Centralized management method and management system for enterprise-level work orders of power supply system

Abstract

The invention discloses a centralized management method and management system for enterprise-level work orders of a power supply system, and relates to the technical field of operation and maintenance management of the power supply system, and the method comprises the steps: collecting and preprocessing work order source data in the power supply system; constructing a work order association graph based on the work order source data, calculating a work order association degree, and forming a work order group; performing time sequence characteristic analysis on the work order group, identifying a periodic fault mode in the work order group by using a sequence mode mining algorithm, and generating a characteristic mark chain; training a dual-channel convolutional recurrent neural network by using the feature mark chain, establishing a fault prediction engine, and outputting a work order priority score table; and allocating a maintenance task instruction based on the work order priority score table, executing a maintenance task, and returning maintenance completion information to the work order association graph to complete work order whole process tracking. According to the method, the quantitative analysis of the association relationship between the work orders is realized by constructing the work order association map and combining the physical distance coefficient and the power supply path overlap ratio.

Description

Technical Field

[0001] The present invention relates to the technical field of power supply system operation and maintenance management, and in particular to a power supply system enterprise-level work order centralized management method and management system. Background Art

[0002] At present, the work order management of power supply system mainly has the following problems: first, the work order allocation lacks systematic analysis, and is often manually dispatched based on experience, which cannot fully consider the correlation between work orders; second, there is a lack of effective identification and prediction mechanism for periodic faults, which leads to repeated occurrence of the same or similar faults; third, the work order priority assessment lacks scientific basis, making it difficult to achieve the optimal allocation of resources; finally, it is difficult to track the entire life cycle of the work order, and the maintenance feedback information cannot be effectively used. These problems have seriously affected the operation and maintenance efficiency and service quality of the power supply system. Summary of the invention

[0003] In view of the problems existing in the work order management of the existing power supply system, the present invention is proposed.

[0004] Therefore, the problem to be solved by the present invention is how to improve the scientificity and intelligence level of work order management of the power supply system, reduce the recurrence rate of faults, and optimize the efficiency of maintenance resource allocation.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: In the first aspect, an embodiment of the present invention provides a method for centralized management of enterprise-level work orders in a power supply system, which includes collecting and preprocessing work order source data in the power supply system, the work order source data including work order serial number, equipment fault code, processing level number, maintenance post code, and fault occurrence time; constructing a work order association map based on the work order source data, calculating the degree of work order association, and forming a work order group; performing time series feature analysis on the work order group, using a sequence pattern mining algorithm to identify periodic fault modes in the work order group, and generating a feature labeling chain; using the feature labeling chain to train a dual-channel convolutional recurrent neural network, establish a fault prediction engine, and output a work order priority scoring table; allocating maintenance task instructions based on the work order priority scoring table, executing maintenance tasks, and transmitting the maintenance completion information back to the work order association map to complete the full process tracking of the work order.

[0006] As a preferred solution of the enterprise-level centralized management method of work orders for power supply systems described in the present invention, the following steps are included: constructing a work order association map based on work order source data: extracting features from the work order source data to establish a work order feature vector; obtaining the spatial coordinates of the corresponding equipment and the power supply line number to which it belongs according to the equipment fault code in the work order feature vector, and constructing a work order location mapping table; calculating the physical distance coefficient between work orders according to the spatial coordinates in the work order location mapping table; calculating the power supply path overlap between work orders according to the power supply line number, and generating a line association coefficient; performing weighted calculation on the physical distance coefficient and the line association coefficient to obtain a work order association degree value; constructing a work order association matrix according to the work order association degree value, and setting a correlation threshold value; traversing the work order association matrix, marking the work order pairs exceeding the correlation threshold value as strongly associated, and grouping the strongly associated work orders using a hierarchical clustering algorithm to form multiple work order groups; assigning an identification number to each work order group, and storing the membership relationship of each work order group and the corresponding correlation degree value in the work order association map.

[0007] As a preferred solution of the enterprise-level centralized management method of work orders for power supply systems described in the present invention, the timing feature analysis of the work order group includes the following steps: establishing a work order timing sequence according to the fault occurrence time of each work order in the work order group, and dividing the work order timing sequence into multiple timing segments according to a preset time window; performing frequency statistics on the equipment fault codes in each timing segment to construct a fault code frequency distribution matrix; using a sequence pattern mining algorithm to process the fault code frequency distribution matrix to calculate the time interval characteristics between adjacent fault codes; analyzing the time interval characteristics to identify periodic fault modes in the work order group; associating the periodic fault modes with the processing level numbers to generate feature marker sequences, and integrating the feature marker sequences into feature marker chains.

