Power grid monitoring bionic robot control method for log generation and defect tracking

Abstract

The invention discloses a log generation and defect tracking-oriented power grid monitoring bionic robot control method, which relates to the technical field of power monitoring, and comprises the following steps of: constructing a dynamic causal reasoning network by utilizing log event associated information in a log event list based on a credible log set, calculating a causal influence coefficient of log events and defects, outputting a defect root cause instruction list when the root cause judgment threshold value is reached; a defect root cause instruction list is used as a drive, state records in the credible log set are linked to carry out defect disposal operation, state changes are tracked in real time, and a closed-loop disposal report is generated; and taking the closed-loop disposal report as feedback, updating the defect condition probability in the dynamic causal reasoning network, adjusting the judgment boundary in the reference spectrum set, and performing closed-loop optimization on the generation process of the credible log set and the defect root cause instruction list. The autonomous and intelligent capabilities of power grid monitoring are enhanced, and the safety and reliability of power grid operation are ensured.

Description

Technical Field

[0001] This invention relates to the field of power monitoring technology, and in particular to a biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking. Background Technology

[0002] With the continuous expansion and increasing complexity of power grids, power grid operation monitoring and fault management technologies have developed rapidly. Traditional power grid monitoring methods mainly rely on centralized monitoring platforms and manual inspections, collecting electrical parameters, equipment status, and alarm information from substations to monitor and analyze the power grid's operational status. In recent years, intelligent monitoring methods based on multi-source data fusion have been gradually applied to power grid operation. By integrating power grid operation data, substation electromagnetic spectrum, and dispatching and centralized control alarm information, abnormal events are identified and recorded. Typically, sensor networks are used to achieve multimodal data acquisition, and event logs are generated through data preprocessing, feature extraction, and similarity matching, providing basic data support for subsequent defect analysis and operation and maintenance decisions.

[0003] While existing technologies have made some progress in multi-source data fusion and fault analysis, they still have limitations in the closed-loop processing capabilities of log generation and defect root cause tracking. Existing methods mostly rely on static rules or a single data source for event determination, lacking a comprehensive analysis of causal relationships and dynamic time sequences between events, resulting in limited accuracy in root cause determination under the influence of multiple factors. Defect handling usually requires manual intervention, and there is a lack of mechanisms to feed the handling results back to the event determination and causal analysis models, making it impossible to achieve closed-loop optimization of the monitoring process. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking to solve the automatic closed-loop optimization problem of log generation and defect root cause tracking.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, this invention provides a biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking. The method includes: collecting power grid operation data and substation electromagnetic spectrum; fusing scheduling and centralized control alarm information; calculating confidence weights based on the accuracy of multimodal sensing sensors; generating a log event list; superimposing the log event list and substation electromagnetic spectrum; calculating the similarity score between the real-time spectrum and a reference spectrum set; generating a confidence arbitration value; and outputting a reliable log set when a confidence standard is met; constructing a dynamic causal inference network based on the reliable log set and utilizing the log event association information in the log event list; calculating the causal influence coefficient between log events and defects; and outputting a defect root cause instruction sheet when a root cause determination threshold is reached; using the defect root cause instruction sheet as the driver, linking the status records in the reliable log set to perform defect handling operations, and tracking status changes in real time to generate a closed-loop handling report; using the closed-loop handling report as feedback, updating the defect conditional probability in the dynamic causal inference network, and adjusting the determination boundary in the reference spectrum set to perform closed-loop optimization of the generation process of the reliable log set and the defect root cause instruction sheet.

[0008] As a preferred embodiment of the biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking described in this invention, the specific steps for generating the log event list are as follows:

[0009] The power grid operation data and substation electromagnetic spectrum and dispatch and centralized control alarm information are aligned in time and space, and multi-dimensional log event information sequences are integrated to output a preliminary log event set;

[0010] Electromagnetic interference features, operating status features, and alarm type features are extracted from the initial log event set, and time-series alignment and normalization are performed to generate a multimodal feature matrix.

[0011] Based on the accuracy of the multimodal sensing sensor and historical measurement errors, the multimodal feature matrix is ​​weighted to generate a confidence weight for each log event and output a list of log events with confidence scores.

[0012] Arrange the list of log events with confidence levels in chronological order to form a log event list.

[0013] As a preferred embodiment of the bionic robot control method for power grid monitoring oriented towards log generation and defect tracking described in this invention, the reference spectrum set includes spectral fingerprints collected under different operating conditions, load states, and environmental conditions, and each spectral fingerprint is obtained by recording the amplitude characteristics, harmonic distribution, and high-frequency pulse characteristics of the frequency band.

[0014] As a preferred embodiment of the biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking described in this invention, the specific steps for outputting a reliable log set when a reliability standard is reached are as follows:

[0015] The log event list is overlaid with the substation electromagnetic spectrum at corresponding time steps to output a fused event-spectrum matrix.

[0016] Amplitude features, harmonic distribution features, and high-frequency pulse features are extracted from the fused event-spectrum matrix and integrated into a multi-dimensional electromagnetic spectrum feature set.

[0017] Multidimensional matching and similarity calculation are performed on the feature vector of each log event in the multidimensional feature set of electromagnetic spectrum and the reference spectrum set to obtain the multidimensional matching score of each log event;

[0018] The multidimensional matching score of each log event is weighted and fused according to amplitude weight, harmonic weight and high-frequency pulse weight to obtain a comprehensive similarity score;

[0019] The overall similarity score is compared with the credibility threshold to generate a credibility arbitration value, and a credibility log set is output when the overall similarity score reaches the credibility threshold.

[0020] As a preferred embodiment of the bionic robot control method for power grid monitoring oriented towards log generation and defect tracking described in this invention, the dynamic causal reasoning network is constructed by building nodes and directed edges based on the log event association information and time sequence in the log event list. Each node represents a single log event, and each edge represents the causal relationship between log events. The network is obtained by traversing the network to calculate the conditional probability change of each log event on the occurrence of potential defects.

