current number of iterations, calculating the iteration progress ratio, and dynamically updating the magnitude of the inertia coefficient according to a preset inertia coefficient interval, wherein the iteration progress ratio is inversely proportional to the magnitude of the inertia coefficient;

[0167] Calculate the electrical distance between adjacent relay protection devices, convert the electrical distance into a time penalty factor, and add a penalty term to the particles that violate the protection coordination time difference constraint;

[0168] Count the velocity distribution characteristics of each particle in the historical search process, construct a probability density function for velocity update, and generate new velocity components by random sampling based on the probability density function;

[0169] Combined with the adjusted protection range of each relay protection device, the time cost and constraint violation degree of each particle in the current particle swarm are calculated, and the particles are stratified by the non-dominated sorting method to obtain multiple non-dominated layers. The particles in the optimal non-dominated layer are selected to update the individual optimal position and the global optimal position.

[0170] Until the preset number of iterations is reached, the optimal position of the particles is determined and the optimal tripping timing is obtained.

[0171] In an optional specific implementation, first, the electrical connection relationship of the relay protection device in the fault area is extracted. By analyzing the primary wiring diagram of the power system and the protection configuration information, the connection relationship between the various relay protection devices is determined, such as which devices have direct current or voltage loops. Then, based on the minimum spanning tree formed by the effective communication link, an adjacent level matrix between the devices is established. Using the minimum spanning tree algorithm in graph theory, the shortest path connecting all devices is found, and the adjacent level between the devices is determined according to the length of the path, for example, the level of the directly connected device is 1, and the higher the level of the indirectly connected device. For example, assuming that there are four devices A, B, C, and D, A and B are directly connected, B and C are directly connected, and C and D are directly connected, then the adjacent level of A and B is 1, the adjacent level of A and C is 2, and the adjacent level of A and D is 3.

[0172] Next, each particle in the particle swarm is initialized. Each particle represents a set of possible tripping timing schemes, including the tripping times of all relay protection devices in the fault area. For example, assuming there are three devices A, B, and C, a particle can be represented as [0.1s, 0.2s, 0.3s], indicating that device A trips 0.1 seconds after the fault, device B trips at 0.2 seconds, and device C trips at 0.3 seconds.

[0173] In each iteration, first detect the current number of iterations and calculate the iteration progress ratio. For example, if the current number of iterations is 50 and the total number of iterations is 100, the iteration progress ratio is 50%. Then, dynamically update the inertia coefficient according to the preset inertia coefficient interval. The inertia coefficient is used to balance the influence of the particle's own historical speed and the global optimal speed. The iteration progress ratio is inversely proportional to the inertia coefficient. For example, the preset inertia coefficient interval is [0.4, 0.9]. When the iteration progress ratio is 50%, the inertia coefficient can be set to 0.65.

[0174] Then, calculate the electrical distance between adjacent relay protection devices. The electrical distance can be calculated based on parameters such as the line length between the devices and the transformer ratio. For example, if the line length between devices A and B is 10 km, the electrical distance is 10. Convert the electrical distance into a time penalty factor. The larger the electrical distance, the larger the time penalty factor. For example, the electrical distance is 10 and the time penalty factor is 0.05. Add a penalty item to the particles that violate the protection coordination time difference constraint. The protection coordination time difference constraint means that the tripping time difference between adjacent devices should meet a certain range, for example, the tripping time of the subsequent device should be 0.1 to 0.3 seconds later than the tripping time of the previous device. If the tripping timing of a particle violates this constraint, a penalty item is added to it, and the size of the penalty item is related to the time difference and the time penalty factor.

[0175] Next, the velocity distribution characteristics of each particle in the historical search process are counted to construct the probability density function of velocity update. For example, the change of velocity of each particle in the past few iterations can be counted, and the velocity distribution characteristics can be fitted with the normal distribution function. Based on the probability density function, random sampling is used to generate new velocity components. For example, a value can be randomly extracted from the normal distribution function as the new velocity component.