[0008] As a preferred solution of the enterprise-level centralized management method of work orders for power supply systems described in the present invention, outputting a work order priority score table includes the following steps: performing second-order differential encoding on the feature tag chain to extract the changing trend characteristics of the periodic fault mode; constructing a dual-channel neural network model, with the first channel processing the feature tag chain and the second channel processing the changing trend characteristics; obtaining a fault prediction engine based on training the dual-channel neural network; using the fault prediction engine to analyze the newly added work order features, and outputting a fault probability prediction value and a reliability evaluation value at the same time; combining the fault probability prediction value and the reliability evaluation value to generate a work order priority score; arranging the work order priority scores in descending order to generate a work order priority score table.

[0009] As a preferred solution of the enterprise-level centralized management method of work orders for power supply systems described in the present invention, the allocation of maintenance task instructions based on the work order priority scoring table includes the following steps: constructing a set of work orders to be allocated based on the work order priority scoring table, the set of work orders to be allocated including work order priority scores, equipment fault codes and maintenance post codes; obtaining the corresponding dispatchable maintenance personnel identification numbers from the maintenance personnel database according to the maintenance post codes, and generating a maintenance personnel mapping table; counting the number of in-transit work orders corresponding to each maintenance personnel identification number in the maintenance personnel mapping table; calculating the remaining workload capacity of each maintenance personnel identification number based on a preset single-person maximum workload threshold; matching each work order with a maintenance personnel identification number that meets the remaining workload capacity based on the work order priority scoring table in turn, and generating a work order-maintenance personnel identification number association table; converting the association table into a maintenance task instruction format, and issuing the maintenance task instruction to the corresponding maintenance personnel terminal.

[0010] As a preferred solution of the enterprise-level centralized management method of work orders for the power supply system described in the present invention, wherein: transmitting the maintenance completion information back to the work order association map includes: receiving maintenance completion feedback information from the maintenance personnel terminal, the maintenance completion feedback information including the work order serial number, the actual fault handling method, the maintenance time and the equipment operating status; locating the corresponding work order group in the work order association map according to the work order serial number; updating the maintenance completion feedback information to the work order attribute field of the work order group; updating the work order association degree value in the work order association map based on the actual fault handling method and the equipment operating status; if the equipment operating status shows that the fault is not resolved, triggering a new work order generation process.

[0011] As a preferred solution of the enterprise-level centralized management method of work orders for power supply systems described in the present invention, the update and maintenance of the fault prediction engine includes the following steps: statistically analyzing the fault prediction accuracy based on the maintenance completion feedback information, and triggering the model retraining process when the fault prediction accuracy is lower than the preset control threshold; extracting the newly added feature marker chain from the work order association map as an incremental training sample; and based on the incremental training sample, performing online updates on the weight parameters in the dual-channel neural network model.

[0012] In the second aspect, an embodiment of the present invention provides an enterprise-level work order centralized management system for a power supply system, which includes a data acquisition and processing module for acquiring and preprocessing work order source data within the power supply system; a work order association analysis module for constructing a work order association map based on the work order source data, calculating the degree of work order association, and forming a work order group; a timing feature extraction module for performing timing feature analysis on the work order group, using a sequence pattern mining algorithm to identify periodic fault modes in the work order group, and generating a feature tag chain; a fault prediction modeling module for training a dual-channel convolutional recurrent neural network using a feature tag chain, establishing a fault prediction engine, and outputting a work order priority scoring table; a work order scheduling management module for allocating maintenance task instructions based on the work order priority scoring table, executing maintenance tasks, and transmitting maintenance completion information back to the work order association map to complete the full process tracking of the work order.

[0013] The beneficial effects of the present invention are as follows: the present invention realizes the quantitative analysis of the association relationship between work orders by constructing a work order association map and combining the physical distance coefficient and the power supply path overlap; innovatively designs a feature tag chain generation mechanism, uses a sequence pattern mining algorithm to identify periodic fault modes, and improves the accuracy of fault prediction; adopts a dual-channel neural network architecture to process the feature tag chain and change trend characteristics respectively, establishes a full-process tracking mechanism from prediction, allocation, execution to feedback, and provides a new solution for the intelligent operation and maintenance of power supply systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0015] Figure 1 This is a framework flow chart of the enterprise-level centralized management method for work orders in power supply systems.

[0016] Figure 2 Construct a flow chart for the work order association map of the enterprise-level work order centralized management method of the power supply system.

[0017] Figure 3 Construct a flow chart for the feature tag chain of the centralized management method of enterprise-level work orders in power supply systems. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0019] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0020] 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 term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0021] Example 1, reference Figure 1~Figure 3 , which is the first embodiment of the present invention, and provides a method for centralized management of enterprise-level work orders for a power supply system. The framework flow chart is as follows Figure 1 As shown, including, S1: Collect and pre-process work order source data within the power supply system.

[0022] Among them, the work order source data includes the work order serial number, equipment fault code, processing level number, maintenance position code, and fault occurrence time.