[0021] As a preferred embodiment of the bionic robot control method for power grid monitoring oriented towards log generation and defect tracking described in this invention, the root cause determination threshold is obtained by analyzing the causal relationship between log events and defects in the trusted log set, as well as the actual contribution of log events to defect determination in manual review feedback, calculating the causal confidence limit for each type of log event, and performing backtracking analysis on the causal relationship of log events based on the handling results after each closed-loop processing, and rolling adjustment.

[0022] As a preferred embodiment of the biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking described in this invention, the specific steps for outputting a defect root cause instruction sheet when the root cause determination threshold is reached are as follows:

[0023] Extract each log event and its time sequence from the trusted log set to generate a log event time series table;

[0024] By utilizing the chronological order of log events and the event association information, each log event is set as a node, and directed edges are used to represent the potential causal relationships between log events, thus constructing a dynamic causal reasoning network.

[0025] In the dynamic causal inference network, the conditional probability changes of each node's log event on the occurrence of potential defects are analyzed one by one to generate causal influence coefficients.

[0026] The causal impact coefficient is compared with the root cause determination threshold to identify log event associations that exceed the root cause determination threshold and generate a defect root cause instruction sheet.

[0027] As a preferred embodiment of the biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking described in this invention, the specific steps for generating a closed-loop handling report are as follows:

[0028] Extract the root cause determination content and corresponding handling plan from the defect root cause instruction sheet, and convert them into an initial handling operation list;

[0029] Based on the comprehensive similarity score in the trusted log set, defect handling operations are performed on the objects in the initial handling operation list, and status change data is recorded.

[0030] The status change data is continuously monitored and dynamically updated. The updated status change data is matched with the initial handling operation list. When the match is consistent, a handling completion mark is generated. When the match is inconsistent, a handling adjustment instruction is generated.

[0031] The completed disposal markers and disposal adjustment instructions are summarized to generate a closed-loop disposal report.

[0032] As a preferred embodiment of the bionic robot control method for power grid monitoring oriented towards log generation and defect tracking described in this invention, the defect conditional probability is obtained by statistically analyzing the handling operation results and state change data in the closed-loop handling report, and feeding back the actual impact of log events on defects to the dynamic causal inference network to adjust the conditional probability distribution between log events and defects.

[0033] As a preferred embodiment of the biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking described in this invention, the specific steps for adjusting the decision boundaries in the reference spectrum set are as follows:

[0034] Extract the matching results and status change data corresponding to each defect root cause instruction sheet from the closed-loop handling report to form a feedback information set;

[0035] The feedback information set is input into the dynamic causal inference network to update the log events and defect condition probability distributions, generating a new set of causal parameters;

[0036] Based on the new set of causal parameters, the decision boundary of the reference spectrum set is re-estimated to generate a revised credibility arbitration standard.

[0037] The comprehensive similarity score is judged by the revised credibility arbitration standard to generate an optimized credibility log set;

[0038] Using the updated defect conditional probability distribution and the optimized trusted log set as input, the root cause determination threshold is calibrated and the determination boundary is adjusted to generate an optimized defect root cause instruction sheet.

[0039] The beneficial effects of this invention are as follows: by constructing a dynamic causal reasoning network and analyzing the conditional probability changes of log events on potential defects, the causal relationship of log events is quantified, enabling the determination of the root cause of defects to be dynamically adjusted according to the actual conditional probability, thereby improving the accuracy and interpretability of the determination; the generated defect root cause instruction sheet can be directly used to guide defect handling operations, realize closed-loop optimization of the defect handling process, shorten response time, improve processing efficiency, and enhance the autonomy and intelligence of power grid monitoring, ensuring the safety and reliability of power grid operation. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a flowchart of the biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking in this invention.

[0042] Figure 2 This is a flowchart illustrating the generation of the log event list in this invention.

[0043] Figure 3 This is a flowchart of the process for outputting a trusted log set in this invention.

[0044] Figure 4 This is a flowchart of the closed-loop processing report generation process in this invention. Detailed Implementation

[0045] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0046] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0047] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0048] Reference Figures 1-4 This is one embodiment of the present invention, which provides a biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking, including the following steps:

[0049] S1. Collect power grid operation data and substation electromagnetic spectrum, integrate dispatch and centralized control alarm information, calculate confidence weights based on the accuracy of multimodal sensing sensors, and generate a log event list.

[0050] The system performs time and space alignment on power grid operation data, substation electromagnetic spectrum, and dispatch and centralized control alarm information, and integrates multi-dimensional log event information sequences to output a preliminary log event set.

[0051] Specifically, power grid operation data includes voltage, current, frequency, power, and operating status information; substation electromagnetic spectrum includes amplitude characteristics, harmonic distribution, and high-frequency pulse characteristics; dispatching and centralized control alarm information includes alarm type, alarm level, and alarm time.

[0052] Align power grid operation data, substation electromagnetic spectrum, and dispatch and centralized control alarm information according to timestamps. For example, use millisecond-level timestamps as a benchmark to interpolate or truncate data from different sources to ensure that corresponding data records are formed at the same point in time.

[0053] The power grid operation data, substation electromagnetic spectrum and dispatch and centralized control alarm information are spatially aligned according to geographical location and collection point number. For example, the substation location code is used as a unified index to integrate the data of different collection points into the corresponding spatial coordinates.

[0054] After completing time and space alignment, the voltage, current, and frequency information in the power grid operation data, the amplitude characteristics, harmonic distribution, and high-frequency pulse characteristics in the electromagnetic spectrum of the substation, and the alarm types, alarm levels, and alarm times in the dispatching and centralized control alarm information are fused into a multi-dimensional log event information sequence in chronological order. The fused multi-dimensional log event information sequence is then output to form a preliminary log event set.

[0055] Electromagnetic interference features, operating status features, and alarm type features are extracted from the initial log event set, and time-series alignment and normalization are performed to generate a multimodal feature matrix.

