Intelligent power distribution system alarm suppression method based on event correlation analysis

Abstract

The invention relates to the technical field of industrial process control and monitoring, in particular to an intelligent power distribution system alarm suppression method based on event correlation analysis, and the method comprises the steps: obtaining transient recording data related to alarm nodes in a power distribution network, and calculating fault excitation energy; theoretical action time difference is dynamically determined based on fault excitation energy, and space-time manifold impedance between an upstream node and a downstream node is determined in combination with the electrical distance and the actual action time difference; accumulating the space-time manifold impedance according to the topological path, and calculating the effective causal flux received by the downstream node; and determining a causal game coefficient based on the game relationship between the fault excitation energy of the downstream node and the effective causal flux, and correcting the root cause confidence to determine whether to suppress the alarm. According to the method, by introducing the inverse time limit characteristic and the dynamic impedance model, the accuracy of alarm traceability is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial process control and monitoring. Specifically, it relates to an intelligent power distribution system alarm suppression method based on event correlation analysis. BACKGROUND

[0002] Power distribution network alarm tracing and suppression technology is a key link in power system fault handling, mainly used for quickly locating the root cause node from massive alarm information when a fault occurs in the power grid, and shielding the derivative alarms caused by the root cause. Its typical work flow includes: collecting the displacement information and protection action signals of each switch node through the SCADA (Supervisory Control and Data Acquisition) system, analyzing the logical correlation between each alarm event using topology analysis or CGI (Causal Graph Inference) algorithm, and finally outputting the fault point positioning result. This technology is widely used in power supply monitoring systems of urban power distribution networks and industrial parks.

[0003] However, the mainstream CGI (Causal Graph Inference) algorithm in the prior art usually uses a fixed time window to determine the sequence of event occurrence when performing event correlation analysis, and calculates the causal probability based on static topological connections. This method ignores the inherent inverse time limit characteristic of relay protection devices in power distribution networks, i.e., the greater the fault current, the faster the protection action, which leads to the failure of the fixed time determination logic when strong and weak faults coexist. At the same time, the existing method ignores the attenuation effect of electrical impedance on fault waveform propagation, which easily causes the wrong correlation of isolated events that are far apart in electrical distance but close in time.

[0004] The above problems can cause the alarm analysis system to break the causal correlation chain due to the chaotic event time sequence logic when facing high-energy fault impact, making it impossible to correctly distinguish between root cause alarms and companion alarms affected by the proximity effect, and thus forming an alarm storm in the monitoring background. This not only increases the difficulty of troubleshooting for operation and maintenance personnel, but also may cause false judgments of fault points, leading to false power outages in non-fault areas, seriously affecting power supply reliability and repair efficiency.

[0005] Therefore, there is an urgent need for a method that can combine fault physical energy and dynamic impedance characteristics for deep event correlation analysis to solve the misjudgment problem caused by ignoring physical constraints in the prior art. SUMMARY

[0006] To solve the problem of alarm flooding and misjudgment caused by the inability to accurately distinguish between root cause faults and companion actions due to the neglect of relay protection inverse time limit characteristics and electrical impedance constraints in the prior art, the present application proposes an intelligent power distribution system alarm suppression method based on event correlation analysis, which includes: Obtaining transient recording data of each node involved in the alarm in the power distribution network, and determining the fault excitation energy of each node according to the difference between the actual current and the setting protection current of the transient recording data and the voltage change rate. For the upstream node and the downstream node existing topology connection relationship in the power distribution network, the electrical distance and the actual action time difference of the upstream node and the downstream node are obtained, the theoretical action time difference is dynamically determined based on the fault excitation energy of the upstream node, and the space-time manifold impedance of the upstream node and the downstream node is determined according to the deviation of the actual action time difference and the theoretical action time difference and the electrical distance; The space-time manifold impedance is accumulated according to the topology path between the upstream node and the downstream node, and the effective causal flux received by the downstream node from the upstream node is calculated in combination with the space-time manifold impedance of the upstream node and the downstream node and the fault excitation energy of the upstream node. Based on the difference between the fault excitation energy of the downstream node and the effective causal flux received by the downstream node from the upstream node, the causal game coefficient is determined, the initial probability of the downstream node as a root cause is corrected by using the causal game coefficient, the root cause confidence of the downstream node is obtained, and the alarm suppression is performed in response to the comparison result of the root cause confidence and the preset threshold.

[0007] This technical scheme breaks through the limitation of traditional causal analysis which only depends on discrete state and static topology, extracts a quantitative index representing the physical impact strength of the fault from the microcosmic transient waveform by introducing the fault excitation energy, and provides a physical benchmark of energy dimension for causal inference; further, a space-time manifold impedance model is constructed, which skillfully uses the inverse time limit protection characteristics to map the fault energy into the theoretical action time sequence, and aligns and verifies with the actual space-time characteristics, so as to eliminate false correlations that violate physical laws; on this basis, the effective causal flux is used to simulate the physical propagation process of the fault influence in the topology network, and a competitive decision mechanism of endogenous energy and external flux is constructed by using the causal game coefficient, so as to accurately identify the associated nodes forced to act by the upstream energy impact, thereby realizing accurate suppression of associated alarms in complex fault conditions and improving the accuracy of power distribution network fault diagnosis.