[0176] Combined with the adjusted protection range of each relay protection device, the time cost and constraint violation degree of each particle in the current particle swarm are calculated. The time cost refers to the product of the total load of the power outage area after the fault and the power outage time, the smaller the better. The constraint violation degree refers to the degree of violation of the protection coordination time difference constraint, the smaller the better. The particles are stratified by the non-dominated sorting method to obtain multiple non-dominated layers. The non-dominated layer refers to a set of particles that are not inferior to each other in terms of time cost and constraint violation degree. The particles in the optimal non-dominated layer are selected to update the individual optimal position and the global optimal position.

[0177] Until the preset number of iterations, such as 100, is reached, the optimal position of the particle is determined and the optimal tripping sequence is obtained. For example, the particles finally obtained are [0.15s, 0.25s, 0.35s], then device A trips 0.15 seconds after the fault, device B trips 0.25 seconds, and device C trips 0.35 seconds.

[0178] In this embodiment, by considering the electrical distance and protection coordination time difference constraints, the method can find a tripping sequence that meets the protection coordination requirements, avoid false tripping and refusal to trip, and thus improve the reliability of the power system; by optimizing the tripping sequence, the fault can be isolated more quickly, the power outage scope and time can be reduced, and thus the power supply reliability can be improved; by dynamically updating the inertia coefficient and the speed update strategy based on the probability density function, the method can accelerate the convergence speed of the particle swarm algorithm, thereby improving the computational efficiency.

[0179] Figure 2 FIG. 1 is a schematic diagram of a structure of a power grid fault location and intelligent relay protection selective control system according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0180] The first unit is used to collect electrical data of multiple measurement units in the power grid, dynamically detect the operation status of the power grid, adjust the sampling window time length according to the operation status of the power grid, calculate the adaptive filtering threshold, filter the electrical data according to the adaptive filtering threshold to obtain filtered data, perform boundary extension on the filtered data to eliminate endpoint effects, perform empirical mode decomposition operation, and decompose to obtain multi-layer fault characteristic component data;

[0181] The second unit is used to input the multi-layer fault characteristic component data into the fault diagnosis model, establish a causal relationship chain between power grid equipment based on the Bayesian network, and calculate the initial probability of the fault; according to the initial probability of the fault, combined with the preset time series evolution rules, generate multiple groups of fault prediction data; calculate the information entropy values ​​of the multiple groups of fault prediction data, and select the position corresponding to the fault prediction data with the minimum information entropy value as the fault location result;

[0182] The third unit is used to determine the fault area that needs to be protected according to the fault location result, establish a local communication network in the fault area, and connect all relay protection devices in the fault area; obtain the real-time operation status of each relay protection device through the local communication network, calculate the protection sensitivity coefficient of each relay protection device, determine the protection range and tripping sequence of each relay protection device according to the protection sensitivity coefficient, and generate a relay protection control instruction; send the relay protection control instruction to each relay protection device through the local communication network, control each relay protection device to perform a tripping operation, and realize selective isolation of the fault area.

[0183] According to a third aspect of the embodiments of the present invention,

[0184] An electronic device is provided, comprising:

[0185] processor;

[0186] a memory for storing processor-executable instructions;

[0187] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0188] A fourth aspect of the embodiments of the present invention is:

[0189] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0190] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0191] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements 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.