[0056] Specifically, the electromagnetic spectrum data of the substation contained in the preliminary log event set is processed to extract electromagnetic interference features, including peak point extraction of frequency band amplitude features, such as identifying abnormal peak points through amplitude curves, Fourier decomposition of harmonic distribution features and extraction of harmonic distribution ratio, and pulse detection of high-frequency pulse features to obtain transient interference signal parameters.

[0057] The power grid operation data contained in the preliminary log event set is processed to extract operation status characteristics, including mean and fluctuation range analysis of voltage data, time-series fluctuation trend calculation of current data, deviation statistics of frequency data, and load curve extraction of power data. For example, the numerical characteristics of voltage, current and power changes over time are recorded under different load conditions.

[0058] The scheduling and centralized control alarm information contained in the preliminary log event set is processed to extract alarm type features, including converting alarm type into independent feature items, alarm level into numerical identifier, and alarm time as time-series index feature. For example, during a power grid switching operation, the corresponding alarm category, level and timestamp are recorded.

[0059] Electromagnetic interference features, operating status features, and alarm type features are time-series aligned and normalized respectively, and the three types of features are combined to generate a multimodal feature matrix.

[0060] Based on the accuracy of the multimodal sensing sensor and historical measurement errors, the multimodal feature matrix is ​​weighted to generate a confidence weight for each log event and output a list of log events with confidence scores.

[0061] Specifically, when performing weighted calculations on the multimodal feature matrix, a correspondence is first established column-by-column according to the channels and measurement types of the multimodal sensing sensor. The accuracy parameters of the multimodal sensing sensor are then read and converted into the accuracy variance of each feature (for example, for a temperature channel accuracy of ±0.1℃, the accuracy variance can be taken as 0.1). 2 The system synchronously obtains the historical error variance of each feature from historical measurement errors, calculates the feature-level confidence weight coefficient according to the inverse variance weighting principle, multiplies the feature-level confidence weight coefficient with the feature vector of the multimodal feature matrix at the same time point element by element and sums them according to the weight normalization, and obtains the log event confidence weight for the corresponding time point. The system then adds log event confidence weight to each log event in chronological order and outputs a list of log events with confidence.

[0062] It should also be noted that historical measurement error refers to the long-term measurement deviation statistics formed by multimodal sensing sensors under the same range and the same measurement conditions. The method of obtaining the historical error is to perform time window statistics on the calibration records of multimodal sensing sensors, the comparison results with reference instruments, and repeated measurement data, and calculate the residual standard deviation, root mean square error, or percentile error of each feature (for example, taking the root mean square error of data from the most recent period as an estimate of historical error variance). Based on this, the historical error variance is provided for the calculation of confidence weight.

[0063] Arrange the list of log events with confidence levels in chronological order to form a log event list.

[0064] Specifically, the timestamp of each log event in the list of log events with confidence is first standardized, then sorted in ascending order by timestamp and in descending order by confidence weight if the timestamps are the same. For log events that are missing millisecond timestamps, linear interpolation is performed on adjacent valid timestamps or the data is truncated to the nearest time step (e.g., aligned to millisecond granularity).

[0065] For duplicate log events occurring within the same time step, perform deduplication and merging according to predetermined rules (e.g., retain the log event with the highest confidence weight within the same time step, or perform a weighted average of the confidence weights and retain one merged result). Renumber the log events in the order of time and verify the time continuity. Output a list of log events arranged in chronological order with confidence weights as the log event list.

[0066] S2. Overlay the log event list and the substation electromagnetic spectrum, calculate the similarity score between the real-time spectrum and the reference spectrum set, generate a credibility arbitration value, and output a credible log set when the credibility standard is met.

[0067] The log event list is overlaid with the substation electromagnetic spectrum at corresponding time steps to output a fused event-spectrum matrix.

[0068] Specifically, the timestamps in the log event list and the electromagnetic spectrum of the substation are standardized in terms of format and time zone, and the time steps are determined (e.g., in milliseconds or seconds), and the standardized time index is output.

[0069] Based on the standardized time index, the electromagnetic spectrum of the substation is resampled or sliced ​​using a window function. Short-time Fourier transform or sliding window energy statistics are used to extract the spectrum frame or frequency band energy in each time step, and the spectrum frame sequence divided by time step is output.

[0070] Map the log event list to a spectrum frame sequence using a normalized time index. If the log event includes a duration, take the spectrum frames within the corresponding time step and aggregate them over time (e.g., take the mean or weighted average), and output the spectrum summary vector corresponding to the event.

[0071] Extract time-series fields and operational values ​​such as voltage, current, frequency, alarm type, and confidence weight from the log event list, and concatenate or fuse them with the corresponding spectrum summary vector (e.g., concatenate amplitude vector or calculate harmonic energy ratio) to output the multimodal feature vector of each event.

[0072] All multimodal feature vectors are arranged in time step order. Missing values ​​are filled by linear interpolation or forward imputation and normalized as necessary. The output is a fused event-spectrum matrix, where the matrix rows correspond to time steps or log events and the matrix columns correspond to running features and spectrum features.

[0073] Amplitude features, harmonic distribution features, and high-frequency pulse features are extracted from the fused event-spectrum matrix and integrated into a multi-dimensional electromagnetic spectrum feature set.

[0074] Specifically, the fused event-spectrum matrix is ​​subjected to a short-time Fourier transform within a predetermined time window, and the spectrum frames are segmented and noise baselines are removed to output a time-by-time spectrum frame sequence. The spectrum envelope, maximum value of the frequency band, root mean square of the frequency band, and peak value are calculated for the time-by-time spectrum frame sequence, and a set of amplitude feature vectors is output.

[0075] The fundamental frequency is estimated for each time-by-time spectral frame sequence, and spectral peak detection and amplitude ratio statistics are performed at integer multiples of the frequency. Harmonic amplitude and harmonic energy proportion are extracted, and a set of harmonic distribution feature vectors is output. High-pass filtering and envelope detection are applied to each time-by-time spectral frame sequence, and transient pulses are identified by energy mutation detection or kurtosis statistics. Pulse start and end times, pulse amplitude, pulse energy, and pulse spectral distribution are extracted, and a set of high-frequency pulse feature vectors is output.