[0008] Preferably, the transient recording data of each node is determined based on the following manner: the actual current and the actual voltage of each node are collected in real time and stored in a circular buffer; when it is monitored that the actual electric quantity flowing through a certain node exceeds a pre-set starting threshold or a trip signal of a protection device is received, a recording interception instruction is triggered; a time window composed of a first pre-set time length before triggering and a second pre-set time length after triggering is locked and extracted from the circular buffer, and a sequence composed of the current value and the voltage value at each time in the time window is determined as the transient recording data of the node.

[0009] The preferred scheme ensures that the acquired data not only contains information after the fault occurs, but also retains the system state at the moment before the fault occurs and the transient characteristics during the fault development process, avoids the energy calculation error caused by data loss, and provides a complete and high-fidelity data basis for subsequent feature extraction.

[0010] Preferably, the fault-induced energy of each node is determined based on the following relationship: ; wherein, is the fault-induced energy of node , is the maximum actual current in the transient recording data of node , is the set protection current of node , is the logarithmic function, is the actual voltage of node , is the rated reference voltage of the node, is the voltage change rate of node , to are the start time and end time corresponding to the transient recording data, used to define the start and end time of voltage distortion integration, is a constant to prevent the denominator from being zero, is the absolute value symbol, is the maximum value function.

[0011] The preferred scheme aims to solve the defect of traditional causal inference method which discretizes continuous electrical quantities and loses fault intensity information. By integrating the overload multiple of current relative to the setting value and the integral of voltage transient change rate, not only the driving moment of fault current is reflected, but also the distortion degree of electromagnetic transient is captured, which can accurately distinguish high-energy short-circuit fault and low-energy load fluctuation, and avoid the mistake of equating events of different nature. It is a dynamic deduction of protection behavior based on the real severity of the fault.

[0012] Preferably, the spacetime manifold impedance of the upstream node and the downstream node is determined based on the following relationship: ; wherein, node is the upstream node, node is the downstream node, is the spacetime manifold impedance of node and node , and are the preset spatial normalization coefficient and time normalization coefficient, respectively, is the hyperbolic tangent function, is a natural constant, is the electrical distance between node and node , is the actual action time difference between node and node , is the fault initiation energy of node , and are fitting constants determined based on the inverse time characteristics of the relay protection, is the absolute value symbol, is a preset positive number to prevent the denominator from being 0.

[0013] The first term of this preferred scheme describes the attenuation effect of the spatial distance on the causal correlation using an exponential function, and the second term describes the correlation degree in the time dimension using a hyperbolic tangent function, in particular, a theoretical time term based on the inverse time characteristics is introduced, so that the model can dynamically adjust the expected action time according to the fault energy. This design enables the algorithm to adapt to different intensities of faults, and only when the actual time difference meets the physical protection characteristics, the impedance is the smallest, and the causal correlation is the strongest.

[0014] Preferably, the effective causal flux received by the downstream node from the upstream node is determined based on the following relationship: ; wherein, is the effective causal flux received by node from node , is the fault initiation energy of node , is the topological path between node and node , is the serial number of the node passed through on the topological path, is the spacetime manifold impedance of the th node and the th node passed through on the topological path, is a natural exponential function, is a preset attenuation coefficient, is the spacetime manifold impedance of node and node .

[0015] This preferred scheme simulates the flux transmission process in physical field theory, which not only considers the influence of directly adjacent nodes, but also considers the attenuation of cross-node propagation through path impedance accumulation, which enables the causal flux received by the downstream node to truly reflect the influence size of the upstream fault on it, providing an accurate basis for subsequent differentiation of its own energy and conducted energy.

[0016] Preferably, the causality game coefficient is determined based on the following manner: The maximum value in the effective causality flux of all upstream nodes received by the downstream node is calculated, the difference between the maximum value and the fault excitation energy of the downstream node is calculated, and the difference is negatively correlated mapped into the interval of 0 to 1 by using a nonlinear decay mapping relationship constructed by a Sigmoid function to obtain the causality game coefficient of the downstream node.

[0017] This preferred scheme compares the received external flux with the self-excited energy to determine whether the current node's action is mainly driven by the outside or driven by the self-fault, and the introduction of the Sigmoid function smoothes the difference between the two physical quantities into a coefficient in the probability space, effectively converting the physical layer to the logical layer and enhancing the robustness of the decision.

[0018] Preferably, the root cause confidence of the downstream node is determined based on the following manner: the initial probability of the downstream node as a root cause is calculated by using a causality graph inference algorithm, the initial probability is multiplied by the causality game coefficient of the downstream node, and the operation result is taken as the root cause confidence of the downstream node to realize the correction of the initial probability.

