Intelligent auxiliary monitoring system and method for transformer substation

Abstract

The invention relates to an intelligent auxiliary monitoring system and method for a transformer substation, and relates to the technical field of power system monitoring, and the system comprises the steps: collecting multi-source heterogeneous data, constructing an equipment dynamic probability state superposition vector, analyzing quantitative data association through cognitive resonance to update the vector and calculate risk entropy, introducing a focusing control module, and carrying out the focusing control. The calculation intensity of the analysis module is dynamically adjusted according to the risk entropy, and efficient utilization of resources is realized while the monitoring sensitivity is ensured; and meanwhile, the system constructs a state evolution path, and performs dynamic time warping matching on the state evolution path and a preset restricted zone topology library to realize prospective evaluation of the risk evolution trend. And finally, an early warning module determines whether to trigger early warning based on a dual-mode criterion formed by a risk entropy absolute value and a path matching result, and generates a structured diagnosis report containing a potential fault mode and traceability evidence. According to the invention, early-stage, efficient and prospective intelligent early warning of latent faults can be realized, and the operation and inspection efficiency is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system monitoring, in particular to a substation intelligent auxiliary monitoring system and method. BACKGROUND

[0002] As a key hub of the power grid, the safe and stable operation of a large number of high-value power equipment in the substation is of great importance. The traditional monitoring system mainly relies on the judgment of the limit value of a single physical quantity (for example, the oil temperature exceeds the alarm value, the gas pressure is lower than the locking value) to trigger protection or alarm. The sensitivity of this method is low, and it can only trigger when the fault characteristics are very obvious. For early latent faults with weak correlation between multiple physical quantities but insignificant changes in a single physical quantity, it is often difficult to effectively identify, resulting in a lag in early warning and missing the best early disposal window.

[0003] To improve the sensitivity of early warning, some existing technologies try to introduce more complex analysis algorithms, hoping to improve the diagnostic ability by analyzing multi-source data. However, these advanced algorithms usually consume high computing resources. In the healthy operation state of most of the equipment, continuously and indiscriminately performing high-intensity calculations on all data streams causes serious resource waste; and if the calculation intensity or frequency is reduced to save resources, the monitoring sensitivity of the system will be sacrificed, making the system react slowly when facing risk mutations, and falling into the dilemma of "high power consumption" and "low efficiency".

[0004] In addition, the existing risk assessment methods are mostly static "snapshot" diagnosis, that is, only the risk state of the equipment at the current time is evaluated, lacking the ability of dynamic tracking and trend prediction of the state evolution process. This makes the monitoring system unable to identify a potential high-risk scenario: although the absolute value of the risk indicator at the current time is not high, the state evolution trajectory of the equipment has shown a "dangerous tendency" highly similar to the historical typical fault. This neglect of the "momentum" of risk evolution makes the system lose the foresight and cannot realize the real forward-looking early warning.

[0005] Furthermore, when the traditional system finally issues an alarm, the information is often isolated and vague, such as only indicating that "a certain device has an abnormal indicator". This "black box" alarm information does not provide deep clues or evidence chain of the fault cause, resulting in that the operation and maintenance personnel still need to start from zero for wide-range and inefficient manual investigation after receiving the alarm, making it difficult to quickly and accurately locate the problem source and delaying the best disposal opportunity. SUMMARY

[0006] The purpose of this application is to provide an intelligent auxiliary monitoring system and method for substations, which solves the problems of existing technologies in effectively detecting early systemic risks caused by the coupling of weak changes in multiple physical quantities when processing massive monitoring data of substations, and lacking an effective early warning mechanism for fault modes that have never occurred before.

[0007] Firstly, this application provides an intelligent auxiliary monitoring system for substations, which includes:

[0008] The data acquisition module is used to acquire multi-source heterogeneous data within the substation in real time.

[0009] The state vector modeling module, connected to the data acquisition module, is used to generate and dynamically maintain a state superposition vector for each monitored device. The state superposition vector is a probability distribution describing the device being in multiple possible states at the same time.

[0010] The risk analysis module is connected to the data acquisition module and the state vector modeling module.

[0011] Focus on the control module, which connects to the risk analysis module;

[0012] The early warning release module is connected to the risk analysis module.