Claims

1. A method for locating power grid faults and selectively controlling intelligent relay protection, characterized in that: include: Collect electrical data of multiple measurement units in the power grid, dynamically detect the operation status of the power grid, adjust the sampling window time length according to the operation status of the power grid, calculate the adaptive filtering threshold, filter the electrical data according to the adaptive filtering threshold to obtain filtered data, perform boundary extension on the filtered data to eliminate endpoint effects, perform empirical mode decomposition operation, and decompose to obtain multi-layer fault characteristic component data; Inputting the multi-layer fault characteristic component data into a fault diagnosis model, establishing a causal relationship chain between power grid devices based on a Bayesian network, and calculating the initial probability of the fault; According to the initial probability of the fault, combined with the preset time series evolution rules, multiple groups of fault prediction data are generated; Calculating information entropy values ​​of the multiple groups of fault prediction data, and selecting a position corresponding to the fault prediction data with the minimum information entropy value as the fault location result; Determine the fault area that needs to be protected according to the fault location result, establish a local communication network in the fault area, and connect all relay protection devices in the fault area; obtain the real-time operation status of each relay protection device through the local communication network, calculate the protection sensitivity coefficient of each relay protection device, determine the protection range and tripping sequence of each relay protection device according to the protection sensitivity coefficient, and generate a relay protection control instruction; The relay protection control instruction is sent to each relay protection device through the local communication network, and each relay protection device is controlled to perform a tripping operation, thereby realizing selective isolation of the fault area.

2. The method according to claim 1, characterized in that: Collect electrical data of multiple measurement units in the power grid, dynamically detect the operation status of the power grid, adjust the sampling window time length according to the operation status of the power grid, calculate the adaptive filtering threshold, filter the electrical data according to the adaptive filtering threshold to obtain filtered data, perform boundary extension on the filtered data to eliminate endpoint effects, perform empirical mode decomposition operation, and decompose to obtain multi-layer fault feature component data including: The electrical data includes voltage data, current data and phase angle data, which are respectively normalized to obtain reference electrical data; Setting an initial sampling window length, calculating the variance value of the reference electrical data within the sampling window, obtaining a voltage variance value, a current variance value, and a phase angle variance value, and multiplying the voltage variance value, the current variance value, and the phase angle variance value by corresponding preset weight coefficients respectively and summing them to obtain a comprehensive state index; Detecting the change trend of the comprehensive state indicator over time, calculating the first-order change rate and the second-order change rate of the comprehensive state indicator, substituting the first-order change rate and the second-order change rate into a preset window adjustment function, dynamically adjusting the time length of the sampling window, and obtaining an adaptive sampling window length; Extracting the deviation degree of the voltage variance value, the current variance value and the phase angle variance value relative to their respective preset reference states, substituting the deviation degree into a preset exponential decay function, and generating an adaptive threshold value; Based on the adaptive threshold, a characteristic mutation interval in the reference electrical data is identified, and based on the characteristic mutation interval, data segments at both ends of the reference electrical data, each accounting for 10% of the total signal length, are intercepted as data segments to be processed; local maximum points and local minimum points are detected in the data segments to be processed to obtain a local extreme point sequence, and the time difference and amplitude difference between adjacent local extreme points in the local extreme point sequence are respectively calculated to obtain a characteristic time series and a characteristic amplitude series; Based on the characteristic time series and the characteristic amplitude series, construct a periodic variation function and an amplitude variation function of the extended signal, and combine the periodic variation function and the amplitude variation function to generate an extended processing function; At the left end of the reference electrical data, a first group of extended data is generated to the left according to the extended processing function; at the right end of the reference electrical data, a second group of extended data is generated to the right according to the extended processing function; the first group of extended data, the reference electrical data and the second group of extended data are smoothly connected to obtain boundary extended processing data; An empirical mode decomposition operation based on cubic spline interpolation is performed on the boundary extension processing data to obtain multiple intrinsic mode components, component characteristic parameters of the multiple intrinsic mode components are calculated, and the multiple intrinsic mode components are screened according to the component characteristic parameters to obtain multi-layer fault characteristic component data.