[0019] This preferred scheme constructs a fusion decision mechanism of statistical prior guidance and physical empirical veto. The initial probability provided by the graph inference only reflects the statistical possibility based on the network topology structure and the historical event association, and reflects how large the fault possibility is. The causality game coefficient is based on the physical nature of energy transmission and confrontation, and answers whether it is a spontaneous root cause fault or a passive accompanying fault by comparing the endogenous energy and the external flux. The two are coupled by multiplication operation. For those nodes that are suspected to be root causes in topological position or time sequence logic, but are obviously suppressed and driven by the huge causality flux input from the upstream in physics, their confidence will be suppressed by the lower game coefficient, so that the logical bias caused by the lack of physical constraints in the pure data-driven algorithm is completely eliminated in mathematics, and it is ensured that the finally locked root cause fault meets both the causality logic and the physical electrical conduction characteristics.

[0020] Preferably, the method for determining whether the alarm of the downstream node needs to be suppressed is: if the root cause confidence is lower than a preset root cause confidence threshold, the alarm of the downstream node is determined as an accompanying alarm and is suppressed, and if the root cause confidence is not lower than the preset root cause confidence threshold, the alarm of the downstream node is determined as a root cause fault and is prewarned.

[0021] Preferably, the fitting constant determined based on the inverse time characteristic of the relay protection is determined based on the following manner: reading the protection type configuration information of the relay protection device configured by the upstream node, determining the standard inverse time characteristic curve followed by the device, obtaining the curve coefficient, time multiplier and inverse time index of the curve equation corresponding to the standard inverse time characteristic curve, and determining the product of the curve coefficient and the time multiplier as the fitting constant , the inverse time index is taken as the fitting constant , so that the theoretical action time difference conforms to the physical action characteristic of the relay protection device.

[0022] Preferably, another determination manner of the space normalization coefficient and the time normalization coefficient is as follows: the average electrical distance between all adjacent nodes in the power distribution network is counted, and the reciprocal of the average electrical distance is taken as the space normalization coefficient, so that the change rate of the space impedance with the distance is adapted to the average span of the power grid; the sampling frequency of the transient recording data and the allowed inherent action error time of the relay protection device are obtained, the reciprocal of the sampling frequency is calculated to obtain the sampling period, and the sum of the allowed inherent action error and the sampling period is determined as the time tolerance interval, and half of the time tolerance interval is taken as the time normalization coefficient.

[0023] The present application has the following effects: The present application constructs an alarm analysis architecture of fusion of physical driving and data driving, calculates the fault excitation energy by using the transient recording data, proposes a space-time manifold impedance based on the inverse time characteristic, effectively eliminates the false cause-effect correlation which does not conform to the physical protection logic, constructs a cause-effect flux propagation and energy game mechanism, and improves the accuracy of identification and suppression of alarms in the intelligent power distribution system. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 is a flowchart of the present application; Figure 2 is a comparison diagram of the fault excitation energy and the effective cause-effect flux of two nodes in the present application; Figure 3 is a cause-effect game coefficient diagram of two nodes in the present application; Figure 4 is a comparison diagram of the alarm suppression judgment result of two nodes based on the root cause confidence and the alarm suppression judgment result of the prior art in the present application. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application.

[0026] Referring to Figure 1 , an intelligent power distribution system alarm suppression method based on event correlation analysis, specifically comprising the following steps: S1: Obtain transient recording data of each node involved in the alarm in the power distribution network, and calculate the fault excitation energy of the node based on the sudden change of transient electrical quantity.

[0027] Before performing causal analysis, the physical strength of the fault must be quantified first. Traditional alarm analysis often only relies on the opening / closing state of the switch, losing the changes in the electrical characteristics of the node, such as the key information of current change and voltage fluctuation, which directly determines the speed of the relay protection device action (inverse time limit characteristic) and the size of the fault influence range.

[0028] Therefore, by analyzing the high-frequency recording data, the overload degree of the current and the distortion degree of the voltage are fused into a unified physical scalar, i.e., the fault excitation energy. The node of the real fault source has huge energy and is spontaneous, while the accompanying fault is affected by the energy and conforms to the law of space-time propagation. This step aims to provide a unique energy input source for subsequent dynamic time benchmarking.

[0029] Embodiment 1: First, since the power distribution network is composed of multiple devices connected by physical topology, the physical topology structure diagram of the power distribution network is obtained, and each device is a node in the topology structure diagram, which cooperates with its upstream and downstream nodes to realize the transmission function of the current.

[0030] Then, when the power grid fails, the transient recording data of each node involved in the alarm is obtained from the massive alarm information. Specifically, the transient recording data of each node is determined based on the following method: The actual current and actual voltage of each node are collected in real time and stored in a circular buffer. When it is monitored that the actual current flowing through a certain node exceeds the starting threshold (1.2 times the rated reference current of the node) or the trip signal of the protection device is received, the recording interception instruction is triggered. The time window composed of 1 minute before the trigger (first preset time length) and 1 minute after the trigger (second preset time length) is locked and extracted from the circular buffer, and the sequence of the actual current and actual voltage at each time in the time window is determined as the transient recording data of the node.