[0023] In the specific implementation process, the system first connects to various business subsystems of the power supply system, including the fault reporting subsystem, equipment management subsystem, and operation and maintenance management subsystem, by configuring a unified data collection interface to collect work order source data. Since the collected work order source data has problems such as inconsistent format, missing data, and duplication, it is necessary to standardize the work order source data, including data format unification, null value processing, duplicate data removal, and abnormal value correction.

[0024] S2: Build a work order association graph based on the work order source data, calculate the degree of work order association, and form a work order group.

[0025] Specifically, the work order association graph construction flow chart is as follows Figure 2 As shown, the following steps are included: S2.1: Extract features from the work order source data and establish a work order feature vector.

[0026] In the feature extraction process, the coding mapping technology is used to map the work order serial number into an increasing integer. The equipment fault code, processing level number, and maintenance post code are converted into a binary feature vector using one-hot encoding. All features are normalized to eliminate dimensional differences and form a unified work order feature vector.

[0027] S2.2: According to the equipment fault code in the work order feature vector, the spatial coordinates of the corresponding equipment and the power supply line number to which it belongs are obtained, and a work order location mapping table is constructed.

[0028] It should be noted that, in this embodiment, the corresponding spatial coordinates and the power supply line number are obtained from the equipment inventory database according to the equipment fault code to form a structured location mapping table.

[0029] S2.3: Calculate the physical distance coefficient between work orders based on the spatial coordinates in the work order location mapping table.

[0030] Specifically, the longitude and latitude coordinates of the devices corresponding to the two work orders are extracted, and the actual physical distance between the devices is calculated using the Euclidean distance formula; the actual physical distance value calculated is mapped to the interval [0,1] using the Gaussian kernel function to obtain the standardized physical distance coefficient. When the device locations corresponding to the two work orders are closer, their physical distance coefficient is closer to 1, indicating that the two work orders have a strong location correlation; conversely, when the device locations are far apart, the physical distance coefficient is closer to 0, indicating that the location correlation is weak.

[0031] S2.4: Calculate the power supply path overlap between work orders based on the power supply line number and generate the line correlation coefficient.

[0032] Specifically, the power supply path information of the equipment involved in the work order is obtained according to the power supply line number, including the trunk line and the branch line. For any pair of work orders, its complete power supply path is extracted, and the ratio of the length of the overlapping part of the path to the length of each path is calculated. The arithmetic mean of the two ratios is used as the line correlation coefficient. This coefficient reflects the degree of correlation between the two work orders in the topological structure of the power supply network. The higher the overlap of the power supply paths of the two work orders, the stronger their correlation in the power supply network.

[0033] S2.5: Perform weighted calculation on the physical distance coefficient and the line correlation coefficient to obtain a work order correlation degree value.

[0034] It should be noted that the weight coefficients are determined based on statistical analysis of historical work order processing data, and the sum of the weight coefficients is 1.

[0035] S2.6: Construct a work order association matrix based on the work order association degree values, and set the association degree threshold value.

[0036] Specifically, the construction process of the work order association matrix is ​​as follows: create an N×N symmetric matrix, where N is the total number of work orders to be processed; fill the work order association degree value into the corresponding position of the matrix to form the work order association matrix. Since the association between work orders is symmetrical, it is only necessary to calculate the element values ​​of the upper triangular matrix. In order to improve the calculation efficiency, a sparse matrix storage method is used to save only non-zero elements.

[0037] The determination of the correlation threshold value adopts an adaptive mechanism and is dynamically adjusted according to the density of work orders in the current time window. When the density of work orders is high, the threshold value is appropriately increased to avoid too many invalid associations; when the density of work orders is low, the threshold value is appropriately lowered to ensure that necessary work order associations are not missed. The specific adjustment strategy is: if no work order pairs exceeding the correlation threshold value are found after traversing the work order association matrix, the correlation threshold value is automatically adjusted. First, the correlation threshold value is gradually reduced to 80% of the original value; if a strongly associated work order pair is still not found, it is further reduced to 60% of the original value; if a strongly associated work order pair is still not found, the system preset minimum correlation threshold is used to ensure that at least one work order group can be formed.

[0038] S2.7: Traverse the work order association matrix, mark the work order pairs exceeding the association threshold as strongly associated, and use a hierarchical clustering algorithm to group the strongly associated work orders to form multiple work order groups.

[0039] Specifically, first traverse the work order association matrix, find all work order pairs whose association values ​​exceed the association threshold value, and mark them as strongly associated work order pairs. Then use the hierarchical clustering algorithm to group the work orders. Initially, each work order is regarded as an independent cluster. The distance between clusters is iteratively calculated, and the two clusters with the closest distance are merged each time. The distance between clusters is calculated using the complete link method, that is, the maximum distance between any work order pairs in the two clusters is taken as the distance between clusters. When there are no clusters that meet the merging conditions, the iterative process is terminated. If no work order pair that meets the association threshold value is found after traversal, the association threshold value is adjusted according to the strategy described in S2.6 until a strongly associated work order pair is found or the system minimum association threshold value is reached.