3. The method according to claim 2, characterized in that Performing an empirical mode decomposition operation based on cubic spline interpolation on the boundary extension processing data to obtain multiple intrinsic mode components, calculating component characteristic parameters of the multiple intrinsic mode components, and screening the multiple intrinsic mode components according to the component characteristic parameters to obtain multi-layer fault characteristic component data includes: Taking the boundary extension processing data as the data to be processed, start envelope screening: Searching for all extreme points in the data to be processed, fitting the extreme points using cubic spline interpolation to generate an upper envelope and a lower envelope; calculating the arithmetic mean of the upper envelope and the lower envelope to generate an average line, subtracting the average line from the boundary extension processing data to generate a candidate component; Searching for all zero-crossing points and local extreme value points in the candidate components, obtaining the number of zero-crossing points and the number of local extreme value points by counting, and calculating the absolute value of the difference between the number of zero-crossing points and the number of local extreme value points; performing envelope analysis on the candidate components, and calculating the symmetry of the upper envelope and the lower envelope of the candidate components about the time axis; When the absolute value is less than or equal to 1 and the symmetry is greater than a preset symmetry threshold, the candidate component is determined to meet the conditions of the intrinsic mode function; when the absolute value is greater than 1 or the symmetry is less than or equal to the preset symmetry threshold, the candidate component is used as new data to be processed, and the envelope screening is repeatedly performed until the conditions of the intrinsic mode function are met, completing a round of envelope screening, and the candidate component is recorded as the first intrinsic mode function; Subtracting the first intrinsic mode function from the boundary extension processing data to generate a first residual signal; searching for all local extreme points in the first residual signal and counting the number of the local extreme points; when the number of local extreme points is greater than a preset threshold, taking the first residual signal as new data to be processed, repeatedly performing envelope screening until a new candidate component that meets the conditions of the intrinsic mode function is obtained, and recording the new candidate component as a new intrinsic mode function; subtracting the new intrinsic mode function from the new data to be processed to generate a new residual signal, until the number of local extreme points is less than or equal to the preset threshold, ending the decomposition process, recording the new residual signal as the last intrinsic mode function, and obtaining multiple intrinsic mode functions; Calculate the energy entropy values ​​of multiple intrinsic mode functions respectively to generate an energy entropy value sequence; calculate the mutual correlation coefficients of multiple intrinsic mode functions and boundary extension processing data respectively to generate a mutual correlation coefficient sequence; multiply each energy entropy value in the energy entropy value sequence by a first preset weight coefficient, multiply each mutual correlation coefficient in the mutual correlation coefficient sequence by a second preset weight coefficient, sum the multiplied results, and generate a feature importance index corresponding to each intrinsic mode function; According to the numerical value of the characteristic importance index, a preset number of intrinsic mode functions are selected in descending order to generate a sequence of intrinsic mode functions to be analyzed; the time domain characteristic parameters, frequency domain characteristic parameters and time-frequency characteristic parameters of each intrinsic mode function in the sequence of intrinsic mode functions to be analyzed are extracted to generate multi-layer fault characteristic component data.

4. The method according to claim 1, characterized in that: Inputting the multi-layer fault characteristic component data into a fault diagnosis model, establishing a causal relationship chain between power grid devices based on a Bayesian network, and calculating the initial probability of the fault; According to the initial probability of the fault, combined with the preset time series evolution rules, multiple groups of fault prediction data are generated; Calculating the information entropy values ​​of the multiple groups of fault prediction data, and selecting the position corresponding to the fault prediction data with the minimum information entropy value as the fault location result includes: Constructing a three-layer Bayesian network, inputting the time domain characteristic parameters in the multi-layer fault characteristic component data into the first layer nodes, inputting the frequency domain characteristic parameters and time-frequency characteristic parameters in the multi-layer fault characteristic component data into the second layer nodes, and setting the power grid equipment nodes as the third layer nodes; Obtaining the feature mean of the second-layer node, calculating the Euclidean distance between the time-domain feature parameter and the feature mean, multiplying the Euclidean distance by a preset attenuation coefficient, taking the negative exponent, and then multiplying the Euclidean distance by a preset adjustment coefficient to obtain a first conditional probability between the first-layer node and the second-layer node; Extract the feature-fault correspondence in the historical fault case data, establish a second conditional probability matrix between the second-layer nodes and the third-layer nodes, and record the probability that the feature of each second-layer node triggers the fault of each power grid device; calculate the posterior probability of the second-layer node under the preset time domain feature parameters, multiply the posterior probability by the second conditional probability matrix, and obtain the initial probability of failure of each power grid device; Based on the physical connection relationship between the power grid devices, determine the fault propagation probability between the power grid devices and establish a time series evolution matrix; set the initial fault probability as an initial state vector, multiply the initial state vector by the time series evolution matrix to obtain a state vector at the next moment, repeat the execution, and generate multiple sets of fault prediction data at different moments; Extract the failure probability of each power grid device in each group of fault prediction data, calculate the product of each fault probability and the corresponding natural logarithm, sum and take the negative value, and obtain the information entropy value of each group of fault prediction data; select the fault prediction data corresponding to the smallest information entropy value, determine the position of the power grid device with the largest failure probability in the corresponding fault prediction data, and determine the fault location result.