[0031] Finally, for any node involved in the alarm , its fault excitation energy is determined based on the following relationship:

[0032] wherein, is the fault excitation energy of the node , reflecting the internal strength of the node fault, i.e., the possibility of the node itself as a fault source, is the maximum actual current in the transient recording data of the node , is the setting protection current of the node , which is an important parameter to ensure the safety and reliability of the equipment in the power system, and is the current value set in the relay protection system to ensure timely cutting off the current when a fault occurs, is the actual voltage of the node , is the rated reference voltage of the node, and is introduced to eliminate the dimension effect, so that the voltage distortion term and the current overrun term are on the same dimensionless level for energy fusion, is the voltage rate of change of the node , to is the integral time window of the transient recording wave data, is a very small normal number used to prevent the denominator from being zero, and is usually set to , is the maximum value function.

[0033] wherein, ensures that when the actual current is smaller than the setting current, the numerator is directly forced to zero, and only when it is larger than the setting current, the square is calculated, which is due to the physical nature of the fault excitation energy is the physical quantity representing the destructive impact strength released by the node in the fault state, which has the characteristics of one-way and threshold triggering, and in mathematical modeling, only when the actual monitoring current exceeds the system set protection reference , the overflow component of the overflow component constitutes a fault impact on the power grid; otherwise, when , the system is in normal load or light load operation area, at this time the physical fault excitation energy should be strictly zero, if the square term without constraint is directly used , due to the even symmetry of the quadratic function, the safety margin (negative difference) during normal operation will be incorrectly converted into positive fault energy after the square operation, therefore, the introduction of the maximum value function essentially imposes a physical causal constraint, which forcibly removes all non-fault interval value noise, ensuring that only the real overcurrent event is responded to, thereby ensuring the rigor of the fault tracing model under all operating conditions.

[0034] wherein, part is the current term, which removes the normal load fluctuation by , and uses the normalized square structure to nonlinearly amplify the overrun degree of the actual current of the node , which physically simulates the contribution of Joule heat effect to the fault, and the thermal destructive power of the fault is proportional to the square of the current, which can better reflect the fault severity than simply using linear current value, and is divided by The index is made dimensionless (normalization operation), which solves the problem of direct comparison between different voltage levels and different devices. Part of the voltage term, using the integral of the voltage rate of change and the logarithmic mapping, captures the severity of voltage change, Reflecting the transient distortion of the voltage, using the logarithmic function is a kind of dynamic signal compression processing method, at the moment of fault, There may be a great peak, if directly integrated, a noise peak may cover the overall characteristics. The logarithmic function suppresses extreme outliers, making the energy index smoother and more robust.

[0035] In summary, it successfully combines the two dimensions of steady-state overcurrent (current square) and transient mutation (voltage rate of change). Only when the current is severely over-limited and the voltage is severely distorted, both indicators will quickly rise, making It can fully represent the physical potential of the node as a fault source, The greater the value, the stronger the destructive power of the node to cause a fault, and according to the inverse time principle, the theoretical time of its protection action should be shorter.

[0036] S2: Based on the fault excitation energy correction time reference, determine the spatiotemporal manifold impedance between nodes.

[0037] After obtaining the excitation energy of each node, in order to solve the problem of time sequence discrimination failure under strong and weak faults caused by the use of fixed time window in traditional algorithms, this step aims to establish a dynamic impedance model driven by energy. The resistance of fault propagation comes from the distance in space and the deviation in time. This step uses the theoretical action time of each node's fault excitation energy to dynamically deduce the actual time deviation, and combines the physical properties of the power grid topology to determine the correlation coefficient, thereby repairing the problem of broken causal chain.

[0038] Example 2: For the upstream nodes and downstream nodes in the distribution network with topological connection relationship, obtain their electrical distance and actual action time difference , and determine the spatiotemporal manifold impedance based on the following relationship:

[0039] Among them, is the spatiotemporal manifold impedance of node and node , reflecting the resistance of causal relationship propagation, the smaller the resistance, the stronger the causal relationship propagation, and are the space normalization coefficient and the time normalization coefficient, respectively, is the hyperbolic tangent function, is the natural constant, is the electrical distance between the node and the node , is the fault energizing energy of the node , and are fitting constants determined based on the inverse-time characteristic of the relay protection, is the absolute value symbol, is a preset positive number for preventing the denominator from being 0, and is usually set as a very small value .

[0040] wherein the fitting constants and are obtained in the following manner: they are not only mathematical constants, but also direct mappings of the physical protection characteristics, and the protection type configuration information (such as the IEC standard inverse-time, the pole inverse-time, etc.) of the relay protection device configured by the upstream node is read to determine the standard inverse-time characteristic curve equation followed by the device, which is usually in the form of wherein, is the action time of the protection device, is the curve coefficient, and for the standard inverse-time characteristic, the value is usually 0.14, is the fault current measurement value, is the setting protection current of the node , is the maximum actual current in the transient recording data of the node , is the inverse-time index, and for the standard inverse-time characteristic, the value is usually 0.02, is the time multiplier, and in the embodiment, the is determined as the fitting constant , considering that the fault energizing energy is proportional to the square of the current (I2), in order to make the finally derived inverse-time action characteristic comply with the standard physical law (I2t), the inverse-time index is determined as the fitting constant , so that accurately calculates the time at which the protection device should theoretically act under the fault energy, and complies with the physical action characteristic of the relay protection device.