[0013] In an optional implementation, the state vector modeling module generates a state superposition vector. The form is:

[0014] ;

[0015] in, For a moment The state superposition vector; To be in a state of perfect health For the first Potential fault or abnormal modes, This represents the total number of potential faults or abnormal modes. For the equipment at all times In state The probability of and satisfying .

[0016] The risk analysis module is used to perform the following operations:

[0017] By configuring a set of cognitive resonance operators, coupling analysis is performed on multi-source heterogeneous data to obtain the resonance intensity that quantifies the synchronicity or causal relationship between different signal change trends.

[0018] Based on the obtained resonance intensity, the state superposition vector maintained by the state vector modeling module is updated. In one optional implementation, the log-probability update method for the state is as follows: ,in, For state At any moment The logarithmic odds; For state At any moment The logarithmic odds; To update the step size coefficient; For the first The first resonance and the second The association strength weight of each fault state; For the first The resonance relationship at time The resonance intensity; The total number of cognitive harmonic operators; summation symbol This indicates that the summation is performed over all resonance relationships.

[0019] Based on the updated state superposition vector, the risk entropy characterizing the uncertainty of the device state is calculated. In an optional implementation, the risk entropy is calculated using the Shannon information entropy formula: ,in, For a moment Risk entropy; For a moment The state superposition vector; For the device in the state superposition vector, it is in the first position. The probability components of a state; This represents the total number of potential faults or abnormal modes.

[0020] The state evolution path formed by the tracking of the state superposition vector changes over time is then matched with a preset restricted area topology library that stores typical high-risk evolution patterns.

[0021] The focus control module is used to generate control signals based on the current value and trend of the risk entropy calculated by the risk analysis module, and send them to the risk analysis module to dynamically adjust the calculation priority and sensitivity of the cognitive resonance operator.

[0022] The early warning release module is used to receive risk entropy and topology matching results from the risk analysis module, and generate early warning information based on a dual-mode criterion.

[0023] The dual-mode criteria include: a risk entropy threshold criterion based on risk entropy and its rate of change, and an evolution path restricted area criterion based on the topology matching results of the state evolution path and the restricted area topology library.

[0024] Secondly, this application provides a method for intelligent auxiliary monitoring of substations, which includes the following steps:

[0025] Real-time acquisition of multi-source heterogeneous data in a substation;

[0026] For each monitored device, a state superposition vector is generated and dynamically maintained, which is a probability distribution describing the device being in multiple possible states simultaneously;

[0027] Based on the multi-source heterogeneous data, multi-source signal coupling analysis is performed by a cognitive resonance operator to obtain resonance strength quantifying the synchronicity or causal correlation between different signal change trends;

[0028] The state superposition vector is updated based on the resonance strength, and a risk entropy representing the uncertainty of the device state is calculated;

[0029] The state evolution path formed by the state superposition vector over time is tracked, and the state evolution path is topologically matched with a preset forbidden zone topology library;

[0030] According to the current value and change trend of the risk entropy, the calculation priority and sensitivity of the cognitive resonance operator are dynamically adjusted;

[0031] Based on the risk entropy and the results of the topological matching, an early warning information is generated and issued.

[0032] In summary, the present application includes at least one of the following beneficial technical effects:

[0033] 1. The present application can integrate the weak signal changes of multiple sensors within the normal range into a quantitative risk indicator representing the overall uncertainty of the system by generating a state superposition vector and using a cognitive resonance operator to analyze the coupling relationship of multi-source heterogeneous data to calculate the risk entropy. This enables the present application to identify systemic early risks represented by the coordinated changes of multiple physical quantities, which cannot be discovered by traditional monitoring methods based on single parameter threshold alarms;

[0034] 2. The present application realizes the monitoring of the state evolution process of the device by tracking the state evolution path formed by the state superposition vector in the multi-dimensional probability space and matching it with the preset forbidden zone topology library. This enables the system to identify specific high-risk evolution patterns, even if the risk entropy itself does not reach the alarm threshold at any instantaneous moment on the evolution path. It provides the ability to identify process risks and predict potential instability trends, surpassing traditional methods that rely solely on instantaneous state for judgment;

[0035] 3.The application sets a focus control module, uses the risk entropy output by the risk analysis module as a feedback control signal to dynamically adjust the calculation priority and sensitivity of the cognitive resonance operator. This closed-loop control mechanism enables the system to run in a low resource consumption mode when the risk entropy is low, and automatically focuses the calculation resources on the signal coupling analysis most related to the risk source when the risk entropy rises, thereby improving the response speed and analysis depth of key risks while optimizing the overall calculation resource utilization efficiency of the system. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 is a system architecture diagram of the application;

[0037] Figure 2 is a method flowchart of the application;

[0038] Figure 3 is a logic diagram of the double-mode criterion of the early warning release module in the embodiment of the application.