[0076] The amplitude feature vector set, harmonic distribution feature vector set, and high-frequency pulse feature vector set are aligned and concatenated according to time index, and then standardized to output a multidimensional feature set of the electromagnetic spectrum.

[0077] It should also be noted that the predetermined time window is set as follows: the minimum resolvable time interval is determined based on the sampling rate of the power grid operation data and the electromagnetic spectrum of the substation. For example, the minimum time interval is 1ms when the sampling rate of the power grid operation data is 1kHz. The appropriate window length is determined based on the typical duration range of the electromagnetic interference characteristics and alarm type characteristics. For example, 200ms is selected for power frequency interference and 20ms is selected for high frequency pulses. Then, combined with the historical measurement error distribution, a window length that can cover the main event cycle and ensure frequency domain resolution is selected as the predetermined time window. Finally, the entire fused event-spectrum matrix is ​​segmented according to the predetermined time window length.

[0078] Multidimensional matching and similarity calculation are performed on the feature vector of each log event in the multidimensional feature set of electromagnetic spectrum and the reference spectrum set to obtain a multidimensional matching score for each log event.

[0079] Specifically, extract the feature vector x = [x] of a single log event from the multidimensional feature set of the electromagnetic spectrum. amp ,x harm ,x pulse [and each reference fingerprint r in the reference spectrum set] k =[r k,amp ,r k,harm ,r k,pulse Corresponding match;

[0080] Where x represents the electromagnetic spectrum multidimensional feature vector of the log event, x amp x represents the magnitude feature vector of a log event. harm The characteristic vector representing the harmonic distribution of log events, x pulse The high-frequency pulse feature vector, r, represents the log event. k Let r represent the multidimensional eigenvector of the electromagnetic spectrum of the k-th reference spectrum in the reference spectrum set. k,amp Let r represent the amplitude eigenvector of the k-th reference spectrum. k,harm Let r represent the harmonic distribution eigenvector of the k-th reference spectrum. k,pulse This represents the high-frequency pulse feature vector of the k-th reference spectrum, where k represents the index in the set of reference spectra;

[0081] Calculate the Euclidean distance and harmonic cosine similarity for the amplitude component, harmonic component, and high-frequency pulse component, respectively.

[0082] The Euclidean distance expression is:

[0083]

[0084] Where, d amp The distance between the log event amplitude feature vector and the k-th reference spectrum amplitude feature vector is represented by i, where i represents the index of the amplitude feature component, and x represents the distance between the log event amplitude feature vector and the reference spectrum amplitude feature vector. amp,i r represents the i-th amplitude component of the log event amplitude feature vector. k,amp,i This represents the i-th amplitude component of the amplitude feature vector of the k-th spectral fingerprint in the reference spectrum set;

[0085] Harmonic cosine similarity, expressed as:

[0086]

[0087] Among them, cos harmThe cosine similarity between the harmonic feature vector of the log event and the harmonic feature vector of the k-th reference spectrum is represented by ||·||, where ||·|| represents the vector norm, i.e., the Euclidean length.

[0088] The high-frequency pulse energy ratio can be expressed as the normalized energy difference:

[0089]

[0090] Among them, e pulse x represents the normalized energy difference between the high-frequency pulse feature vector of the log event and the high-frequency pulse feature vector of the k-th reference spectrum. pulse,j The j-th component of the high-frequency pulse feature vector of the log event is represented by r. k,pulse,j ε represents the j-th component of the high-frequency pulse feature vector of the k-th spectral fingerprint in the reference spectrum set, ε represents the smallest positive number to prevent division by zero, and j represents the index of the high-frequency pulse feature component;

[0091] Normalize the distance / similarity metric into similarity components, such as magnitude similarity, with the expression:

[0092]

[0093] Among them, s amp This represents the normalized amplitude similarity score. This represents the maximum Euclidean distance between the amplitude characteristic components in the reference spectrum set, used for normalization;

[0094] The harmonic similarity is taken as follows:

[0095]

[0096] Among them, s harm This represents the normalized harmonic similarity score.

[0097] The pulse similarity is taken as follows:

[0098] s pulse =1-e pulse ;

[0099] Among them, s pulse This represents the normalized high-frequency pulse similarity score.

[0100] Weighted by amplitude w amp Harmonic weight w harm With high-frequency pulse weight w pulse (and w) amp +w harm +w pulse =1) Weighted fusion yields a multidimensional matching score for the k-th reference fingerprint, expressed as:

[0101] score k =w amp s amp +w harm s harm +w pulse s pulse ;

[0102] Among them, score k This represents the multidimensional matching score of the log event for the k-th reference spectrum;

[0103] The multidimensional matching score for each log event is obtained by repeatedly calculating the reference fingerprints in the reference spectrum set.

[0104] The multidimensional matching scores for each log event are weighted and fused according to amplitude weight, harmonic weight, and high-frequency pulse weight to obtain a comprehensive similarity score.

[0105] Specifically, the matching score of log event amplitude feature is proportionally allocated according to amplitude weight, the matching score of log event harmonic distribution feature is proportionally allocated according to harmonic weight, and the matching score of log event high-frequency pulse feature is proportionally allocated according to high-frequency pulse weight; the weighted parts of amplitude feature, harmonic feature, and high-frequency pulse feature are summed to generate a comprehensive similarity score of log event.

[0106] The comprehensive similarity scores are organized according to the time sequence of the log events to form a fusion score list for subsequent credibility arbitration. For example, when the amplitude weight is 0.5, the harmonic weight is 0.3, and the high-frequency pulse weight is 0.2, a comprehensive similarity score can be obtained.

