Digital-twin-based operation and maintenance service quality intelligent evaluation method and system

Abstract

This invention discloses a method and system for intelligent evaluation of operation and maintenance service quality based on digital twins, specifically relating to the field of operation and maintenance service quality evaluation technology. The method receives work order instructions, locates virtual nodes and associated nodes in the digital twin space, and extracts initial fault impact features by fusing interface latency, business backlog, and operational load sequences. It analyzes the work order time, location trajectory set, and service recovery timestamp, matching spatiotemporal deviations to construct maintenance execution deviation features. It extracts the business dissipation rate, maps it to generate business recovery lag features, and cross-validates these features with the maintenance execution deviation features to generate a single work order quality score. The quality score is then used to calculate single-time performance data with billing parameters, aggregates periodic data execution trend projections, and outputs an evaluation result including performance rating and cost adjustment ratio. The system described above is used to implement the above method.

Description

Technical Field

[0001] This invention relates to the field of operation and maintenance service quality assessment technology, specifically to an intelligent assessment method and system for operation and maintenance service quality based on digital twins. Background Technology

[0002] As medical institutions and large enterprises deepen their digital transformation, daily operations management involves not only the operation of core business systems but also a large amount of equipment maintenance, information system operation support, and logistical services. These tasks are wide-ranging, involve numerous collaborating departments and service providers, and have a high volume of concurrent daily transactions. They typically require the integration of internal and external business flow data through an information-based monitoring platform to drive a shift in management models from experience-driven to data-driven, refined approaches.

[0003] In actual operation supervision and service evaluation, the arrival time and fault resolution time reported by service personnel in the work order system are highly susceptible to human intervention. This can lead to situations where personnel clock in early or falsely report completed repairs, resulting in superficially meeting performance indicators for a single work order while underlying system interface congestion and service queuing persist. Furthermore, existing evaluation mechanisms typically use fixed, static performance indicators to directly score and discount service results, ignoring the differences in the initial damage caused by different node failures to the overall network topology and service load. In addition, periodic service provider performance accounting largely relies on managers' experience-based judgment of superficial business reports, leading to highly subjective ratings. This results in a severe disconnect between the issuance of maintenance renewal fees and the actual effectiveness of service guarantees, causing ineffective investment of maintenance funds and making it difficult to guarantee the overall stability of the business system. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method and system for intelligent evaluation of operation and maintenance service quality based on digital twins to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent assessment method for operation and maintenance service quality based on digital twins, comprising the following steps: S1, receiving a work order instruction, locating the virtual node and topology-related nodes of the faulty device in the digital twin space, extracting the interface latency and service backlog of the topology-related nodes before the trigger, collecting the operating load sequence of the virtual node, and mapping the interface latency and service backlog to the operating load sequence to extract the initial impact features of the fault; S2, parsing the arrival time and troubleshooting time reported in the work order, extracting the positioning trajectory set of the maintenance terminal and the service recovery timestamp of the virtual node, spatially matching the arrival time with the positioning trajectory set to calculate the position offset, and matching the troubleshooting time with... S3. Calculate the repair lag by aligning the service recovery timestamp with the repair lag, and construct the maintenance execution deviation feature by combining the location offset with the repair lag; S4. Extract the business dissipation rate of the topology-related nodes after the service recovery timestamp, perform correlation mapping between the business dissipation rate and the initial impact feature of the fault to extract the business recovery lag feature, cross-validate the maintenance execution deviation feature with the business recovery lag feature, and generate a single work order quality score; S5. Read the billing parameters from the contract database, cross-calculate the single work order quality score with the billing parameters to generate single performance data, aggregate the single performance data within the cycle to perform trend extrapolation, and output the operation and maintenance service quality assessment result including performance rating and cost adjustment ratio.

[0006] Preferably, the specific process of receiving a work order instruction, locating the virtual node and topology-related node of the faulty device in the digital twin space, and extracting the interface latency and service backlog of the topology-related node before triggering is as follows: Parse the device identification code contained in the work order instruction, retrieve the three-dimensional spatial coordinates bound to the device identification code in the digital twin space to establish the virtual node; traverse the system operation topology diagram associated with the virtual node, extract the service nodes connected to the upstream and downstream of the service data flow to establish the topology-related node; extract the pre-observation time window based on the triggering time point of the work order instruction, capture the network packet round-trip time sequence of the topology-related node within the pre-observation time window to establish the interface latency, and aggregate the total number of queued tasks in the underlying database of the topology-related node to establish the service backlog.