[0041] wherein, and are obtained in the following manner: The rate of spatial impedance decay with distance is controlled by the average electrical distance between all adjacent nodes in the statistical distribution network, and the reciprocal of the average electrical distance is taken as This design makes the spatial impedance term have significant discrimination when the average distance is reached, avoiding the failure of spatial weight due to dimensional problems, The tolerance of time deviation is defined, the sampling frequency of transient recording data and the allowable inherent action error time of the relay protection device are obtained, the reciprocal of the sampling frequency is calculated to obtain the sampling period, and the sum of the allowable inherent action error and the sampling period is determined as the time tolerance interval. Half of the time tolerance interval is taken as This ensures that the time deviation within the allowable error range will not cause the impedance to increase dramatically.

[0042] Wherein, the first term of the formula is the spatial part, which describes the barrier of the physical space, and adopts an exponential saturation model. In the physical world, if two nodes are infinitely far apart, their correlation should tend to 0 (the upper limit of impedance tends to 1), rather than linearly growing to infinity, The farther the distance, the larger the value, and the impedance saturates rapidly, reflecting the spatial propagation characteristics of fault impact; the second term is the time part, which no longer compares with a fixed constant, but compares with the deviation of the theoretical time term , strictly corresponding to the inverse time equation of relay protection, which means that only the action time that conforms to the physical law will be considered as low impedance and strong causal correlation, The function is used to map to the interval of 0 to 1. When the fault excitation energy of the upstream node is large, the theoretical action time becomes shorter. At this time, if the actual time difference is also small and close to the theoretical value, the numerator tends to 0, the value is very small, the impedance is very low, and the closer to 0, the stronger the causal correlation; on the contrary, if the fault excitation energy is large, but the time difference is large, the impedance increases dramatically, and the closer to 1, the causal correlation is weakened. Divide by is to eliminate the dimensional difference. The unit of the numerator is second. Here, the impedance is a dimensionless relative value, which must be first divided by to become a dimensionless relative value.

[0043] S3: Calculate the effective causal flux received by the node using the spacetime manifold impedance and the fault excitation energy. ​

[0044] After the spatiotemporal manifold impedance between nodes is determined, considering that the transmission of fault energy in the network is not a point-to-point jump, but a process of continuous attenuation along the topological path, in order to avoid the point-to-point impedance ignoring the cumulative effect of other nodes on the path, this step further analyzes the residual energy when the upstream node is conducted to the downstream node, that is, the effective causal flux, which embodies the process of energy overcoming impedance to do work.

[0045] Embodiment 3: Calculate downstream nodes Receive the effective causal flux from the upstream node :

[0046] Wherein, is the effective causal flux received by node to node , which is used to simulate the flow and attenuation of fault energy in the power grid topology, calculate the passive impact on the downstream node, so that the causal relationship becomes a measurable index. The greater the effective causal flux, the stronger the fault impact on the downstream node, and the stronger the causal relationship between upstream and downstream. Conversely, the causal relationship is weaker. is the fault excitation energy of node , is the topological path between node and node , is the serial number of the node passed on the topological path, is the spatiotemporal manifold impedance of the first node and the first node on the topological path, is a natural exponential function, is a preset attenuation coefficient for adjusting the suppression degree of direct impedance to flux transmission. In this embodiment, the value of is set to 0.5. In the physical field propagation model, energy usually decays exponentially with impedance. Setting 0.5 is a compromise based on engineering experience. If is too large, it will cause the flux to decay too fast, so that even if there is a real causal relationship, the flux transmitted to the downstream is very small, which is easy to cause missed judgment; if is too small, the attenuation effect is not obvious, and a weak disturbance at a long distance can also produce a large flux, which is easy to cause misjudgment. The setting of 0.5 ensures that when the direct impedance is small (strong correlation), the flux remains high, and when the impedance increases, the flux decreases rapidly, which is consistent with the actual attenuation characteristics of the fault waveform in the distribution network line, is node spatiotemporal manifold impedance of the path.

[0047] The relationship constitutes a typical physical field transmission model, The denominator part reflects the energy transmission law that the longer the path and the more intermediate nodes, the greater the energy transmission loss, and the numerator provides the initial driving force, and the denominator integrates the local impedance along the propagation path, and the physical meaning is that the fault energy is conducted from to , and the impedance exists every time a line is passed and every time a protection cooperation is passed, and the exponential term is a classic signal attenuation model, which further strengthens the influence of the direct two-point spatiotemporal matching degree. If the impedance between the two nodes is large, the exponential term will quickly approach 0, almost cutting off the causal relationship between the two nodes, further strengthening the influence of the direct two-point spatiotemporal matching degree. Only when the fault excitation energy of the upstream node is strong enough and the cumulative impedance of the path is small enough, the downstream node receives more effective causal flux.