[0107] Furthermore, the reference spectrum set includes spectral fingerprints collected under different operating conditions, load states, and environmental conditions. Each spectral fingerprint is obtained by recording the amplitude characteristics, harmonic distribution, and high-frequency pulse characteristics of the frequency band.

[0108] Specifically, the electromagnetic spectrum of the substation corresponding to the power grid operation data is collected under different operating conditions, and the amplitude characteristics of the frequency band are recorded. For example, the spectrum signal with an amplitude range of 0–100dB is collected under high load and low load conditions respectively; the electromagnetic spectrum of the substation is collected under different load conditions, and the harmonic distribution characteristics are extracted, such as the amplitude and phase information of the fundamental frequency and the second and third harmonics.

[0109] Electromagnetic spectra of substations are collected under different environmental conditions to extract high-frequency pulse characteristics, such as the amplitude and duration of short-duration high-amplitude pulses. The amplitude characteristics, harmonic distribution characteristics, and high-frequency pulse characteristics are integrated according to time and acquisition conditions to form each spectral fingerprint, and all spectral fingerprint sets are used to generate a reference spectrum set.

[0110] The overall similarity score is compared with the credibility threshold to generate a credibility arbitration value, and a credibility log set is output when the overall similarity score reaches the credibility threshold.

[0111] Specifically, the overall similarity score is compared with the credibility threshold one by one. Log events that meet or exceed the credibility threshold are marked as high credibility events. At the same time, the comparison result of each log event is recorded to generate a credibility arbitration value. When the overall similarity score reaches the credibility threshold, the corresponding log events are summarized and output as a credibility log set. For example, when the overall similarity score reaches 0.85 or above, the included log events are included in the credibility log set.

[0112] It should also be noted that the steps for setting a trust threshold are as follows: collect historical trust log events and comprehensive similarity scores under different operating conditions, load states, and environmental conditions; perform statistical analysis on the comprehensive similarity scores, such as calculating the mean, standard deviation, or quantiles, to determine the distribution characteristics of the comprehensive similarity scores; select a trust threshold according to the required stringency of trust judgment, for example, selecting the high quantile value of the score distribution as an example of the trust threshold to obtain a value for subsequent comparison; finally, record the trust threshold and apply it to the trust judgment process of the comprehensive similarity score.

[0113] S3. Based on the trusted log set, utilize the log event association information in the log event list to construct a dynamic causal reasoning network, calculate the causal influence coefficient between log events and defects, and output a defect root cause instruction sheet when the root cause determination threshold is reached.

[0114] Extract each log event and its time sequence from the trusted log set to generate a log event time series table.

[0115] Specifically, each log event and timestamp information is read from the trusted log set and sorted in ascending order according to the timestamp to generate a log event time series table arranged in chronological order. For example, the time format is unified to YYYY-MM-DDHH:MM:SS and log events with time intervals less than the minimum interval set based on the sampling frequency are merged or marked, ultimately forming a continuous log event time series table, ensuring that each log event corresponds to a unique time step in the sequence table.

[0116] By utilizing the chronological order of log events and the event association information, each log event is set as a node, and directed edges are used to represent the potential causal relationships between log events, thus constructing a dynamic causal reasoning network.

[0117] Furthermore, the dynamic causal reasoning network is constructed by building nodes and directed edges based on the correlation information and chronological order of log events in the log event list. Each node represents a single log event, and each edge represents the causal relationship between log events. The network is traversed to calculate the conditional probability change of each log event in relation to the occurrence of potential defects.

[0118] Specifically, each log event and its associated information are read sequentially from the log event list. Each log event is set as a node, and the possible causal direction is determined based on the order of the log events in the time series table. Directed edges between nodes are determined based on the association information between log events. For example, when event A precedes event B and there is a significant correlation, a directed edge is established from node A to node B. The constructed node and edge network is traversed, the conditional probability change of each log event on the occurrence of potential defects is calculated, the conditional probability update value of each node is recorded, and finally, a complete dynamic causal reasoning network is generated.

[0119] In the dynamic causal reasoning network, the conditional probability changes of each node's log events on the occurrence of potential defects are analyzed one by one to generate causal influence coefficients.

[0120] Specifically, the log events of each node are read sequentially from the dynamic causal inference network to obtain the initial conditional probability and the joint probability of the potential defect of the node; the causal path of the node is traversed along the outgoing edges of the node and the subsequent nodes, and the difference between the occurrence and non-occurrence of the node on the conditional probability of the potential defect is calculated; the changes in the conditional probabilities of all paths are accumulated or averaged to obtain the causal influence coefficient of the current node on the potential defect, for example by calculating and normalizing the difference in the defect probability under the condition of the node occurring and not occurring; the steps are repeated until the causal influence coefficients of all nodes are generated.

[0121] The causal impact coefficient is compared with the root cause determination threshold to identify log event associations that exceed the root cause determination threshold and generate a defect root cause instruction sheet.

[0122] Specifically, the causal impact coefficient of each node log event in the dynamic causal inference network is read sequentially and compared with the root cause determination threshold. Log events and their associated paths with causal impact coefficients higher than the root cause determination threshold are identified. The log events and their corresponding associated information are organized and encoded in chronological order to generate a defect root cause instruction sheet. For example, the instruction sheet records the log event identifier, occurrence time, type of defect affected, and associated path information. This process is repeated until all log events that meet the conditions are associated and a defect root cause instruction sheet is generated.

[0123] Furthermore, the root cause determination threshold is obtained by analyzing the causal relationship between log events and defects in the trusted log set, as well as the actual contribution of log events to defect determination in manual review feedback, calculating the causal confidence limit for each type of log event, and then performing a backtracking analysis on the causal relationship of log events based on the handling results after each closed-loop processing, and making rolling adjustments.

[0124] Specifically, statistical analysis is performed on the causal relationship between each log event and the defect in the trusted log set, and the conditional probability distribution of each type of log event for defect judgment is calculated; feedback information from manual review is collected to quantify the contribution of log events to defect occurrence in actual judgment, such as by recording the correct judgment rate or the intensity of influence; the conditional probability distribution is combined with the contribution of manual review to calculate the causal confidence limit for each type of log event, forming a preliminary root cause judgment threshold.