[0040] S2.8: Assign an identification number to each work order group, and store the membership relationship of each work order group and the corresponding association degree value into the work order association map.

[0041] Specifically, a unique identification number is assigned to the work order group obtained by clustering. A graph data structure is constructed to store the work order association information, where a node represents a single work order, an edge represents the association relationship between work orders, and an edge weight is the association degree value between work orders.

[0042] S3: Perform time series feature analysis on the work order group, use the sequence pattern mining algorithm to identify the periodic failure mode in the work order group, and generate a feature tag chain.

[0043] Specifically, the feature tag chain construction flow chart is as follows Figure 3 As shown, the following steps are included: S3.1: Establish a work order time sequence according to the fault occurrence time of each work order in the work order group, and divide the work order time sequence into multiple time sequence segments according to a preset time window.

[0044] Specifically, first extract the fault occurrence time of each work order in the work order group, and arrange them in ascending order according to the fault occurrence time to form a work order time series. Among them, the length of the preset time window is T, and the length of the time window is determined according to the fault characteristics and maintenance cycle; the overlap rate of adjacent time windows is R, and the sliding window method is used, starting from the start position of the sequence, each time sliding backward The time length is adjusted until the end of the sequence, and multiple time sequence segments are obtained.

[0045] S3.2: Perform frequency statistics on the equipment fault codes in each time sequence segment and construct a fault code frequency distribution matrix.

[0046] Specifically, an M×N dimensional fault code frequency distribution matrix is ​​established, where M is the number of time series segments, N is the total number of equipment fault code types, and the element values ​​in the matrix represent the frequency of occurrence of the corresponding equipment fault code in the time series segment. At the same time, the total number of work orders in each time series segment is recorded.

[0047] S3.3: Use the sequential pattern mining algorithm to process the fault code frequency distribution matrix and calculate the time interval characteristics between adjacent fault codes.

[0048] In one embodiment, the specific execution process of the sequence pattern mining algorithm includes: when traversing each time segment of the fault code frequency distribution matrix, when it is detected that the occurrence frequency of a certain device fault code in the current time segment is greater than zero, searching backward from the current time segment until the next occurrence position of the device fault code is found; calculating the time difference corresponding to the two occurrence positions, and storing the difference as a time interval feature in the time interval sequence of the corresponding device fault code.

[0049] Furthermore, a hash table structure is used to store the time interval sequence of each device fault code to achieve efficient query and self-balancing characteristics, wherein the key value is the device fault code and the corresponding value is the time interval sequence of the device fault code.

[0050] S3.4: Analyze time interval characteristics and identify periodic failure patterns in the work order group.

[0051] Specifically, the time interval sequence of each device fault code is analyzed. First, the average time interval of each device fault code is calculated based on the stored time interval sequence. and the standard deviation of the time interval ; Set the time interval threshold range to , when a device fault code appears in 3 or more consecutive time sequence segments, and the corresponding time interval values ​​all fall within When the device fault code meets the periodic condition, the average time interval is recorded as the period value.

[0052] S3.5: Associate the periodic failure mode with the processing level number, generate a signature sequence, and integrate the signature sequence into a signature chain.

[0053] Specifically, for the identified periodic fault mode, a triplet of feature tags is constructed, including the device fault code, the period value, and the processing level number. For each device fault code, the feature tags of all periodic faults of the device in the work order group are connected in sequence according to the fault occurrence time to form a feature tag chain for the device.

[0054] S4: Use the feature labeling chain to train a dual-channel convolutional recurrent neural network, build a fault prediction engine, and output a work order priority score table.

[0055] Specifically, the following steps are included: S4.1: Perform second-order differential encoding on the feature marker chain to extract the change trend characteristics of the periodic fault mode.

[0056] Specifically, a periodic value sequence is extracted from each group of feature marker chains for differential coding analysis; a first-order difference calculation is performed on the periodic value sequence to obtain a first-order difference sequence; a further difference calculation is performed on the first-order difference sequence to obtain a second-order difference sequence; the numerical characteristics of the second-order difference sequence are analyzed, and the coding rules and exception handling thresholds of the trend characteristics are set; the trend feature coding of the group of feature marker chains is determined according to the coding rules, and the trend feature coding is associated with the original feature marker chain as a whole and saved.

[0057] Among them, the first-order difference reflects the speed characteristics of the periodic change, and the second-order difference reflects the acceleration characteristics of the periodic change. And the coding rules are: when the second-order difference value is 0, it is coded as C0 to indicate a periodic stable trend; when the second-order difference value is a positive value, it is coded as C3 to indicate a periodic accelerated growth trend; when the second-order difference value is a negative value, it is coded as C4 to indicate a periodic decelerated downward trend; when the first-order difference value is a positive number, it is coded as C1 to indicate a periodic linear growth trend; when the first-order difference value is a negative constant, it is coded as C2 to indicate a periodic linear downward trend; when the second-order difference value alternates between positive and negative values, it is coded as C5 to indicate a periodic fluctuation trend.