[0039] Reference signs: 10, data acquisition module; 20, state vector modeling module; 30, risk analysis module; 40, focus control module; 50, early warning release module. DETAILED DESCRIPTION

[0040] The application will be further described in detail below in combination with the accompanying drawings. Figure 1 The application will be further described in detail below in combination with the accompanying drawings.

[0041] Embodiment 1: An intelligent auxiliary monitoring system for a substation, comprising:

[0042] A data acquisition module 10 is configured to acquire multi-source heterogeneous data in the substation in real time.

[0043] The data acquisition module 10 is the sensing basis of the system in the embodiment of the application, is responsible for acquiring raw data from various monitoring devices in the substation, and performs synchronization preprocessing on the data to provide a time-dimensionally aligned and unified data set for subsequent analysis modules.

[0044] In a specific embodiment, the multi-source heterogeneous data acquired by the data acquisition module 10 includes but is not limited to the following three types:

[0045] Device electrical quantity data, which is from a substation automation system. These data reflect the core operating conditions of the device, and specifically include: three-phase voltage effective value, three-phase current effective value, active power, reactive power, and power grid frequency. These data are usually recorded at a sampling rate of several hertz to represent the macroscopic electrical characteristics of the device.

[0046] Device non-electrical quantity data, which is from various special sensors installed at key parts of the device to monitor physical characteristics caused by electrical or mechanical reasons.

[0047] Environmental data, which are collected from environmental sensors installed around the equipment. These data are mainly used as correction or auxiliary judgment, including: ambient temperature, relative humidity, SF6 gas concentration for monitoring the status of gas insulated switchgear or circuit breaker, and smoke detector readings for fire warning.

[0048] Since the above multi-source heterogeneous data have different sampling rates and reporting periods, one key function of the data acquisition module 10 is to perform strict time synchronization processing. In the present embodiment, this processing is achieved by the following steps: first, all data acquisition front-end units in the system are synchronized with the station master clock server through the precision time protocol, ensuring the unity of the station time reference. Second, each sensor or monitoring device attaches a high-precision time stamp when generating a data record.

[0049] Subsequently, a data processing unit inside the data acquisition module 10 establishes a common time grid with a preset uniform time step (e.g. 1 second). For the collected non-synchronized data stream, this unit aligns it to the common time grid according to the data characteristics using different methods: for slowly changing data (such as ambient temperature or dissolved gas concentration in oil), a forward filling method is used, i.e. maintaining the last valid value within the time step; for faster changing data (such as current or vibration signals), a linear interpolation method is used to calculate the estimated value at the grid time point. Through this process, the data acquisition module 10 finally outputs a time-synchronized multi-dimensional state observation matrix, each row of which corresponds to a certain time point, and each column corresponds to a specific monitored physical quantity.

[0050] The state vector modeling module 20, connected with the data acquisition module, is used to generate and dynamically maintain a state superposition vector for each monitored device, which is a probability distribution describing the device being in multiple possible states simultaneously.

[0051] The state vector modeling module 20 is connected with the data output end of the data acquisition module 10, and its own data output end is connected with the input end of the risk analysis module 30. The core function of this module is to establish and maintain a dynamic, probabilistic state description model for each monitored device object, replacing the traditional, binary "normal / fault" judgment model.

[0052] The model established by the state vector modeling module 20 for the device is called a state superposition vector. This vector is a probability distribution vector in mathematics, and its physical meaning is that the device is not absolutely in a single state at any time, but is in a superposition state of multiple possible states with different probabilities under the influence of uncertainty factors. This modeling method can quantitatively describe the intermediate fuzzy state of the device evolving from a completely healthy state to a failure state.

[0053] The specific mathematical form of the state superposition vector at time t is defined and maintained by the state vector modeling module 20 as follows:

[0054] ;

[0055] Wherein:

[0056] : represents the state superposition vector of the device at time t.

[0057] : represents the evaluated completely healthy state, which is a reference state in which all physical indicators of the device are in the ideal range.