[0007] Preferably, the specific process of collecting the virtual node's operating load sequence and mapping interface latency and service backlog to the operating load sequence to extract the initial impact features of the fault is as follows: Collect the dynamic curve of physical resource consumption of the virtual node within the pre-observation time window to generate the operating load sequence; extract the time-series sampling points of the operating load sequence, align the interface latency and service backlog according to the execution timestamps of the time-series sampling points to generate a multi-dimensional service state cross-section; project the multi-dimensional service state cross-section onto the multi-dimensional tensor space where the operating load sequence is located, and extract the feature co-occurrence distribution density of the multi-dimensional service state cross-section and the operating load sequence in the multi-dimensional tensor space to generate the initial impact features of the fault.

[0008] Preferably, the specific process of parsing the arrival time and exclusion time in the work order, extracting the positioning trajectory set of the maintenance terminal and the service recovery timestamp of the virtual node, and spatially matching the arrival time with the positioning trajectory set to calculate the position offset is as follows: Parse the work order text to extract the arrival time and exclusion time; retrieve the maintenance terminal movement coordinate sequence collected by the indoor positioning network node to generate the positioning trajectory set; extract the log record time point in the system operation log when the core service process status switches to normal response to establish the service recovery timestamp; traverse the positioning trajectory set and extract the terminal spatial coordinates that have a temporal mapping relationship with the arrival time; perform spatial geometric mapping comparison between the terminal spatial coordinates and the three-dimensional spatial coordinates of the virtual node, and extract the spatial physical distance divergence as the position offset.

[0009] Preferably, the specific process of aligning the exclusion time with the service recovery timestamp to calculate the repair lag, and constructing the maintenance execution deviation feature by combining the location offset with the repair lag, is as follows: Map the exclusion time and service recovery timestamp to a unified reference time axis, and extract the absolute time span of the exclusion time and service recovery timestamp on the unified reference time axis as the repair lag; construct a two-dimensional feature matrix containing spatial and temporal dimensions, fill the spatial dimension of the two-dimensional feature matrix with the location offset, fill the temporal dimension of the two-dimensional feature matrix with the repair lag, and perform feature flattening and dimensionality reduction splicing of the two-dimensional feature matrix to generate the maintenance execution deviation feature.

[0010] Preferably, the specific process of extracting the service dissipation rate of the topology-related nodes after the service recovery timestamp and performing correlation mapping between the service dissipation rate and the initial impact features of the fault to extract the service recovery hysteresis features is as follows: Set a feature observation time window after the service recovery timestamp, and collect the service queue backlog decay sequence of the topology-related nodes within the feature observation time window; extract the first-order difference features of the service queue backlog decay sequence at discrete time points to form the service dissipation rate; construct a multi-dimensional correlation mapping space, map the service dissipation rate to the dynamic response dimension in the multi-dimensional correlation mapping space, map the initial impact features of the fault to the initial state dimension in the multi-dimensional correlation mapping space, extract the decay deviation of the dynamic response dimension relative to the initial state dimension, and generate the service recovery hysteresis features.

[0011] Preferably, the specific process of cross-validating maintenance execution deviation features and business recovery hysteresis features to generate a single work order quality score is as follows: The maintenance execution deviation features are converted into a deviation polarity matrix, and the business recovery hysteresis features are converted into a hysteresis weight matrix; a matrix inner product transformation is performed on the deviation polarity matrix and the hysteresis weight matrix to extract a joint negative feature map; a preset full-score benchmark feature scalar is extracted, and the full-score benchmark feature scalar and the joint negative feature map are input into a nonlinear weighted dimensionality reduction network to extract the dimensionality reduction output scalar value of the full-score benchmark feature scalar under the feature constraints of the joint negative feature map, thereby generating a single work order quality score.

[0012] Preferably, the specific process of reading the billing parameters from the contract database and cross-calculating the single work order quality score with the billing parameters to generate single performance data is as follows: Parse the operation and maintenance service agreement data entity in the contract database, extract the single maintenance basic pricing vector and service level threshold parameters to generate billing parameters; perform boundary limit comparison between the single work order quality score and the service level threshold parameters to extract the quality assessment deviation; perform multi-dimensional spatial mapping calculation between the quality assessment deviation and the single maintenance basic pricing vector to generate single performance data including resource reduction coefficients and quality anomaly classification labels.

[0013] Preferably, the specific process of performing trend extrapolation on single-time performance data within the aggregation period and outputting an operation and maintenance service quality assessment result including performance rating and cost adjustment ratio is as follows: Extract all single-time performance data arranged in a timestamp sequence within the accounting period; extract the resource reduction coefficient and quality anomaly classification label from the single-time performance data to construct a performance time-series feature matrix; extract the feature decay slope of the performance time-series feature matrix on the time axis of the accounting period; perform distribution similarity matching between the feature decay slope and the standard rating feature library to establish the performance rating; aggregate and extract the mean distribution feature of the resource reduction coefficient within the performance time-series feature matrix; map the mean distribution feature to the billing compensation rule matrix to extract the cost adjustment ratio; combine the performance rating and cost adjustment ratio to generate the operation and maintenance service quality assessment result.