[0048] As shown in Figure 2 , in a typical fault scenario, the root cause node and the companion node have significant differences in the physical layer. The node is the source of the short-circuit fault, and the current mutation and voltage distortion are severe. The calculated fault excitation energy is extremely high, about 8.5, and since it is the source of energy, it almost does not receive causal flux from the outside, approaching 0; on the contrary, the node has a low fault excitation energy of about 3.2, and receives a large effective causal flux from the upstream node , up to 5.8. By introducing the physical index of effective causal flux, the physical fact of the transmission of fault energy from upstream to downstream is captured at the data bottom layer. This self-fault excitation energy is taken as the internal cause, and the causal flux from the outside is taken as the external cause, which provides an accurate physical criterion for distinguishing between self-driven root cause faults and passive conduction companion faults, overcoming the defect that the traditional method cannot distinguish the fault nature only by the switch state.

[0049] S4: Determine the causal game coefficient based on the game between the fault excitation energy and the effective causal flux, and optimize the causal graph inference algorithm to obtain the root cause confidence of the node.

[0050] In the last link of the causal chain, in order to achieve accurate root cause positioning, the system must make physical corrections to the preliminary probability calculated by the traditional algorithm, which can be physically modeled as a game process: which is the dominant factor, the internal cause of the node itself or the external cause conducted from the outside world.

[0051] This step calls the fault excitation energy of the node itself and the received flux of the node received from the nodes located upstream of the node for confrontation calculation, and uses the game result to nonlinearly modulate the probability parameter.

[0052] Embodiment 4: First, get all the upstream nodes of each node, get the effective causal flux of the node received from each upstream node of the node according to step S3, and then select a maximum effective causal flux, for example, for node , get the effective causal flux of the node received from each upstream node of the node , and the maximum effective causal flux is ; Then, calculate the difference between and the fault excitation energy of node , and use the nonlinear decay mapping relationship constructed by the Sigmoid function to negatively correlate the difference to the interval [0, 1] to obtain the causal game coefficient of the downstream node, satisfying the following relationship:

[0053] Wherein, is the causal game coefficient of node , the value range is , through the antagonistic relationship between internal and external causes, it is determined whether the alarm at the node is the root cause or the companion, and the comparison of physical quantities is converted into a probability coefficient in the range of [0, 1], realizing the smooth transition from the physical layer to the logical layer. is the fault excitation energy of node , exp() is the natural exponential function, and the whole is a nonlinear decay form constructed by the Sigmoid function, part represents the game difference, if node produces a root cause fault, its fault excitation energy is extremely large, and because it is the source, the received external flux is small, at this time the difference is a very small negative number, the exponential tends to 0, the denominator tends to 1, . If node produces a companion fault, although it also has certain electrical fluctuations ( ), but it receives a huge flux from the upstream root cause node, at this time the difference is a very large positive number, the exponential term is extremely large, the denominator tends to infinity, This physical model suppresses false alarms.

[0054] Reference Figure 3 This further reveals the causal game coefficient This is how the Sigmoid function mechanism is used to generate the value. The horizontal axis in the graph represents the game difference, which is equal to the maximum effective causal flux received minus the energy of its own fault excitation. In this part, the vertical axis represents the coefficients of the causal game generated. For node A, since its own energy is much greater than the external flux, the game difference is negative (approximately -8.4). This point falls in the flat region of the upper left of the Sigmoid curve, making the calculated... A stable approach to 1 indicates that the algorithm determines its action is driven by internal factors and should retain its status as a root cause. For node B, since the external flux it receives is much greater than its own energy, the game difference is positive, approximately 2.6. This point falls in the steep drop-off region to the lower right of the Sigmoid curve, leading to... It decays rapidly and approaches 0. This figure vividly illustrates the core logic of the causal game theory of this invention: by using a nonlinear mapping mechanism, the energy difference at the physical level is transformed into an inhibition coefficient at the logical level. This mechanism is not a simple linear threshold cut, but a dynamic smooth inhibition based on energy confrontation. It can adaptively identify nodes suppressed by external energy, thereby eliminating the accompanying alarm interference from a physical nature before logical inference.

[0055] Finally, using The initial probability of each node being the root cause is calculated by the causal graph inference algorithm. After correction, the root cause confidence score for each node is obtained:

[0056] in, For nodes The root cause confidence level, For nodes As the initial probability of the root cause, For nodes The causal game coefficient. It is a probability inferred based on the power grid topology and historical fault logic, reflecting the node The probability of a failure point. Representing physical determinism, it is a coefficient calculated based on the game between fault energy and causal flux. It reflects the physical nature of fault energy outbreaks and transmission at nodes. The likelihood of it being a point of failure.

[0057] In summary, this is a fusion of data-driven and physics-driven results that can be seen as prior probability (statistical inference based on topology and historical data), and likelihood correction based on physical evidence, the multiplication operation embodies the short board effect, even if the traditional algorithm considers that a certain node is probably the root cause, but if the physical game coefficient is very low (proving that it is a passive impact), it will be lowered. It mathematically implements a physical filter that ensures that the final judgment result must satisfy both the topological logic and the physical energy self-consistency conditions.