[0125] After each closed-loop process, the processing results are obtained, and the causal relationship changes of the relevant log events are analyzed retrospectively. For example, the conditional probability weight of log events to defects is adjusted, and the causal confidence limits of each type of log event are continuously corrected based on the updated causal analysis to obtain the dynamically adjusted root cause determination threshold.

[0126] It should be noted that the temporal information and event correlations of multidimensional log data are transformed into a dynamic causal inference network, and the conditional probability change of each log event on potential defects is quantified to generate causal influence coefficients. Root cause determination no longer relies on static rules or empirical judgments, but can dynamically calculate the probability of defect occurrence based on actual observation data, thereby outputting a defect root cause instruction sheet to directly guide the handling operation. This realizes log data-driven causal quantitative analysis and automated root cause identification, improving the accuracy of determination, interpretability, and defect response efficiency.

[0127] S4. Driven by the defect root cause instruction sheet, the defect handling operation is carried out in conjunction with the status records in the trusted log set, and the status changes are tracked in real time to generate a closed-loop handling report.

[0128] Extract the root cause determination content and corresponding handling plan from the defect root cause instruction sheet, and convert them into an initial handling operation list.

[0129] Specifically, for each record in the defect root cause instruction sheet, the root cause determination content is identified, for example, by identifying the log event category, potential defect type, and related causal information, which will be used as the root cause determination content; the handling plan associated with each root cause determination content is read, for example, by matching the operation steps, remedial measures, or adjustment suggestions specified in the instruction sheet with the corresponding root cause determination content; based on the matching results, each root cause determination content and its corresponding handling plan are integrated to form an initial handling operation list. Each list item includes the root cause determination content identifier, related log events, and specific handling operations, such as the operation object, operation type, and operation sequence, completing the structured transformation from the defect root cause instruction sheet to the initial handling operation list.

[0130] Based on the comprehensive similarity score in the trusted log set, defect handling operations are performed on the objects in the initial handling operation list, and status change data is recorded.

[0131] Specifically, for each operation record in the initial handling operation list, the handling object is identified. The handling object specifies the specific equipment, component, or functional unit that needs to be handled, such as a substation switch, relay protection device, or communication link. Based on the comprehensive similarity score in the trusted log set, the corresponding defect handling operation is performed on each handling object in descending order of score. This includes operations such as adjustment, reset, replacement, or parameter optimization. At the same time, the status change data of the handling object is recorded during the operation, including the status before the operation, the status after the operation, and the operation time. Finally, each handling operation and its corresponding status change data are summarized to form a complete defect handling execution record, completing the transformation from the initial handling operation list to the defect handling status record.

[0132] The status change data is continuously monitored and dynamically updated. The updated status change data is matched with the initial handling operation list. When the match is consistent, a handling completion mark is generated. When the match is inconsistent, a handling adjustment instruction is generated.

[0133] Specifically, continuously collected status change data is recorded and updated. Each updated status change data is organized in chronological order and matched one by one with the corresponding disposal object and expected status in the initial disposal operation list. If the updated status change data is completely consistent with the expected status of the disposal object in the initial disposal operation list, the disposal is marked as completed in the disposal object record, for example, a disposal completion mark is generated. If the updated status change data differs from the expected status, a disposal adjustment instruction is generated in the disposal object record, indicating that further defect disposal operations need to be performed, such as readjustment, reset, or parameter optimization.

[0134] The completed disposal markers and disposal adjustment instructions are summarized to generate a closed-loop disposal report.

[0135] Specifically, the completion markers and adjustment instructions generated for all disposal objects are organized and summarized according to time sequence and disposal object. The completion markers record the completed defect disposal operation status for each disposal object, and the adjustment instructions record the adjustment operations that still need to be performed for each disposal object and the corresponding reasons. The log order and disposal object order form a structured entry, and all entries are integrated to generate a closed-loop disposal report. The closed-loop disposal report includes the final disposal status of each disposal object, descriptions of incomplete or adjustment operations, as well as timestamps and operation sequence examples, to fully record the closed-loop disposal process.

[0136] S5. Using the closed-loop handling report as feedback, update the defect conditional probability in the dynamic causal reasoning network and adjust the decision boundary in the reference spectrum set to perform closed-loop optimization on the generation process of the trusted log set and the defect root cause instruction sheet.

[0137] Extract the matching results and status change data corresponding to each defect root cause instruction sheet from the closed-loop handling report to form a feedback information set.

[0138] Specifically, based on the identification information of each defect root cause instruction sheet in the closed-loop handling report, the corresponding matching results and status change data are extracted. The matching results record whether the handling object of each defect root cause instruction sheet is consistent with the initial handling operation list. The status change data records the operation status and change information of each defect root cause instruction sheet during the handling process. All items are organized according to the order of defect root cause instruction sheets and time sequence, and integrated into a feedback information set. The feedback information set includes the matching results, status change data, timestamps and operation sequence examples of each defect root cause instruction sheet, which are used for subsequent analysis and adjustment of the closed-loop handling results.

[0139] The feedback information set is input into the dynamic causal inference network to update the log events and defect conditional probability distributions, generating a new set of causal parameters.

[0140] Specifically, the feedback information set is mapped to nodes in the dynamic causal inference network according to the identifier and time sequence of each log event. The conditional probability distribution of each log event is updated. The conditional probability of the log event for the occurrence of potential defects is adjusted based on the matching results in the feedback information set. The probability values ​​of causal edges in the network are corrected by combining state change data. The entire dynamic causal inference network is traversed to complete the conditional probability update of all nodes and edges. The update parameters of each log event and related causal edges are organized to generate a new causal parameter set. The new causal parameter set contains the update conditional probability of each log event and an example of the update probability of each causal edge, which is used for subsequent closed-loop optimization.