[0014] The intelligent assessment system for operation and maintenance service quality based on digital twins is used to execute the aforementioned intelligent assessment method for operation and maintenance service quality based on digital twins. It includes: a status awareness module, used to receive work order instructions, locate the virtual node and topology-related nodes of the faulty device in the digital twin space, extract the interface latency and service backlog of the topology-related nodes before triggering, collect the operating load sequence of the virtual node, and map the interface latency and service backlog to the operating load sequence to extract the initial impact features of the fault; and a deviation comparison module, used to parse the arrival time and resolution time reported in the work order, extract the positioning trajectory set of the maintenance terminal and the service recovery timestamp of the virtual node, spatially match the arrival time with the positioning trajectory set to calculate the position offset, and compare the resolution time with the location offset. The system calculates repair lag by aligning with the service recovery timestamp sequence, and constructs maintenance execution deviation features by combining location offset with repair lag. The quality scoring module extracts the business dissipation rate of topology-related nodes after the service recovery timestamp, performs correlation mapping between the business dissipation rate and initial fault impact features to extract business recovery lag features, cross-validates maintenance execution deviation features with business recovery lag features, and generates a single work order quality score. The performance accounting module reads billing parameters from the contract database, cross-calculates the single work order quality score with billing parameters to generate single performance performance data, aggregates single performance performance data within a cycle to perform trend extrapolation, and outputs an operation and maintenance service quality assessment result including performance rating and cost adjustment ratio.

[0015] The technical effects and advantages of this invention are as follows: (1) In the technical solution described in this invention, when receiving a work order instruction, the virtual node and topology associated node of the faulty equipment in the digital twin space are located, and the interface delay and business backlog are mapped to the running load sequence to extract the initial impact features of the fault; the work order filling time is analyzed, the maintenance terminal positioning trajectory set and the service recovery timestamp of the virtual node are extracted, the arrival time is matched with the spatial matching of the positioning trajectory set to calculate the position offset, and the troubleshooting time is aligned with the service recovery timestamp to calculate the repair lag, and the maintenance execution deviation features are constructed. Thus, the problem of service personnel falsely reporting working hours and cheating on location in the prior art is solved, the physical authenticity of the troubleshooting execution process is restored from the underlying objective data level, and the fault of a single node is objectively bound to the degree of global business obstruction; (2) In the technical solution described in this invention, after the system service is restored, the business dissipation rate of the topology-related nodes is tracked, and the business recovery lag feature is extracted by performing correlation mapping with the initial impact features of the fault. The maintenance execution deviation feature and the business recovery lag feature are cross-validated to generate a single work order quality score. The billing parameters in the contract library are read, and the single work order quality score and billing parameters are cross-calculated to generate single performance performance data. The data execution trend is aggregated within the cycle, and the results including the performance rating and the cost adjustment ratio are output. Thus, the problems of the existing evaluation mechanism using fixed indicators for scoring, which leads to the disconnect between the evaluation results and the actual business, and the problem of the issuance amount not matching the actual guarantee effect caused by manual experience-based rating, are solved. The objective accounting of the performance cost of periodic outsourced service providers is realized from the cross-validation of single troubleshooting efficacy to the objective accounting of the performance cost of periodic outsourced service providers is achieved.

[0016] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0017] Figure 1 This is a flowchart of the intelligent evaluation method for operation and maintenance service quality based on digital twins according to the present invention. Figure 2 This is a schematic diagram illustrating the attenuation characteristics of the service queue backlog in an embodiment of the present invention. Figure 3 This is a schematic diagram illustrating the trend of performance time sequence characteristics in an embodiment of the present invention; Figure 4 This is a flowchart of the intelligent evaluation system for operation and maintenance service quality based on digital twins, as described in this invention. Detailed Implementation

[0018] This application's embodiments address the problems of traditional operation and maintenance service quality evaluation being easily interfered with by false manual reporting and the high subjectivity of assessment and rating leading to uncontrolled performance costs through a digital twin-based intelligent evaluation method and system for operation and maintenance services.

[0019] The overall approach of the solution in this application is as follows: First, the initial impact features of the fault are extracted by integrating the operating load of virtual nodes and the blocked status of related services in the digital twin space. Then, the time stamps of the terminal's physical trajectory and the recovery of the system's core services are compared with the manually reported time to quantify the spatial and temporal deviations in the maintenance process. Subsequently, the dissipation of service queues after fault recovery is tracked, and an objective quality score for each work order is generated by combining the initial impact of the fault with the execution deviation. Finally, the single work order score and contract billing parameters are used for calculation and cycle extrapolation to output the service provider's performance rating and subsequent fee adjustment ratio.

