CN-122022764-A - Property facility intelligent operation and maintenance management system and method based on deep learning

CN122022764ACN 122022764 ACN122022764 ACN 122022764ACN-122022764-A

Abstract

The invention discloses a deep learning-based intelligent operation and maintenance management system and method for property facilities, which belong to the technical field of operation and maintenance management of property facilities, wherein firstly, facilities with fault risks are identified and defined as potential fault points by analyzing operation data of the property facilities; generating a fault influence link map based on the physical connection path and functional association among facilities, positioning all the affected associated facilities according to the link map as check points, carrying out priority classification and parameter matching according to the conduction sequence, the facility level and the influence range, integrating the fault points and all the check points into a task group, matching maintenance detection personnel according to the link map, planning a continuous detection path, and finally packaging fault characteristics, influence links, classification strategies and execution schemes in the processing process into an operation and maintenance scene template, and realizing intelligent multiplexing and continuous improvement of the operation and maintenance strategies by comparing the characteristic similarity of the history template.

Inventors

MAO FEIFEI
PENG WEI
FENG YAN

Assignees

扬州新盛物业管理有限公司

Dates

Publication Date: 20260512
Application Date: 20260113

Claims (8)

1. The intelligent operation and maintenance management method for the property facility based on deep learning is characterized by comprising the following steps: Step 100, acquiring real-time and historical operation data of each property facility, identifying facilities with fault risks by analyzing the degree of deviation of the real-time data from the historical data, and defining the facilities as potential fault points; Step 200, positioning all affected associated facilities as checking points to be detected according to the fault influence link diagram, classifying the checking points according to the conduction sequence of the checking points in the link diagram, the preset level of the facilities and the fault influence range, and matching corresponding detection parameters and abnormality judgment thresholds for the checking points of different levels; Step 300, integrating potential fault points and all the check points into a task group, firstly matching and adapting operation and maintenance personnel to complete maintenance and dispatch of the fault points based on a fault influence link diagram, and then planning a continuous detection path of the check points and distributing personnel for check by combining the priority and position distribution of each check point; step 400, collecting maintenance and investigation data fed back on site, packaging fault characteristics, influence links, investigation point grading strategies and task execution schemes in the task group into operation and maintenance scene templates, storing the operation and maintenance scene templates in an operation and maintenance scene knowledge base, realizing quick matching and multiplexing of similar fault operation and maintenance strategies by comparing and analyzing the characteristic similarity of the new templates and the history templates, and creating new operation and maintenance scene templates in the operation and maintenance scene knowledge base for storage when the similarity is lower than a set threshold value.
2. The intelligent operation and maintenance management method of a property facility based on deep learning as set forth in claim 1, wherein the step S100 comprises: Step 110, acquiring and acquiring operation data from an industrial facility management platform and a sensor network deployed at each facility in real time, simultaneously, acquiring periodic operation records and maintenance logs of each facility from a historical database to form a data source together, comparing the real-time data with the statistical baseline, if the indexes continuously exceed the fluctuation range or the peak value threshold, the indexes are marked as potential fault points directly, and if the indexes do not exceed the fluctuation range or the peak value threshold, the indexes are marked as potential fault points directly, constructing an abnormal prediction model by adopting a long-short-term memory network, taking the periodic operation records and the maintenance logs of each facility in the historical database as training data sources, screening time sequence data sequences with normal operation marks, dividing the time sequence data sequences into a training set, a verification set and a test set in proportion, intercepting time sequence data fragments with fixed lengths by adopting a sliding time window method as model input samples, finally outputting the index prediction values for a period, and comparing the real-time data with the statistical baseline, and judging the fault points if the indexes continuously exceed the fluctuation range or the peak value threshold, the indexes are marked as potential fault points, and the fault trend, and the fault level is the fault level, if the deviation of the model prediction values and the fault level exceeds the threshold, the fault level is judged to be the fault level, and the fault level.