5. The method according to claim 1, characterized in that The initial probability of the fault is set as an initial state vector, the initial state vector is multiplied by the time series evolution matrix to obtain a state vector at the next moment, and the process is repeated to generate multiple groups of fault prediction data at different moments; including: Collecting the operating condition data of the power grid equipment, extracting the load change cycle, temperature parameters, and humidity parameters from the operating condition data, and setting the evolution time threshold and time step according to the operating condition data; Based on the load change cycle, the load change rate is calculated, a load disturbance function is established, and the load change rate is input into the load disturbance function to obtain a load disturbance coefficient; the temperature parameter and the humidity parameter are input into a preset environmental model to calculate the environmental disturbance coefficient; Start state evolution calculation, set the initial fault probability as the initial state vector, multiply the initial state vector by the time series evolution matrix to obtain a first state vector; multiply the first state vector by the load disturbance coefficient and the environmental disturbance coefficient respectively to obtain a second state vector; multiply the second state vector by the time series evolution matrix to obtain a third state vector, and execute state evolution calculation repeatedly until the evolution time threshold is reached to generate a state vector sequence; Setting the number of Monte Carlo iterations, repeatedly executing the state evolution calculation at each time step, and obtaining multiple groups of state vector sequences; Calculate the mean and variance of multiple groups of state vector sequences at each of the time steps, set confidence parameters, calculate the upper limit and lower limit of the confidence interval based on the mean and the variance, and retain the state vector sequences within the confidence interval; organize the retained state vector sequences according to the time step to generate multiple groups of fault prediction data at different times.

6. The method according to claim 1, characterized in that Determine the fault area that needs to be protected according to the fault location result, establish a local communication network in the fault area, and connect all relay protection devices in the fault area; obtain the real-time operation status of each relay protection device through the local communication network, calculate the protection sensitivity coefficient of each relay protection device, determine the protection range and tripping sequence of each relay protection device according to the protection sensitivity coefficient, and generate a relay protection control instruction; Sending the relay protection control instruction to each relay protection device through the local communication network, controlling each relay protection device to perform a tripping operation, and realizing selective isolation of the fault area includes: Receive the fault location result, and determine the fault area to be protected according to the fault location result, wherein the fault area includes the line segment where the fault point is located and adjacent equipment electrically connected to the corresponding line segment; Extract the relay protection devices in the fault area as communication nodes, measure the physical distance between the communication nodes, detect the communication bandwidth and transmission delay between the nodes, add the physical distance, the inverse of the communication bandwidth and the transmission delay according to the preset weight, calculate the communication link cost, traverse all the node connections in the fault area, sort them from small to large according to the communication link cost, and form a candidate link set; Traversing each communication link in the candidate link set, determining whether the nodes at both ends of the current link are in the same connected component, if not, determining the current link as a valid communication link, and merging the connected components where the nodes at both ends are located, repeating the process until all nodes are connected and there is no loop, combining all valid communication links, determining a minimum spanning tree, and establishing a local communication network based on the minimum spanning tree; The local communication network is started, and the fault current detected by each relay protection device is collected according to a preset sampling period, and a time window is set according to a preset window length, and the ratio of the fault current to the set current is accumulated and calculated within the time window to obtain the protection sensitivity coefficient of each device; Determine the protection sensitivity coefficient of each device, and when the protection sensitivity coefficient exceeds a first preset threshold, expand the protection range of the corresponding device according to a first adjustment coefficient; when it is lower than a second preset threshold, reduce the protection range of the corresponding device according to a second adjustment coefficient; record the adjusted protection range of each device; Construct a protection coordination optimization model, set the minimum time interval constraint of protection action, the coordination time difference constraint of upper and lower protection, and the upper and lower limit constraints of action time, and use the particle swarm algorithm to iteratively solve and obtain the optimal tripping sequence that meets the constraints; In combination with the protection range of each relay protection device and the optimal tripping sequence, the action time, action sequence and protection setting value of each relay protection device are determined, control strategy parameters are generated, and the control strategy parameters are combined with the corresponding relay protection device identification information to form a relay protection control instruction; The relay protection control instruction is encapsulated, and the corresponding information is written into the instruction, and the control instruction is sent to each relay protection device through the local communication network.