[0058] : represents the pre-defined th potential failure or abnormal mode, wherein the value range of is 1 to .

[0059] represents the total number of pre-defined potential failure or abnormal modes.

[0060] : represents the probability of the device being in state at time , which is a real number between 0 and 1.

[0061] The sum of all probability components of the vector is always 1, i.e. it satisfies the probability normalization condition: .

[0062] In this model, the state set represents all possible states of the system, wherein usually represents the normal operating state, and is the content of the potential failure or abnormal mode to be identified. The content of the potential failure or abnormal mode is stored in a configurable mode library. The establishment of the mode library is based on:

[0063] Failure mode and effects analysis report provided by the manufacturer of the device.

[0064] High frequency failure types derived from statistical analysis of historical operation data, maintenance records and failure reports of the same type of equipment.

[0065] Failure types derived from industry standards or guidelines, such as those related to dissolved gas analysis in transformer oil.

[0066] The state vector modeling module 20 is responsible for the initialization of the state superposition vector. When a new device is put into operation or an existing device completes a major repair and passes all offline tests, the module initializes the state superposition vector corresponding to the device to The initial vector indicates that at the initial time, the probability of the device being in a completely healthy state is 1, and the probability of being in any potential failure mode is 0. After the system is put into operation, the module will continuously transmit the state superposition vector at the current time to the risk analysis module 30 and receive the update instructions calculated by the risk analysis module 30 to dynamically maintain the real-time state of the vector.

[0067] The risk analysis module 30, connected with the data acquisition module and the state vector modeling module, is adapted to perform multi-source signal coupling analysis through cognitive resonance operators to obtain resonance strength, update the state superposition vector based on the resonance strength, calculate the risk entropy, and track the state evolution path formed by the change of the state superposition vector over time, and perform topological matching between the state evolution path and the preset forbidden zone topology library.

[0068] The risk analysis module 30 is the core processing unit of the system in the embodiment of the present application. The data input end of the module is connected with the data acquisition module 10 and the state vector modeling module 20 respectively, to receive the time-synchronized multi-dimensional state observation matrix and the current state superposition vector of the device. The output end of the module is connected with the focus control module 40 and the early warning release module 50 respectively. The risk analysis module 30 internally integrates three major functions of cognitive resonance analysis, state update and risk entropy calculation, and state evolution path analysis.

[0069] The cognitive resonance analysis function is used to analyze the coupling correlation between multi-source heterogeneous data. The analysis does not directly judge whether the amplitude of a single physical quantity is out of limit, but quantifies the synchronization or causal correlation between different time series signals in the change trend through a set of configurable cognitive resonance operators. In a specific embodiment, the module is configured with the following two types of cognitive resonance operators:

[0070] Time-frequency resonance operator, used to analyze the time-varying coupling relationship between non-stationary signals. One specific implementation is to use wavelet coherence analysis. For two signals from the data acquisition module 10 (e.g., pulse count signal of partial discharge and device load current signal), the operator first performs continuous wavelet transform on both signals to obtain their energy distribution and phase information on the time-frequency plane. Then, by calculating the cross-wavelet spectrum and the individual wavelet power spectrum at the same time-frequency point, a wavelet coherence coefficient value between 0 and 1 is obtained. If the value is close to 1, it means that the two signals have stable phase relationship and energy synchronization at that specific time and frequency scale. The resonance strength output by the operator is the time-averaged coherence coefficient calculated on the critical frequency band.

[0071] Causal resonance operator, used to quantify the strength of directional driving or information flow between different signals. One specific implementation is to use transfer entropy. For two time series signals X and Y (e.g., transformer top layer oil temperature signal X and dissolved gas CH4 concentration signal Y), the operator quantifies how much additional information the history of X can provide for predicting the future value of Y, given the history of Y. If the calculated transfer entropy value from X to Y is significantly greater than that from Y to X, it indicates that signal X has a causal driving relationship with signal Y. The resonance strength output by the operator is the net transfer entropy value calculated within a specific time window.

[0072] State update and risk entropy calculation function, used to integrate the results of cognitive resonance analysis into the state assessment of the device. The function first updates the state superposition vector based on the obtained resonance strength . In this embodiment, the update process does not directly modify the probability, but is implemented by updating the log odds of each state, with the specific update formula as follows:

[0073] ;

[0074] where:

[0075] and represent the log odds of state at time and time , respectively.