[0058] For example, for a pseudo-root cause (companion alarm), in the traditional algorithm, a certain downstream node will be misjudged as a high-probability root cause because it receives a cascading trip signal from the upstream and is located at the end of the topology, but in this scheme, the node receives a huge external causal flux, resulting in a game coefficient , Although it is topologically similar to a fault point, the physical energy proves that it is only a passive node that receives an impact, suppressing false positives. For a true root cause (source fault), its own energy is huge and almost no external flux is received, so tends to 1, resulting in ， The physical layer confirms the statistical result, retaining the original algorithm's judgment of topological association.

[0059] Among them, the initial probability of each node as a root cause calculated by the causal graph inference algorithm usually includes the following steps: Construct a Bayesian network or causal graph: Use the single-line diagram (topology) of the power distribution network to define each switch, transformer, and line segment as a node in the graph. According to the current flow direction, define the directed edge. The upstream element failure will cause the downstream element to lose power or act.

[0060] Set conditional probability table: Set the probability based on the relay protection action logic. For example: If line L1 fails, the probability of switch CB1 tripping is 0.99; if L1 is not faulty, the probability of CB1 mis-tripping is 0.01; Input observation evidence: Real-time alarm information collected by the SCADA system (such as switch X displacement, protection Y action) as evidence input into the network, fixing the state of the corresponding node as occurring; Posterior probability reasoning: Through the Noisy-OR model or belief propagation algorithm, the posterior probability of each node failure is inferred in reverse according to the observed alarms. This calculated posterior probability is the initial probability of each node as a root cause in this scheme.

[0061] S5: determining whether the alarm of the node needs to be suppressed in response to a comparison result of the root cause confidence and a preset threshold.

[0062] The root cause confidence threshold is set to 0.8, which is based on a comprehensive consideration of statistical significance and industrial site fault tolerance. In the power system, false positives (mistaking accompanying errors for root causes) will lead to false maintenance scheduling, and false negatives (missing real root causes) may cause power outage incidents to expand. A confidence level of 0.8 (i.e., 80%) is generally considered a highly reliable boundary in probability theory. Since the invention introduces the causal game coefficient , for real accompanying alarms, since the external flux is much larger than the self-energy, tends to 0 quickly, resulting in which is much less than 0.8; and for real root causes, tends to 1, and the final probability mainly depends on the judgment of the original algorithm. Setting 0.8 leaves about 20% safety margin, which can effectively filter out ambiguous interference terms and ensure that real faults with significant physical characteristics are reliably captured.

[0063] If , the node is determined to be a root cause fault node, and the system automatically generates a repair work order and pushes it to the operation and maintenance personnel; if , the alarm of the node is determined to be an accompanying alarm, and the system performs logical folding or suppression display on it in the monitoring interface to prevent interference.

[0064] Referring to Figure 4 , due to the lack of physical energy dimension constraints, the traditional CGI algorithm only relies on time sequence and static topology for inference. In the scenario of cascading trips triggered by strong faults, the action time sequence of node and node is extremely close, making the algorithm unable to distinguish between them, giving a root cause probability of 0.9 or higher (both exceeding the threshold), thus alarming both nodes, and the operation and maintenance personnel cannot determine the true fault point. After the introduction of the causal game coefficient correction in the invention, the root cause confidence of node is accurately suppressed to below 0.1 (much lower than the threshold), and it is determined by the system to be an accompanying alarm and is shielded, while the confidence of node remains at a high level of 0.9 or higher, and it is accurately locked as a root cause fault, and its alarm is notified in time.

[0065] The invention effectively solves the long-standing alarm false alarm problem in power distribution network fault analysis through the fusion strategy of physical driving and data driving, realizes accurate noise reduction of accompanying interference signals while retaining real fault warnings, and improves the automation level of power grid fault disposal.

[0066] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for suppressing alarms in an intelligent power distribution system based on event correlation analysis, characterized in that, include: Acquire transient waveform data of each node involved in alarm in the distribution network, and determine the fault excitation energy of each node based on the difference between the actual current and the set protection current and the voltage change rate of the transient waveform data. For upstream and downstream nodes with topological connections in a distribution network, the electrical distance and actual action time difference of the upstream and downstream nodes are obtained. The theoretical action time difference is dynamically determined based on the fault excitation energy of the upstream node. Based on the deviation between the actual action time difference and the theoretical action time difference and the electrical distance, the spatiotemporal manifold impedance of the upstream and downstream nodes is determined. The spatiotemporal manifold impedance is accumulated based on the topological path between the upstream and downstream nodes, and the effective causal flux received by the downstream node from the upstream node is calculated by combining the spatiotemporal manifold impedances of the upstream and downstream nodes and the fault excitation energy of the upstream node. Based on the difference between the fault-induced energy of the downstream node and the effective causal flux received by the downstream node from the upstream node, the causal game coefficient is determined. The initial probability of the downstream node as the root cause determined by the causal graph inference algorithm is corrected using the causal game coefficient to obtain the root cause confidence of the downstream node. In response to the comparison between the root cause confidence and the preset threshold, alarm suppression is performed.