[0058] S4.2: Construct a dual-channel neural network model, the first channel processes the feature labeling chain, and the second channel processes the change trend characteristics.

[0059] Specifically, the feature tag chain and its corresponding trend feature code are input into two channels respectively. The first channel inputs the original feature tag chain as the main feature, and the second channel inputs the trend feature code of each group of feature tag chains as the auxiliary feature. In the data preprocessing stage, the equipment fault code in the feature tag chain is one-hot encoded, the cycle value is normalized to the minimum-maximum, and the processing level is converted to an integer value and normalized. The trend feature codes C0~C5 are converted into 6-bit one-hot codes, that is, C0 is represented as [1,0,0,0,0,0], C1 is represented as [0,1,0,0,0,0], and so on, to ensure that the discrete characteristics of the trend feature are retained.

[0060] Furthermore, in the network architecture design, the first channel includes an input layer, an embedding layer with residual connections, a convolution layer, and a bidirectional LSTM layer in sequence, wherein the embedding layer is used for feature encoding and dimensionality reduction, the convolution layer is used for feature extraction, and the LSTM layer is used to process sequence information; the second channel includes a trend feature input layer and a multi-layer fully connected layer for feature mapping; wherein the dimension of the trend feature input layer is 6, corresponding to the unique hot representation of six trend feature codes; feature fusion is performed through the attention mechanism, the two channel features are merged, and feature interaction is performed through the fully connected layer, and finally the output layer uses the Sigmoid activation function to output the fault probability prediction value and the reliability evaluation value. The reliability evaluation value is obtained by calculating the entropy value of the model output feature vector. The lower the entropy value, the more certain the model prediction is, and the higher the reliability evaluation value is.

[0061] S4.3: Train the dual-channel neural network based on the loss function to obtain a fault prediction engine.

[0062] Firstly, the core framework of the fault prediction engine is constructed, which includes a data preprocessing module, a dual-channel neural network model, a prediction result processing module, a model update management module and a system interface module.

[0063] Before model training, the feature tag chain and its corresponding trend feature encoding were randomly divided into training set and validation set in a ratio of 8:2. The training process adopted a combined loss function, including cross entropy loss for fault prediction and mean square error loss for reliability assessment. The training parameters were set as follows: initial learning rate 0.001, cosine annealing learning rate scheduling, batch-size 64, 200 training rounds, and early stopping mechanism enabled until the validation set performance converged.

[0064] The update and maintenance process of the fault prediction engine after deployment is as follows: the fault prediction accuracy is counted based on the maintenance completion feedback information. When the fault prediction accuracy is lower than the preset control threshold, the model retraining process is triggered; the newly added feature tag chain is extracted from the work order association map as an incremental training sample; based on the incremental training samples, the weight parameters in the dual-channel neural network model are updated online; the model update version number and its corresponding prediction performance evaluation results are recorded to form a complete model iteration log.

[0065] The following optimization strategies are adopted during the incremental training process: set the sliding time window W to retain only the most recent samples; use differentiated weights for new and old samples (new sample weight 1.0, historical sample weight decays at 0.95); control the ratio of new and old samples in each mini-batch to 1:3; use a low learning rate of 0.0001 for 10 rounds of fine-tuning training; and use an early stopping strategy to terminate training when there is no improvement in the performance of the validation set for three consecutive rounds.

[0066] S4.4: Use the fault prediction engine to analyze the features of the newly added work order and output the fault probability prediction value and reliability assessment value at the same time.

[0067] S4.5: The failure probability prediction value and the reliability assessment value are combined and calculated to generate a work order priority score.

[0068] Specifically, the calculation formula for the work order priority score is as follows: ; Among them, G is the predicted value of failure probability, R is the reliability evaluation value, and is the weight coefficient.

[0069] S4.6: Arrange the work order priority scores in descending order to generate a work order priority score table.

[0070] Specifically, the work order priority score table contains the following fields: work order serial number, work order priority score, fault probability prediction value, reliability evaluation value, and recommended processing time limit. The recommended processing time limit is determined based on the work order priority score, including within 2 hours (greater than 0.8), within 4 hours (between 0.6 and 0.8), within 8 hours (between 0.4 and 0.6), and within 24 hours (less than 0.4).

[0071] S5: Allocate maintenance task instructions based on the work order priority score table, execute maintenance tasks, and send maintenance completion information back to the work order association map to complete the full process tracking of the work order.