3. The intelligent operation and maintenance management method of a property facility based on deep learning as set forth in claim 2, wherein the step S100 further comprises: Step S120, based on engineering drawings, topology configuration libraries and historical operation and maintenance records of property facilities, extracting physical connection paths and logical function dependency relations among all facilities, setting the whole property facility group as a directed graph model G= (V, E), wherein a node set V represents each independent facility, an edge set E represents connection relations among facilities, and for any two facility nodes Va and Vb, the connection relations comprise direct connection and indirect connection, and specific judgment rules comprise: a direct connection, namely adding a directed edge E (Va- > Vb) into the diagram if the output of the facility Va is directly used as the input of the facility Vb or the normal operation of the Vb is a precondition of the normal operation of the Va; if the facilities Va and Vb depend on another facility Vc together or the correlation coefficient of the historical operation sequence is higher than a threshold value theta according to the historical data calculation, establishing bidirectional correlation or indirectly correlating through an intermediate node; And (3) taking the identified current fault point F as a starting node, applying a graph traversal algorithm on the directed graph model G, searching all facility nodes directly or indirectly affected by the F fault state along the directed edge direction, and recording all reachable paths from the F to each affected facility node to form a fault affected link graph L.
4. The intelligent operation and maintenance management method of a property facility based on deep learning as set forth in claim 3, wherein the step S200 comprises: Step S210, extracting all relevant nodes except a potential fault point F in the graph to form an affected relevant facility to form a to-be-detected check point set according to the fault influence link graph L, extracting a conduction distance of each check point in the fault influence link graph L, a preset facility importance level and a potential influence range of each check point, wherein the conduction distance is the shortest path length from the potential fault point F to the point, the preset facility importance level is a functional level based on setting of facilities in a property management platform, and the potential influence range is the number of downstream facilities with the check point as a starting point and associated in the fault influence link graph L; Step S220, dynamically sorting and grading the check point set according to the principle that the conduction distance is closer, the facility importance level is higher and the potential influence range is larger, wherein the priority is higher, the specific grading process comprises the steps of firstly sorting the check points from the near to the far according to the conduction distance, sorting the check points from the high to the low according to the facility importance level in the same group of the conduction distance, sorting the check points from the high to the low according to the potential influence range in the same level for three times, sorting the check points from the high to the low according to the potential influence range in the same level, sorting the check points into three high, medium and low priorities according to the final sorting result, and matching corresponding detection parameters and abnormality judgment thresholds for the check points with different priorities based on equipment technical manuals, preset operation parameters of the facility and a historical normal working condition database.
5. The intelligent operation and maintenance management method of a property facility based on deep learning as set forth in claim 1, wherein the step S300 comprises: Step S310, integrating the potential fault point F and all the check points into a predictive operation task group T= { F, P1, P2, & gt, pn }, wherein Pn is the nth check point; extracting and packaging a group of scheduling characteristics Tt= [ l, type, S, P and d ] for each node in a task group, wherein l represents the three-dimensional physical positions of potential fault points and check points in the task node, type represents task types for distinguishing whether the node is a maintenance task or a detection task, S represents specific professional skills required for processing the node, P represents scheduling priority, potential fault points F are set as the highest priority by default, d represents estimated processing time, historical type fault average maintenance time is taken for the potential fault points F, and average detection time of corresponding priority is taken for the check points Pn; The method comprises the steps of S320, simultaneously constructing a subsection characteristic vector V_wi at each operation and maintenance point which is each subsection W1, W2, and Wm, wherein i=1, 2, m is the total number of subsections, wm represents the mth operation and maintenance subsection, the subsection characteristic vector V_wi comprises three core attribute variables, S_wi is a skill type set possessed by operation and maintenance personnel of the ith subsection, A_wi is a responsible physical area of the ith subsection, N_wi is the number of schedulable personnel of the ith subsection, the characteristic vector T of a potential fault point F in a task group T is used as a matching standard, the matching condition is the three-dimensional physical coordinate E A_wi of the potential fault point F, if a plurality of subsections meeting the condition exist, the subsection with the largest schedulable personnel number is selected as a target subsection W_opt with the highest matching degree, the subsection meeting the matching condition and the target subsection with the highest matching degree is screened, and the task staff in the task group can be directly assigned from the task group W_opt according to the task type of the task group.