[0141] Furthermore, the defect conditional probability is obtained by statistically analyzing the handling operation results and status change data in the closed-loop handling report, and feeding back the actual impact of log events on defects to the dynamic causal inference network to adjust the conditional probability distribution between log events and defects.

[0142] Specifically, the handling operation results corresponding to each defect root cause instruction in the closed-loop handling report are extracted. The status change data of the handled object recorded in the handling operation results are compared with the initial handling operation list. The frequency of each log event causing the defect to occur or not occur under different handling operations is counted. The actual impact ratio of each log event on the occurrence of the defect is calculated. Based on the ratio, the conditional probability distribution between each log event and the defect in the dynamic causal inference network is updated to form the defect conditional probability. The handling operation result refers to the status change of the handled object after each defect root cause instruction is executed, such as the example of normal or abnormal status change.

[0143] Based on the new set of causal parameters, the decision boundary of the reference spectrum set is re-estimated, and a revised credibility arbitration standard is generated.

[0144] Specifically, amplitude features, harmonic distribution, and high-frequency pulse features are extracted from each spectral fingerprint in the reference spectrum set. Combined with the conditional probability distribution between each log event and defect in the new causal parameter set, the matching boundary between the spectral fingerprint and the log event is analyzed one by one. This allows the credibility score of each spectral fingerprint under the influence of the new conditional probability to accurately reflect the probability of potential defects occurring, and ultimately forms the revised credibility arbitration standard.

[0145] The revised credibility arbitration standard is used to judge the comprehensive similarity score, and an optimized credibility log set is generated.

[0146] Specifically, the comprehensive similarity score of each log event is compared item by item with the amplitude feature threshold, harmonic distribution threshold, and high-frequency pulse threshold in the revised credibility arbitration standard. When the comprehensive similarity score meets the judgment conditions of the revised credibility arbitration standard, the corresponding log event is included in the optimized trusted log set; otherwise, the log event is excluded. For example, when the amplitude feature score is greater than the example threshold of 0.6, the harmonic distribution score is within the range of the example threshold of 4–7, and the high-frequency pulse score is greater than the example threshold of 0.3, the log event is recorded and added to the optimized trusted log set.

[0147] It should also be noted that the amplitude characteristic threshold, harmonic distribution threshold, and high-frequency pulse threshold are obtained by analyzing the spectral fingerprints collected under different operating conditions, load states, and environmental conditions in the reference spectrum set. The amplitude characteristics, harmonic distribution, and high-frequency pulse characteristics of each type of spectral fingerprint are statistically calculated. For example, the mean and standard deviation of the amplitude characteristics, the typical frequency range of the harmonic distribution, and the peak distribution of the high-frequency pulse characteristics are calculated. The statistical results are used as the initial thresholds, and the initial thresholds are then adjusted in a rolling manner in combination with the closed-loop processing feedback information to obtain the amplitude characteristic threshold, harmonic distribution threshold, and high-frequency pulse threshold used for comprehensive similarity scoring.

[0148] Using the updated defect conditional probability distribution and the optimized trusted log set as input, the root cause determination threshold is calibrated and the determination boundary is adjusted to generate an optimized defect root cause instruction sheet.

[0149] Specifically, each log event in the optimized trusted log set is matched with the updated defect conditional probability distribution to determine the conditional probability value of each log event for the occurrence of the defect. The root cause determination threshold is adjusted based on the statistical results. For example, if the conditional probability of a log event exceeds the root cause determination threshold, it is determined to be a potential root cause. At the same time, the boundary of the root cause determination threshold is modified to adapt to different defect types. Log events that meet the adjusted determination conditions and their corresponding handling solutions are summarized to generate an optimized defect root cause instruction sheet.

[0150] It should be noted that after generating the optimized defect root cause instruction sheet, it can provide precise operational basis and judgment rules for the control method of bionic robot for power grid monitoring oriented towards log generation and defect tracking. The instruction sheet clarifies the correspondence between each log event and potential defects and the priority order of handling, enabling the bionic robot to accurately locate abnormal points and perform targeted operations during inspection, defect identification and handling. It can also dynamically adjust the inspection path and handling strategy according to real-time log updates, thereby achieving closed-loop defect tracking and efficient control.

[0151] By using closed-loop handling reports as feedback to update the defect conditional probabilities in the dynamic causal inference network in real time and adjusting the decision boundaries of the reference spectrum set, adaptive optimization of the generation of trusted log sets and defect root cause instruction sheets can be achieved. Compared with existing technologies that typically rely on static decision criteria and fixed conditional probabilities, this method can dynamically correct causal relationships and decision boundaries, making the correlation analysis between log events and defects more accurate and the root cause determination more reliable. It also improves the response speed and autonomy of defect handling, thereby significantly optimizing the decision-making efficiency and operational reliability of power grid monitoring.

[0152] This embodiment also provides a computer device applicable to the control method of a bionic robot for power grid monitoring oriented towards log generation and defect tracking, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the control method of a bionic robot for power grid monitoring oriented towards log generation and defect tracking as proposed in the above embodiment.

[0153] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0154] This embodiment also provides a storage medium on which a computer program is stored. When executed by a processor, the program implements the control method for a bionic robot for power grid monitoring, which is oriented towards log generation and defect tracking, as proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0155] In summary, this invention quantifies the causal relationships of log events by constructing a dynamic causal reasoning network and analyzing the conditional probability changes of log events on potential defects. This allows the determination of root causes of defects to be dynamically adjusted based on actual conditional probabilities, thereby improving the accuracy and interpretability of the determination. The generated root cause instruction sheet can be directly used to guide defect handling operations, achieving closed-loop optimization of the defect handling process, shortening response time, improving processing efficiency, and enhancing the autonomy and intelligence of power grid monitoring, ensuring the safety and reliability of power grid operation.