[0072] Specifically, allocating maintenance task instructions based on the work order priority scoring table includes the following steps: constructing a set of work orders to be allocated based on the work order priority scoring table, the set of work orders to be allocated including work order priority scores, equipment fault codes and maintenance post codes; normalizing the work order priority scores, unifying different scoring dimensions into the [0,1] interval; obtaining the corresponding dispatchable maintenance personnel identification number from the maintenance personnel database according to the maintenance post code, and generating a maintenance personnel mapping table; and counting the number of in-transit work orders corresponding to each maintenance personnel identification number in the maintenance personnel mapping table.

[0073] Furthermore, the maximum workload threshold for a single person is determined according to the maintenance personnel's years of service, skill level and historical completion rate, wherein the weights of years of service, skill level and historical completion rate are 0.3, 0.4 and 0.3 respectively; wherein the years of service are assigned values ​​of 0.6, 0.8 and 1.0 for less than 5 years, 5 to 10 years and more than 10 years respectively, the skill level is assigned values ​​of 0.6, 0.8 and 1.0 for junior, intermediate and senior respectively, and the historical completion rate is taken as the average value of the most recent 3 months; based on the preset maximum workload threshold for a single person, the remaining workload capacity of each maintenance personnel identification number is calculated; when multiple maintenance personnel meet the remaining workload requirements at the same time, the maintenance personnel are prioritized according to the weighted scores of historical completion rate, order response time and geographical distance, wherein the weights of historical completion rate, response time and geographical distance are 0.4, 0.3 and 0.3 respectively, the response time is calculated in hours, the geographical distance is calculated in kilometers and the straight-line distance is calculated, and the maintenance personnel with the highest weighted score are given priority.

[0074] Furthermore, based on the work order priority scoring table, each work order is matched with a maintenance personnel identification number that meets the remaining workload capacity in turn, and a work order-maintenance personnel identification number association table is generated; when generating the association table, the work order allocation timestamp, estimated arrival time, and estimated completion time are also recorded; the association table is converted into a maintenance task instruction format, the maintenance task instruction contains key information such as the work order description, equipment location, required tools, safety precautions, etc., and the maintenance task instruction is sent to the corresponding maintenance personnel terminal.

[0075] Furthermore, transmitting the maintenance completion information back to the work order association map includes: receiving maintenance completion feedback information from the maintenance personnel terminal, the maintenance completion feedback information includes the work order serial number, actual fault handling method, maintenance time and equipment operating status; the maintenance completion feedback information also includes detailed records such as fault cause analysis, specific maintenance measures taken, replaced spare parts information and its coding, maintenance process pictures, etc.; locating the corresponding work order group in the work order association map according to the work order serial number; updating the maintenance completion feedback information to the work order attribute field of the work order group; calculating the deviation between the actual completion time and the expected completion time, and including the time deviation indicator in the historical performance evaluation of the maintenance personnel; updating the work order association degree value in the work order association map based on the actual fault handling method and the equipment operating status; if the equipment operating status shows that the fault is not resolved, the new work order generation process is triggered, and the newly generated work order is associated with the original work order. The association types include "fault is not resolved" and "need to upgrade".

[0076] Furthermore, the present embodiment also provides an enterprise-level work order centralized management system for a power supply system, including a data acquisition and processing module for acquiring and preprocessing work order source data within the power supply system; a work order association analysis module for constructing a work order association map based on the work order source data, calculating the degree of work order association, and forming a work order group; a timing feature extraction module for performing timing feature analysis on the work order group, using a sequence pattern mining algorithm to identify periodic fault modes in the work order group, and generating a feature tag chain; a fault prediction modeling module for training a dual-channel convolutional recurrent neural network using a feature tag chain, establishing a fault prediction engine, and outputting a work order priority scoring table; a work order scheduling management module for allocating maintenance task instructions based on the work order priority scoring table, executing maintenance tasks, and transmitting maintenance completion information back to the work order association map to complete full-process tracking of the work order.

[0077] In summary, the present invention realizes the quantitative analysis of the association relationship between work orders by constructing a work order association map and combining the physical distance coefficient and the overlap of the power supply path; innovatively designs the generation mechanism of the feature tag chain, and uses the sequence pattern mining algorithm to identify periodic fault modes, thereby improving the accuracy of fault prediction; adopts a dual-channel neural network architecture to process the feature tag chain and change trend characteristics respectively, and establishes a full-process tracking mechanism from prediction, allocation, execution to feedback, providing a new solution for the intelligent operation and maintenance of the power supply system.

[0078] Example 2, reference Figure 1~Figure 3 , which is the second embodiment of the present invention, and this embodiment provides a centralized management method for enterprise-level work orders in a power supply system. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0079] In order to verify the effectiveness of the enterprise-level centralized management method of work orders in the power supply system, the work order source data of a provincial power company from January to December 2023 were selected for experimental analysis. Some of the work order source data are shown in Table 1.