6. The intelligent operation and maintenance management method of a property facility based on deep learning as set forth in claim 5, wherein the step S300 further comprises: Step S330, after the maintenance personnel of the potential fault point F are assigned by the target subsection W_opt, screening personnel with corresponding skill qualification from the rest schedulable personnel according to the required specific expertise S defined in the corresponding task node of each check point to form a check personnel pool, when the distribution is carried out, the high-priority check points are preferentially matched with the rest detection personnel in the required expertise personnel and the medium-low-priority matching personnel pool, after the distribution of the check personnel is completed, the check points are grouped from near to far according to the conduction distance by combining the priority of each check point with the three-dimensional physical coordinates, and then all priority sub-paths are formed in series in the same group according to the principle of the near physical distance, finally, the continuous detection paths which cover all check points are integrated and distributed to the corresponding check personnel are constructed, and meanwhile, a shared operation and maintenance map is constructed based on the fault influence link map, the task node distribution, the personnel assignment result and the planning path, and the real-time position and the task progress of the check points are marked in real time.
7. The intelligent operation and maintenance management method of a property facility based on deep learning as set forth in claim 1, wherein the step S400 comprises: Step S410, collecting the whole flow core data of a task group T, including the fault characteristics of potential fault points F, the core topology information of a fault influence link diagram L, the check point grading strategy, the corresponding detection parameters and the task execution record, carrying out standardized arrangement on the collected data, unifying the data format and the coding rules to form a structured data set, packaging the structured operation and maintenance scene template based on the data set, wherein the template comprises the fault characteristics, the influence links, the grading strategy and the execution scheme, and converting the fault characteristics, the influence links, the grading strategy and the execution scheme into a unified numerical value type characteristic vector; Step S420, extracting feature vectors of historical templates in an operation and maintenance scene knowledge base, calculating similarity between the feature vectors of the new templates and each historical template vector by adopting a cosine similarity algorithm, presetting a similarity threshold, directly multiplexing operation and maintenance strategies corresponding to the historical templates if the historical templates with the similarity being more than or equal to the similarity threshold exist, supplementing execution effect data of the new scene to the historical templates, judging that the new operation and maintenance scene is a brand-new operation and maintenance scene if the similarity of all the historical templates is less than the similarity threshold, associating scene labels of the new templates with the operation and maintenance scene knowledge base, and perfecting the coverage range of the knowledge base.
8. The intelligent operation and maintenance management system for the property facilities based on deep learning is characterized by comprising a fault intelligent prediction module, an associated facility dynamic grading module, an operation and maintenance task intelligent scheduling module and an operation and maintenance scene management module; The intelligent fault prediction module acquires real-time and historical operation data of all property facilities, identifies facilities with fault risks by analyzing the degree of deviation of the real-time data from the historical data, and defines the facilities as potential fault points; the related facility dynamic grading module is used for positioning all related facilities to be used as checking points to be detected according to the fault influence link diagram, grading the checking points in priority according to the conduction sequence of each checking point in the link diagram, the preset grade of the facility and the fault influence range, and matching corresponding detection parameters and abnormality judgment thresholds for the checking points of different grades; The operation and maintenance task intelligent scheduling module integrates potential fault points and all the check points into a task group, and based on a fault influence link diagram, matching and adapting operation and maintenance personnel to complete maintenance and dispatch of the fault points, and planning a continuous detection path of the check points and distributing personnel for check by combining the priority and position distribution of the check points; The operation and maintenance scene management module collects maintenance and investigation data fed back on site, packages fault characteristics, influence links, investigation point grading strategies and task execution schemes in the task group into an operation and maintenance scene template, stores the operation and maintenance scene template into an operation and maintenance scene knowledge base, realizes quick matching and multiplexing of similar fault operation and maintenance strategies by comparing and analyzing characteristic similarity of the new template and the history template, and creates a new operation and maintenance scene template in the operation and maintenance scene knowledge base to store when the similarity is lower than a set threshold value.