7. The method according to claim 6, characterized in that The particle swarm algorithm comprises: Extracting the electrical connection relationship of the relay protection devices in the fault area, establishing an adjacent level matrix between the devices based on the minimum spanning tree formed by the effective communication link, and initializing each particle in the particle swarm so that each particle contains the tripping time of all the relay protection devices in the fault area; Detecting the current number of iterations, calculating the iteration progress ratio, and dynamically updating the magnitude of the inertia coefficient according to a preset inertia coefficient interval, wherein the iteration progress ratio is inversely proportional to the magnitude of the inertia coefficient; Calculate the electrical distance between adjacent relay protection devices, convert the electrical distance into a time penalty factor, and add a penalty term to particles that violate the protection coordination time difference constraint; Count the velocity distribution characteristics of each particle in the historical search process, construct a probability density function for velocity update, and generate new velocity components by random sampling based on the probability density function; Combined with the adjusted protection range of each relay protection device, the time cost and constraint violation degree of each particle in the current particle swarm are calculated, and the particles are stratified by the non-dominated sorting method to obtain multiple non-dominated layers. The particles in the optimal non-dominated layer are selected to update the individual optimal position and the global optimal position. Until the preset number of iterations is reached, the optimal position of the particles is determined and the optimal tripping timing is obtained.

8. A power grid fault location and intelligent relay protection selective control system, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used to collect electrical data of multiple measurement units in the power grid, dynamically detect the operation status of the power grid, adjust the sampling window time length according to the operation status of the power grid, calculate the adaptive filtering threshold, filter the electrical data according to the adaptive filtering threshold to obtain filtered data, perform boundary extension on the filtered data to eliminate endpoint effects, perform empirical mode decomposition operation, and decompose to obtain multi-layer fault characteristic component data; The second unit is used to input the multi-layer fault characteristic component data into a fault diagnosis model, establish a causal relationship chain between power grid equipment based on a Bayesian network, and calculate the initial probability of the fault; According to the initial probability of the fault, combined with the preset time series evolution rules, multiple groups of fault prediction data are generated; Calculating information entropy values ​​of the multiple groups of fault prediction data, and selecting a position corresponding to the fault prediction data with the minimum information entropy value as the fault location result; The third unit is used to determine the fault area to be protected according to the fault location result, establish a local communication network in the fault area, and connect all relay protection devices in the fault area; obtain the real-time operation status of each relay protection device through the local communication network, calculate the protection sensitivity coefficient of each relay protection device, determine the protection range and tripping sequence of each relay protection device according to the protection sensitivity coefficient, and generate a relay protection control instruction; The relay protection control instruction is sent to each relay protection device through the local communication network, and each relay protection device is controlled to perform a tripping operation, thereby realizing selective isolation of the fault area.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images