2. The alarm suppression method for an intelligent power distribution system according to claim 1, characterized in that, The transient waveform data of each node is determined based on the following method: The actual current and voltage of each node are collected in real time and stored in a circular buffer. When the actual power flowing through a node exceeds the preset start threshold or a trip signal from the protection device is received, a waveform recording interception command is triggered. The time window consisting of the first preset duration before triggering and the second preset duration after triggering is locked and extracted from the circular buffer. The sequence of current and voltage values ​​at each moment within the time window is determined as the transient waveform data of that node.

3. The alarm suppression method for an intelligent power distribution system according to claim 1, characterized in that, The fault excitation energy of each node is determined based on the following relationship: ;in, For nodes The fault triggers energy. For nodes The maximum actual current in the transient waveform data, For nodes The set protection current, It is a logarithmic function. For nodes The actual voltage, This is the rated reference voltage for that node. For nodes The rate of change of voltage, to These are the start and end times corresponding to the transient waveform data, used to define the start and end times of the voltage distortion integral. To prevent constants with a denominator of zero, To determine the absolute value sign, This is the function for finding the maximum value.

4. The alarm suppression method for an intelligent power distribution system according to claim 1, characterized in that, The spatiotemporal manifold impedances of upstream and downstream nodes are determined based on the following relationship: Among them, nodes It is an upstream node, a node It is a downstream node. It is a node and nodes The spacetime manifold impedance, and These are the preset spatial normalization coefficients and time normalization coefficients, respectively. It is the hyperbolic tangent function. It is a natural constant. It is a node and nodes Electrical distance between them It is a node and nodes The actual time difference of action, It is a node The fault triggers energy. and All of these are fitting constants determined based on the inverse-time characteristics of relay protection. To determine the absolute value sign, This is a preset positive number used to prevent the denominator from being 0.

5. The alarm suppression method for an intelligent power distribution system according to claim 1, characterized in that, The effective causal flux received by a downstream node from an upstream node is determined based on the following relationship: ； in, For nodes Received node The effective causal flux, It is a node The fault triggers energy. For nodes and nodes Topological paths between them This represents the sequence number of the nodes traversed along the topological path. For the th traversed on this topological path The node and the first The spatiotemporal manifold impedance of each node, It is a natural exponential function. It is the preset attenuation coefficient. For nodes and nodes The spacetime manifold impedance.

6. The alarm suppression method for an intelligent power distribution system according to claim 1, characterized in that, The causal game coefficients are determined based on the following method: Calculate the maximum effective causal flux received by the downstream node from all upstream nodes, calculate the difference between the maximum value and the fault excitation energy of the downstream node, and use the nonlinear decay mapping relationship constructed by the Sigmoid function to negatively correlate the difference to the interval between 0 and 1 to obtain the causal game coefficient of the downstream node.

7. The alarm suppression method for an intelligent power distribution system according to claim 1, characterized in that, The root cause confidence of downstream nodes is determined based on the following method: The initial probability of a downstream node as a root cause is calculated using a causal graph inference algorithm. The initial probability is then multiplied by the causal game coefficient of the downstream node, and the result is used as the root cause confidence of the downstream node to correct the initial probability.

8. The alarm suppression method for an intelligent power distribution system according to claim 1, characterized in that, The method for determining whether alarms from downstream nodes need to be suppressed is as follows: If the root cause confidence level is lower than the preset root cause confidence level threshold, the alarm of the downstream node is determined to be a co-occurring alarm and suppressed. If the root cause confidence level is not lower than the preset root cause confidence level threshold, the alarm of the downstream node is determined to be a root cause failure and an early warning is issued.

9. The alarm suppression method for an intelligent power distribution system according to claim 4, characterized in that, The fitting constants determined based on the inverse-time characteristics of relay protection are determined in the following way: Read the protection type configuration information of the relay protection device configured in the upstream node, determine the standard inverse time characteristic curve followed by the device, obtain the curve coefficients, time multiplier, and inverse time exponent of the curve equation corresponding to the standard inverse time characteristic curve, and determine the product of the curve coefficients and the time multiplier as the fitting constant. Using the inverse time-limited exponent as the fitting constant This is to ensure that the theoretical operating time difference matches the physical operating characteristics of the relay protection device.

10. The alarm suppression method for an intelligent power distribution system according to claim 4, characterized in that, Another way to determine the spatial normalization coefficient and the time normalization coefficient is as follows: The average electrical distance between all adjacent nodes in the distribution network is statistically analyzed, and the reciprocal of the average electrical distance is taken as the spatial normalization coefficient so that the rate of change of spatial impedance with distance is adapted to the average span of the power grid. The sampling frequency of transient waveform data and the allowable inherent operating error time of the relay protection device are obtained. The reciprocal of the sampling frequency is calculated to obtain the sampling period. The sum of the allowable inherent operating error and the sampling period is determined as the time tolerance interval, and half of the time tolerance interval is taken as the time normalization coefficient.

Images