[0156] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking, characterized in that: include, Collect power grid operation data and substation electromagnetic spectrum, integrate dispatch and centralized control alarm information, combine the accuracy of multimodal sensing sensors to calculate confidence weights, and generate a log event list; The log event list and the electromagnetic spectrum of the substation are superimposed to calculate the similarity score between the real-time spectrum and the reference spectrum set, generate a credibility arbitration value, and output a credibility log set when the credibility standard is met. Based on a trusted log set, a dynamic causal reasoning network is constructed using the log event association information in the log event list. The causal influence coefficient between log events and defects is calculated, and a defect root cause instruction sheet is output when the root cause determination threshold is reached. Driven by the defect root cause instruction sheet, the system links the status records in the trusted log collection to perform defect handling operations, tracks status changes in real time, and generates a closed-loop handling report. Using the closed-loop handling report as feedback, the defect conditional probability in the dynamic causal reasoning network is updated, and the decision boundary in the reference spectrum set is adjusted to perform closed-loop optimization on the generation process of the trusted log set and the defect root cause instruction sheet.

2. The biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking as described in claim 1, characterized in that: The specific steps for generating the log event list are as follows. The power grid operation data and substation electromagnetic spectrum and dispatch and centralized control alarm information are aligned in time and space, and multi-dimensional log event information sequences are integrated to output a preliminary log event set; Electromagnetic interference features, operating status features, and alarm type features are extracted from the initial log event set, and time-series alignment and normalization are performed to generate a multimodal feature matrix. Based on the accuracy of the multimodal sensing sensor and historical measurement errors, the multimodal feature matrix is ​​weighted to generate a confidence weight for each log event and output a list of log events with confidence scores. Arrange the list of log events with confidence levels in chronological order to form a log event list.

3. The biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking as described in claim 1, characterized in that: The reference spectrum set includes spectral fingerprints collected under different operating conditions, load states, and environmental conditions. Each spectral fingerprint is obtained by recording the amplitude characteristics, harmonic distribution, and high-frequency pulse characteristics of the frequency band.

4. The biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking as described in claim 1, characterized in that: The process of outputting a trusted log set when the trust standard is met is as follows: The log event list is overlaid with the substation electromagnetic spectrum at corresponding time steps to output a fused event-spectrum matrix. Amplitude features, harmonic distribution features, and high-frequency pulse features are extracted from the fused event-spectrum matrix and integrated into a multi-dimensional electromagnetic spectrum feature set. Multidimensional matching and similarity calculation are performed on the feature vector of each log event in the multidimensional feature set of electromagnetic spectrum and the reference spectrum set to obtain the multidimensional matching score of each log event; The multidimensional matching score of each log event is weighted and fused according to amplitude weight, harmonic weight and high-frequency pulse weight to obtain a comprehensive similarity score; The overall similarity score is compared with the credibility threshold to generate a credibility arbitration value, and a credibility log set is output when the overall similarity score reaches the credibility threshold.

5. The biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking as described in claim 1, characterized in that: The dynamic causal reasoning network is constructed by building nodes and directed edges based on the correlation information and chronological order of log events in the log event list. Each node represents a single log event, and each edge represents the causal relationship between log events. The network is traversed to calculate the conditional probability change of each log event in relation to the occurrence of potential defects.

6. The biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking as described in claim 1, characterized in that: The root cause determination threshold is obtained by analyzing the causal relationship between log events and defects in the trusted log set, as well as the actual contribution of log events to defect determination in manual review feedback, calculating the causal confidence limit for each type of log event, and performing a backtracking analysis on the causal relationship of log events based on the handling results after each closed-loop processing, and making rolling adjustments.

7. The biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking as described in claim 1, characterized in that: When the root cause determination threshold is reached, a defect root cause instruction sheet is output. The specific steps are as follows: Extract each log event and its time sequence from the trusted log set to generate a log event time series table; By utilizing the chronological order of log events and the event association information, each log event is set as a node, and directed edges are used to represent the potential causal relationships between log events, thus constructing a dynamic causal reasoning network. In the dynamic causal inference network, the conditional probability changes of each node's log event on the occurrence of potential defects are analyzed one by one to generate causal influence coefficients. The causal impact coefficient is compared with the root cause determination threshold to identify log event associations that exceed the root cause determination threshold and generate a defect root cause instruction sheet.

8. The biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking as described in claim 1, characterized in that: The specific steps for generating the closed-loop handling report are as follows. Extract the root cause determination content and corresponding handling plan from the defect root cause instruction sheet, and convert them into an initial handling operation list; Based on the comprehensive similarity score in the trusted log set, defect handling operations are performed on the objects in the initial handling operation list, and status change data is recorded. The status change data is continuously monitored and dynamically updated. The updated status change data is matched with the initial handling operation list. When the match is consistent, a handling completion mark is generated. When the match is inconsistent, a handling adjustment instruction is generated. The completed disposal markers and disposal adjustment instructions are summarized to generate a closed-loop disposal report.

9. The biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking as described in claim 1, characterized in that: The defect conditional probability is obtained by statistically analyzing the handling operation results and status change data in the closed-loop handling report, and feeding back the actual impact of log events on defects to the dynamic causal inference network to adjust the conditional probability distribution between log events and defects.

10. The biomimetic robot control method for power grid monitoring oriented towards log generation and defect tracking as described in claim 1, characterized in that: The specific steps for adjusting the decision boundaries in the reference spectrum set are as follows. Extract the matching results and status change data corresponding to each defect root cause instruction sheet from the closed-loop handling report to form a feedback information set; The feedback information set is input into the dynamic causal inference network to update the log events and defect condition probability distributions, generating a new set of causal parameters; Based on the new set of causal parameters, the decision boundary of the reference spectrum set is re-estimated to generate a revised credibility arbitration standard. The comprehensive similarity score is judged by the revised credibility arbitration standard to generate an optimized credibility log set; Using the updated defect conditional probability distribution and the optimized trusted log set as input, the root cause determination threshold is calibrated and the determination boundary is adjusted to generate an optimized defect root cause instruction sheet.

Images