[0080] Table 1 Part of the work order source data Work order serial number Equipment fault code Processing level number Maintenance post code Failure time WO2023010001 F001 L2 M103 2023-01-01 08:15:23 WO2023010002 F003 L1 M101 2023-01-01 09:30:45 WO2023010003 F001 L2 M103 2023-01-01 10:20:12 WO2023010004 F002 L3 M102 2023-01-01 11:45:33 ...... ...... ...... ...... ...... The experimental environment uses a server with Intel Xeon Gold 6248R CPU, 256GB memory, and NVIDIA A100 GPU, and the operating system is Ubuntu 20.04 LTS. In the data preprocessing stage, the work order data collection frequency is set to 5 minutes, and more than 1 million original work order records are cleaned and standardized, and finally 892,456 valid work order data are obtained.

[0081] In the process of constructing the work order association map, the weights of the physical distance coefficient and the line association coefficient are set to 0.4 and 0.6 respectively, and the association threshold is initially set to 0.75. Through experiments, it is found that when the work order density exceeds 3 per square kilometer, raising the threshold to 0.85 can effectively avoid excessive association; when the work order density is less than 1 per square kilometer, lowering the threshold to 0.65 can ensure that necessary work order associations are not missed.

[0082] In the time series feature analysis phase, the preset time window length T is set to 7 days, and the adjacent window overlap rate R is set to 0.3. In the process of sequence pattern mining, the time interval threshold range is set to ±20%. In the experimental data set, a total of 2,847 periodic fault modes were identified, among which the periodic faults of transformer equipment accounted for the highest proportion, reaching 38.5%.

[0083] The dual-channel neural network model training uses 714,965 training samples and 177,491 validation samples. The fault prediction accuracy of the model on the validation set reached 87.3%, and the mean absolute error of reliability assessment was 0.086. In the work order priority score calculation, the weight α of the fault probability prediction value was set to 0.7, and the weight β of the reliability assessment value was set to 0.3. The performance comparison between the method of the present invention and the prior art is shown in Table 2.

[0084] Table 2 Performance comparison between the method of the present invention and the prior art Evaluation Metrics Method of the present invention Traditional rule-based approach Single Neural Network Fault prediction accuracy 87.3% 65.8% 78.4% Average response time 2.5h 6.8h 4.2h Improved work order processing efficiency 45.6% - 25.3% Maintenance personnel utilization rate 92.4% 75.2% 83.7% System stability 99.2% 85.6% 92.8% From the comparative data in Table 2, it can be seen that the method of the present invention is significantly superior to the traditional rule-based method and the single neural network method in terms of key indicators such as fault prediction accuracy, average response time, work order processing efficiency, maintenance personnel utilization rate and system stability. Among them, the fault prediction accuracy rate increased by 21.5 percentage points, the average response time was shortened by 63.2%, and the maintenance personnel utilization rate increased by 17.2 percentage points, which fully demonstrated the advanced nature and practical value of the method of the present invention.

[0085] During the actual deployment phase, the system ran continuously for three months, processing a total of 215,647 work orders, reducing the average work order response time from 6.8 hours to 2.5 hours, and increasing the average number of work orders completed by maintenance personnel from 5.2 to 7.6 per day. Through the analysis of the feedback from the completion of maintenance, the accuracy of the system's prediction of fault types reached 87.3%, and the consistency between the work order priority score and the actual urgency reached 85.6%. In addition, the system successfully predicted and prevented 23 major equipment failures, avoiding huge economic losses.

[0086] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A centralized management method for enterprise-level work orders in a power supply system, characterized by: include, Collect and pre-process the work order source data in the power supply system, the work order source data includes the work order serial number, equipment fault code, processing level number, maintenance post code, and fault occurrence time; Building a work order association graph based on the work order source data, calculating the work order association degree, and forming a work order group; Performing time series feature analysis on the work order group, identifying periodic fault patterns in the work order group using a sequence pattern mining algorithm, and generating a feature tag chain; The feature tag chain is used to train a dual-channel convolutional recurrent neural network, establish a fault prediction engine, and output a work order priority score table; Based on the work order priority score table, maintenance task instructions are assigned, maintenance tasks are executed, and maintenance completion information is fed back to the work order association map to complete the full process tracking of the work order.

2. The enterprise-level centralized management method for power supply system work orders according to claim 1, characterized in that: Constructing a work order association graph based on the work order source data includes the following steps: Extracting features from the work order source data to establish a work order feature vector; According to the equipment fault code in the work order feature vector, the spatial coordinates of the corresponding equipment and the power supply line number to which it belongs are obtained, and a work order location mapping table is constructed; Calculate the physical distance coefficient between work orders according to the spatial coordinates in the work order location mapping table; According to the power supply line number, the power supply path overlap between the work orders is calculated to generate a line correlation coefficient; Performing weighted calculation on the physical distance coefficient and the line association coefficient to obtain a work order association degree value; Constructing a work order association matrix according to the work order association degree values, and setting an association degree threshold value; Traversing the work order association matrix, marking work order pairs exceeding the association threshold value as strongly associated, and grouping the strongly associated work orders using a hierarchical clustering algorithm to form a plurality of work order groups; An identification number is assigned to each work order group, and the membership relationship of each work order group and the corresponding association degree value are stored in the work order association map.