[0076] : represents a preset update step coefficient, which is a positive real number, used to control the update rate.

[0077] : represents the total number of cognitive resonance operators configured in the system.

[0078] : indicates the first A cognitive resonant operator at time 1 The calculated resonance intensity.

[0079] : Represents a preset correlation strength weight, used to characterize the relationship between the k-th resonance relationship and the th resonance relationship. Failure modes The degree of correlation. This weight matrix is ​​configured based on expert knowledge or historical data statistics.

[0080] After calculating all states After calculating the logarithmic odds at time step 30, the risk analysis module 30 normalizes them using the Softmax function to obtain the updated probability components. Thus, a new state superposition vector is obtained. Subsequently, the module is based on the updated state overlay vector. The risk entropy at the current moment is calculated using the Shannon information entropy formula. This is used to quantify the overall uncertainty of the equipment's condition.

[0081] ;

[0082] The higher the risk entropy value, the more dispersed the probability distribution of the device state, and the higher the uncertainty of the system.

[0083] The state evolution path analysis function is used to evaluate the dynamic process of equipment state changes. The risk analysis module 30 superimposes a vector sequence of states at consecutive time points. This constructs a state evolution path for the device in an (n+1)-dimensional probability space. Simultaneously, this module pre-stores a restricted area topology library, which contains multiple typical high-risk state evolution paths derived from historical failure cases or simulation analysis.

[0084] The risk analysis module 30 performs real-time topology matching between the device's current state evolution path and each path in the restricted area topology library. A specific matching algorithm is Dynamic Time Warping (DTW). The DTW algorithm constructs a distance matrix, finds the optimal nonlinear alignment between two paths, and calculates a cumulative distance as a similarity score between the two paths. If the calculated similarity score is below a preset threshold, it indicates that the device's current state evolution trend is highly similar to a known high-risk evolution pattern.

[0085] Finally, the risk analysis module 30 will calculate the risk entropy. and its rate of change over time Transmitted to the focus control module 40. Simultaneously, the risk entropy... The topological matching results of the state evolution path (e.g., the ID of the most similar restricted path and its similarity score) are transmitted to the early warning release module 50.

[0086] The focus control module 40 is connected to the risk analysis module and is used to dynamically adjust the calculation priority and sensitivity of the cognitive resonance operator based on the current value and changing trend of the risk entropy.

[0087] The data input terminal of the focus control module 40 is connected to the output terminal of the risk analysis module 30, while its control output terminal is connected to the control input terminal of the risk analysis module 30, forming a feedback control loop. The function of this module is to dynamically adjust the calculation behavior of the risk analysis module 30 based on the current risk situation of the equipment, in order to optimize the allocation of computing resources and prioritize responses to key risks.

[0088] The focus control module 40 continuously receives two key indicators from the risk analysis module 30: the risk entropy at the current moment. and the time change rate of risk entropy This module is internally configured with two levels of control thresholds: a low-risk threshold and a low-risk threshold. and a high-risk threshold The focus control module 40 determines the system's operating mode based on the relationship between the received indicators and preset thresholds, and sends corresponding control commands to the risk analysis module 30.

[0089] The working mechanism of this module is specifically divided into the following three modes:

[0090] Baseline monitoring mode: when risk entropy Below the low risk threshold At this point, the system is deemed to be in a healthy and stable state. The focus control module 40 then issues a control command to the risk analysis module 30, instructing it to operate in a low-power baseline mode. In this mode, the risk analysis module 30 activates only a small set of pre-defined, low-computational-commonness cognitive resonance operators (e.g., performing simple time-domain correlation analysis on key electrical quantities) and performs state updates and risk entropy calculations at relatively long time intervals (e.g., once per minute).

[0091] Alertness Analysis Mode: When Risk Entropy Exceeding the low-risk threshold But below the high-risk threshold or the rate of change of risk entropy over time When the risk entropy exceeds a preset positive value, the system is identified to be in an uncertainty growth alert state. At this time, the focus control module 40 issues an instruction to switch the risk analysis module 30 to an alert analysis mode. In this mode, the risk analysis module 30 activates a larger range of cognitive resonance operators of medium complexity and increases the frequency of state updates to a higher level (for example, once every 5 seconds).