Description

Property facility intelligent operation and maintenance management system and method based on deep learning Technical Field The invention relates to the technical field of property facility operation and maintenance management, in particular to a deep learning-based intelligent property facility operation and maintenance management system and method. Background With the promotion of the urban process, the facility scale of large-scale property scenes such as business complexes, industrial parks and the like is continuously expanded to cover various types of equipment such as power supply and distribution, vertical traffic, heating ventilation and air conditioning, water supply and drainage, security protection and the like, and currently, the field of property facility operation and maintenance generally adopts a separate equipment independent operation and maintenance mode, namely, dedicated monitoring and operation and maintenance systems are deployed for different types of equipment, and each system independently collects and stores data and executes local operation and maintenance decisions, so that a main operation and maintenance technology system is formed. However, the operation and maintenance technical system has many technical defects in practical application, namely the existing scheme is mostly dependent on passive feedback after faults occur, only can realize isolated fault identification of single equipment, is difficult to combine physical connection and function dependency relationship among facilities, and is difficult to predict possible diffusion paths of the faults, so that risks and ranges of chain reactions of related equipment can not be systematically evaluated, after the faults occur, operation and maintenance personnel usually determine an inspection sequence according to experience or rules, operation and maintenance resource scheduling and routing inspection paths depend on manual experience, personnel dispatch and routing inspection routing are often arranged based on an on-duty table or a general partition, the requirements and the depth of real-time fault scenes are not combined, response delay and resource mismatch are easily caused, and the fault maintenance and inspection efficiency of the whole property facility is affected. Disclosure of Invention The invention aims to provide a deep learning-based intelligent operation and maintenance management system and method for a property facility, which are used for solving the problems in the background technology. In order to solve the technical problems, the invention provides the technical scheme that the intelligent operation and maintenance management method for the property facility based on deep learning specifically comprises the following steps: Step 100, acquiring real-time and historical operation data of each property facility, identifying facilities with fault risks by analyzing the degree of deviation of the real-time data from the historical data, and defining the facilities as potential fault points; Step 200, positioning all affected associated facilities as checking points to be detected according to the fault influence link diagram, classifying the checking points according to the conduction sequence of the checking points in the link diagram, the preset level of the facilities and the fault influence range, and matching corresponding detection parameters and abnormality judgment thresholds for the checking points of different levels; Step 300, integrating potential fault points and all the check points into a task group, firstly matching and adapting operation and maintenance personnel to complete maintenance and dispatch of the fault points based on a fault influence link diagram, and then planning a continuous detection path of the check points and distributing personnel for check by combining the priority and position distribution of each check point; step 400, collecting maintenance and investigation data fed back on site, packaging fault characteristics, influence links, investigation point grading strategies and task execution schemes in the task group into operation and maintenance scene templates, storing the operation and maintenance scene templates in an operation and maintenance scene knowledge base, realizing quick matching and multiplexing of similar fault operation and maintenance strategies by comparing and analyzing the characteristic similarity of the new templates and the history templates, and creating new operation and maintenance scene templates in the operation and maintenance scene knowledge base for storage when the similarity is lower than a set threshold value. Further, step S100 includes: Step 110, acquiring and acquiring operation data from an industrial facility management platform and a sensor network deployed at each facility in real time, simultaneously, acquiring periodic operation records and maintenance logs of each facility from a historical database to jointly form a data