3. The enterprise-level centralized management method for power supply system work orders according to claim 1, characterized in that: Performing time series feature analysis on the work order group includes the following steps: Establishing a work order time sequence according to the fault occurrence time of each work order in the work order group, and dividing the work order time sequence into multiple time sequence segments according to a preset time window; Perform frequency statistics on the equipment fault codes in each time series segment and construct a fault code frequency distribution matrix; Processing the fault code frequency distribution matrix using a sequence pattern mining algorithm to calculate the time interval characteristics between adjacent fault codes; Analyzing the time interval characteristics to identify periodic failure modes in the work order group; The periodic fault mode is associated with the processing level number to generate a signature sequence, and the signature sequence is integrated into a signature chain.

4. The enterprise-level centralized management method for power supply system work orders according to claim 1, characterized in that: The output work order priority scoring table comprises the following steps: Performing second-order differential encoding on the characteristic marker chain to extract the change trend characteristics of the periodic fault mode; Construct a dual-channel neural network model, where the first channel processes the feature labeling chain and the second channel processes the change trend characteristics; Training the dual-channel neural network based on the loss function to obtain a fault prediction engine; Utilize the fault prediction engine to analyze the features of the newly added work order and output the fault probability prediction value and reliability assessment value at the same time; The failure probability prediction value and the reliability evaluation value are combined and calculated to generate a work order priority score; Arrange the work order priority scores in descending order to generate a work order priority score table.

5. The enterprise-level centralized management method for power supply system work orders according to claim 1, characterized in that: Allocating maintenance task instructions based on the work order priority scoring table includes the following steps: Constructing a set of work orders to be assigned based on the work order priority score table, wherein the set of work orders to be assigned includes work order priority scores, equipment fault codes, and maintenance position codes; According to the maintenance post code, the corresponding identification number of the dispatchable maintenance personnel is obtained from the maintenance personnel database to generate a maintenance personnel mapping table; Counting the number of in-transit work orders corresponding to each maintenance personnel identification number in the maintenance personnel mapping table; Based on the preset single-person maximum workload threshold, calculate the remaining workload capacity of each maintenance personnel identification number; Based on the work order priority scoring table, matching each work order with a maintenance personnel identification number that satisfies the remaining workload capacity in turn, and generating a work order-maintenance personnel identification number association table; The association table is converted into a maintenance task instruction format, and the maintenance task instruction is sent to the corresponding maintenance personnel terminal.

6. The enterprise-level centralized management method for power supply system work orders according to claim 1, characterized in that: The method of returning the maintenance completion information to the work order association graph includes: Receive maintenance completion feedback information from the maintenance personnel terminal, wherein the maintenance completion feedback information includes the work order serial number, actual fault handling method, maintenance time and equipment operation status; Locate the corresponding work order group in the work order association map according to the work order serial number; Updating the maintenance completion feedback information to the work order attribute field of the work order group; Based on the actual fault handling method and the equipment operation status, updating the work order association degree value in the work order association map; If the equipment operation status shows that the fault is not resolved, a new work order generation process is triggered.

7. The enterprise-level centralized management method for power supply system work orders according to claim 4, characterized in that: The updating and maintenance of the fault prediction engine includes the following steps: The fault prediction accuracy is calculated based on the maintenance completion feedback information. When the fault prediction accuracy is lower than the preset control threshold, the model retraining process is triggered; Extract new feature tag chains from the work order association graph as incremental training samples; Based on the incremental training samples, the weight parameters in the dual-channel neural network model are updated online.

8. A power supply system enterprise-level work order centralized management system, based on the power supply system enterprise-level work order centralized management method according to any one of claims 1 to 7, characterized in that: Also includes, Data acquisition and processing module, used to collect and pre-process work order source data in the power supply system; A work order association analysis module, used to construct a work order association map based on the work order source data, calculate the degree of association of the work orders, and form a work order group; A time series feature extraction module is used to perform time series feature analysis on the work order group, identify periodic fault patterns in the work order group using a sequence pattern mining algorithm, and generate a feature tag chain; A fault prediction modeling module is used to train a dual-channel convolutional recurrent neural network using the feature tag chain, establish a fault prediction engine, and output a work order priority scoring table; The work order scheduling management module is used to assign maintenance task instructions based on the work order priority score table, execute maintenance tasks, and return maintenance completion information to the work order association map to complete the full process tracking of the work order.

Images