[0092] Focused diagnosis mode: when the risk entropy exceeds a high-risk threshold , the system is identified to be in a high-risk state. At this time, the focus control module 40 switches the risk analysis module 30 to a focused diagnosis mode. In this mode, the module first identifies which one or several fault modes in the current state superposition vector have the largest probability component that contributes to the growth of the risk entropy.

[0093] Based on the query results, the focus control module 40 generates and issues a precise control instruction. This instruction not only requires the risk analysis module 30 to activate all relevant cognitive resonance operators that are preset to have high computational complexity (for example, transfer entropy calculation or high-resolution wavelet coherence analysis for specific signal pairs), but also specifies the data channels that these operators should prioritize for analysis. For example, if the probability of the "partial discharge" mode is identified as the main source of entropy growth, the control instruction will specify that the risk analysis module 30 should prioritize high-precision calculation of the time-frequency resonance relationship between the ultra-high frequency signal and the phase of the power frequency voltage. In this way, the system's computing resources are precisely directed to the most critical risk source analysis.

[0094] The early warning release module 50 is connected to the risk analysis module and is used to generate and release early warning information based on the results of risk entropy and topology matching.

[0095] The data input end of the early warning release module 50 is connected to the output end of the risk analysis module 30. This module is used to receive the final calculation results of the risk analysis module 30 and, based on a certain bimodal criterion, to decide whether to generate and what level of early warning information to generate, and finally to push the structured early warning report to the human-machine interface of the monitoring system or the superior dispatching system.

[0096] Referring to the accompanying Figure 3 , in an embodiment of the present application, the input information received by the early warning release module 50 includes the risk entropy , the time rate of change of the risk entropy , and the topology matching results of the state evolution path and the forbidden zone topology library. The bimodal criterion integrated in this module is composed of a risk entropy threshold criterion and an evolution path forbidden zone criterion. When either criterion is met, the system generates an early warning.

[0097] The first criterion is the risk entropy threshold criterion. This criterion sets multiple trigger conditions based on the absolute value of the risk entropy and its rate of change. For example:

[0098] Level 1 warning condition: Current risk entropy The value exceeds a preset level 1 warning threshold. And its rate of change over time The slope remains consistently greater than a preset positive slope threshold. This condition corresponds to the initial stage where the uncertainty of the device status is significantly increasing.

[0099] Level 2 Alarm Conditions: Current Risk Entropy The value directly exceeds a preset secondary alarm threshold, which is higher than the primary warning threshold. This condition indicates that the device status has entered a region of high uncertainty.

[0100] The second criterion is the evolution path no-go criterion. This criterion is independent of the instantaneous risk entropy value and focuses on assessing whether the dynamic process of equipment state evolution is similar to known hazardous patterns.

[0101] The triggering condition is: Topology matching alarm condition: The dynamic time warp distance between the current state evolution path of the device calculated by the risk analysis module 30 and any high-risk path in the restricted area topology library is less than a preset similarity threshold. This condition indicates that even if the absolute value of the current risk entropy is not high, the state evolution trend of the equipment has reproduced a typical historical failure evolution path, suggesting that the subsequent state will tend towards a serious failure.

[0102] When any of the above conditions are met, the early warning issuance module 50 generates a structured diagnostic report. This report is not a simple alarm signal, but rather structured data containing the following specific information fields:

[0103] Basic information includes the affected device ID, the timestamp of the alert, and the alert level determined based on the triggering conditions (e.g., Level 1 alert or Level 2 alarm).

[0104] Triggering criteria: Clearly indicate which criterion (risk entropy threshold or evolution path forbidden zone) triggered this warning.

[0105] State snapshot: Provides a state overlay vector at the moment of early warning. It visually displays the most suspicious potential failure modes in the form of probability distribution.

[0106] Quantitative indicators: providing risk entropy values ​​at the time of early warning. , and (if applicable) the path ID that best matches the restricted area topology library and its DTW similarity score.

[0107] Traceable evidence chain: list the specific signal pairs involved in the most contributive resonant relationship identified by the risk analysis module in the process of risk entropy growth (e.g. "time-frequency resonance of winding vibration signal and load current signal"), providing direct clues for the maintenance personnel to troubleshoot problems.

[0108] Embodiment 2: refer to the attached Figure 2 A substation intelligent auxiliary monitoring method, comprising the following steps:

[0109] Step S1, the data acquisition module 10 continuously acquires multi-source heterogeneous data related to the transformer, and performs time synchronization processing to form a unified multi-dimensional state observation matrix. The data stream collected includes: load current, top oil temperature, partial discharge ultra-high frequency (UHF) signal, box vibration signal, and H2, CH4, C2H2, etc. Characteristic gas concentration reported by the online oil chromatographic analysis system.

[0110] Step S2, the state vector modeling module 20 initializes the state superposition vector of the transformer as , indicating that it is determined to be in a completely healthy state . The vector is transmitted to the risk analysis module 30.

[0111] With the passage of time, the internal insulation of the transformer appears slight deterioration, and early physical signs of inter-turn short circuit failure begin to appear. These signs are very weak, and the change of any single physical quantity does not reach the traditional alarm threshold. At this time, the method process of the present application enters the core analysis stage:

[0112] Step S3, the risk analysis module 30 receives the new state observation matrix and performs cognitive resonance analysis. In this stage, a time-frequency resonance operator (wavelet coherence analysis) is used to analyze the relationship between the UHF signal and the power frequency voltage phase. The analysis result shows that near 50Hz and its multiples, the coherence coefficient of the two is significantly increased, indicating that there is a local discharge stable with the power frequency phase. The operator outputs a strength value representing this resonant relationship. At the same time, another causal resonance operator (transfer entropy) is used to analyze the relationship between the load current and the change of the C2H2 concentration in the oil, and the resonant strength value output by it is close to zero.

[0113] Step S4, the risk analysis module 30 performs state update and risk entropy calculation based on the obtained resonant strength. Since the resonant strength is not zero, and its correlation weight with the "partial discharge" fault mode and the "winding overheating" fault mode and Since the values ​​are significantly positive, the probabilities of these two failure modes are calculated using the log-odds update formula. and Starting from zero, it grows slightly, while the probability of a healthy pattern... Then it decreases accordingly. Therefore, the state superposition vector changes, and the risk entropy... It changes from zero to a small positive value.

[0114] Step S5: Focus control module 40 continuously monitors risk entropy. The value of and its rate of change. When The value exceeds the preset low-risk threshold. At that time, the module sends a control command to the risk analysis module 30, requesting it to increase the state update frequency and activate more cognitive resonance operators related to winding faults, such as operators that specifically analyze the coupling relationship between the enclosure vibration signal and the load current harmonic components.

[0115] Step S6: The risk analysis module 30 continuously records the sequence of state superposition vectors, forming a state evolution path in a multidimensional probability space. The trajectory of this path shows that the device state point is gradually deviating from the coordinate origin representing the healthy state and moving along a specific direction.

[0116] Step S7: The early warning release module 50 receives the analysis results from the risk analysis module 30 and makes a judgment based on the dual-mode criterion. When the DTW distance is lower than the preset similarity threshold... Even if the absolute value of the risk entropy has not yet reached the alarm threshold, the evolution path forbidden zone criterion is triggered. The early warning release module 50 generates a level-one early warning report, which clearly states that the early warning was triggered by "evolution path matching," and that the current evolution trend is highly similar to the "initial development of inter-turn short circuit" mode. The most suspicious fault modes are "partial discharge" and "winding overheating," and a time-frequency resonance diagram of the UHF signal and power frequency voltage phase is attached as evidence for tracing the source. This report is sent to the monitoring center for early intervention by maintenance personnel.

[0117] The embodiments described in this specific implementation are preferred embodiments of this application and are not intended to limit the scope of protection of this application. Identical components are represented by the same reference numerals. Therefore, all equivalent changes made to the structure, shape, and principle of this application should be covered within the scope of protection of this application.

Claims

1. A substation intelligent auxiliary monitoring system, characterized in that, The method comprises the following steps: A data acquisition module is used to acquire multi-source heterogeneous data in a substation in real time; A state vector modeling module is connected with the data acquisition module and is used to generate and dynamically maintain a state superposition vector for each monitored device, the state superposition vector being a probability distribution describing the device being in multiple possible states at the same time; A risk analysis module is connected with the data acquisition module and the state vector modeling module and is adapted to perform multi-source signal coupling analysis by using a cognitive resonance operator to obtain resonance strength, update the state superposition vector based on the resonance strength, calculate risk entropy, and track a state evolution path formed by the state superposition vector changing over time, and perform topological matching of the state evolution path with a preset forbidden region topological library; A focus control module is connected with the risk analysis module and is used to dynamically adjust the calculation priority and sensitivity of the cognitive resonance operator according to the current value and change trend of the risk entropy; An early warning issuing module is connected with the risk analysis module and is used to generate and issue early warning information according to the risk entropy and the result of the topological matching.

2. The intelligent auxiliary monitoring system for a substation of claim 1, wherein, The state superposition vector generated by the state vector modeling module has the following form: ； wherein, is the state vector at time ; is the fully healthy state, is the th potential failure or abnormal mode, is the total number of potential failure or abnormal modes; is the probability that the device is in state at time , and satisfies .

3. The intelligent auxiliary monitoring system for a substation of claim 1, wherein, When calculating the risk entropy, the risk analysis module uses a Shannon information entropy formula: ； where, is the risk entropy at time is the state superposition vector at time is the probability component of the state superposition vector that the device is in state is the total number of potential failure or anomaly modes; summation symbol denotes a summation over all states from the healthy state to all potential failure modes.​​​ 4. The intelligent auxiliary monitoring system for a substation of claim 1, wherein, When updating the state superposition vector based on the resonance strength, the risk analysis module updates the log-odds of each state by the following method: ； in, For state At any moment The logarithmic odds; For state At any moment The logarithmic odds; To update the step size coefficient; For the first The first resonance and the second The association strength weight of each fault state; For the first The resonance relationship at time The resonance intensity; The total number of cognitive harmonic operators; summation symbol This indicates that the summation is performed over all resonance relationships.

5. The intelligent auxiliary monitoring system for a substation of claim 1, wherein, When performing topological matching, the risk analysis module is adapted to use a dynamic time warping algorithm or a Frechet distance algorithm to calculate the similarity of the state evolution path to each forbidden region topology in the forbidden region topological library in real time.

6. The intelligent auxiliary monitoring system for a substation of claim 1, wherein, The early warning information issued by the early warning issuing module is generated based on a double-mode criterion; The double-mode criterion comprises a risk entropy threshold criterion based on the risk entropy and its change rate, and an evolution path forbidden region criterion based on the topological matching result of the state evolution path and the forbidden region topological library.

7. The intelligent auxiliary monitoring system for a substation of claim 1, wherein, The cognitive resonance operator used by the risk analysis module comprises at least one of a time-frequency resonance operator and a causal resonance operator; The time-frequency resonance operator is used to analyze the energy synchronicity of non-stationary signals in a specific time-frequency domain; The causal resonance operator is used to quantify the driving or causal influence strength between different signals.

8. The intelligent auxiliary monitoring system for a substation of claim 1, wherein, When dynamically adjusting, the focus control module is adapted to activate a cognitive resonance operator associated with a potential fault mode that contributes most to the increase of entropy and preset to have high calculation complexity when the risk entropy increases.

9. The intelligent auxiliary monitoring system for a substation of claim 1, wherein, The multi-source heterogeneous data comprises device electrical quantity data, device non-electrical quantity data, and environmental data; The device non-electrical quantity data comprises at least one of infrared thermal imaging data, device operating sound signals, partial discharge ultrahigh frequency signals, vibration signals, and online monitoring data of dissolved gas analysis in transformer oil.

10. A substation intelligent auxiliary monitoring method based on the substation intelligent auxiliary monitoring system of any one of claims 1-9, characterized in that, The method comprises the following steps: Real-time acquisition of multi-source heterogeneous data in a substation; Generation and dynamic maintenance of a state superposition vector for each monitored device, the state superposition vector being a probability distribution describing the device being in multiple possible states at the same time; Based on the multi-source heterogeneous data, multi-source signal coupling analysis is performed by a cognitive resonance operator to obtain resonance strength quantifying the synchronicity or causal correlation between different signal change trends; Based on the resonance strength, the state superposition vector is updated, and risk entropy representing the uncertainty of the device state is calculated; The state evolution path formed by the change of the state superposition vector over time is tracked, and the state evolution path is topologically matched with a preset forbidden region topology library; According to the current value and change trend of the risk entropy, the calculation priority and sensitivity of the cognitive resonance operator are dynamically adjusted; Based on the risk entropy and the results of the topological matching, warning information is generated and released.

Images