[0020] Example 1; please refer to Figure 1This invention provides a technical solution: an intelligent evaluation method for operation and maintenance service quality based on digital twins, comprising the following steps: S1, receiving a work order instruction, locating the virtual node and topology-related nodes of the faulty device in the digital twin space, extracting the interface latency and service backlog of the topology-related nodes before the trigger, collecting the operating load sequence of the virtual node, and mapping the interface latency and service backlog to the operating load sequence to extract the initial impact features of the fault; S2, parsing the arrival time and troubleshooting time reported in the work order, extracting the positioning trajectory set of the maintenance terminal and the service recovery timestamp of the virtual node, spatially matching the arrival time with the positioning trajectory set to calculate the position offset, and matching the troubleshooting time with the service recovery timestamp of the virtual node. S3. Calculate the repair lag by aligning the recovery timestamp with the repair lag, and construct the maintenance execution deviation feature by combining the location offset with the repair lag; S4. Extract the service dissipation rate of the topology-related nodes after the service recovery timestamp, perform association mapping between the service dissipation rate and the initial impact feature of the fault to extract the service recovery lag feature, cross-validate the maintenance execution deviation feature with the service recovery lag feature, and generate a single work order quality score; S5. Read the billing parameters from the contract database, cross-calculate the single work order quality score with the billing parameters to generate single performance data, aggregate the single performance data within the cycle to perform trend extrapolation, and output the operation and maintenance service quality assessment result including performance rating and cost adjustment ratio.

[0021] In this implementation scheme, after receiving the work order instruction, step S1 establishes a virtual node of the faulty device and topology-related nodes with data flow associations in the digital twin space. It obtains the interface latency parameters of the topology nodes in network data interaction and the backlog of business tasks at the underlying database level before the work order is triggered. It also collects the dynamic operating load sequence formed by the processor and memory resource consumption of the virtual node itself. Subsequently, the interface latency and business backlog status are matched and aligned to the operating load curve through time dimension matching to form a feature mapping, extracting the initial impact features of the fault reflecting the degree of system obstruction. Here, the digital twin space refers to a three-dimensional digital mapping environment built on the basis of physical space, capable of reflecting the spatial coordinates and operating topology of the physical device in real time; interface latency specifically refers to the time consumed during network packet round-trip communication between business systems. This step ensures that a single device failure is no longer an isolated event, but is placed under the pressure of global business flow for objective state preservation, providing an initial basis for subsequently measuring the urgency of troubleshooting.

[0022] Step S2 parses the manually entered arrival time and troubleshooting time from the work order records submitted by maintenance personnel, and retrieves the mobile positioning trajectory set of the maintenance terminal generated by the indoor positioning base station network. Simultaneously, it extracts the service recovery timestamp from the system operation log to show the core service process returning to normal response status after a crash. By spatially comparing the arrival time with the physical coordinates of the corresponding time point in the positioning trajectory set, the location offset is calculated. The manually reported troubleshooting time is compared with the underlying objective recovery timestamp on the same timeline to calculate the repair lag. Finally, the location offset and repair lag are combined to form the maintenance execution deviation characteristics. The service recovery timestamp refers to the precise log system record time of the core daemon process of the business support system resuming normal port listening and request / response operations. This step eliminates false information in manually reported reports and, using the terminal's physical coordinates and system-level time logs, intuitively reflects whether the maintenance process was performed on-site and whether there are delays where superficial repairs are performed but internal services have not been restarted.

[0023] Step S3 extracts a feature observation window after the service recovery timestamp, monitors the dissipation of accumulated tasks within the topology-related nodes to obtain the business dissipation rate, and correlates this dissipation rate with the previously extracted initial fault impact features to determine whether the efficiency of clearing backlogged business matches the impact range before the fault, thereby extracting business recovery lag features. Then, it performs weighted calculations or cross-validation of maintenance execution deviation features (recording human error) and business recovery lag features (reflecting objective business smoothness) to output a single work order quality score. The business dissipation rate characterizes the processing speed and queue decay slope of suspended transactions in the historical queue after the system service process recovers. This step achieves in-depth verification of the effectiveness of troubleshooting actions, identifies superficial fixes that do not address the root cause, and ensures that the generated service quality score is entirely derived from objective system-level smoothness indicators.

[0024] Step S4 retrieves billing parameters such as the basic rate and service rating threshold signed with the external service provider from the system storage module. It then compares and maps the single-work order quality score output in Step S3 with these billing parameter execution limits to generate single-work order performance data containing resource reduction rules. Finally, it extracts the decay slope and performs similarity deduction on all single-work order performance data distributed over the entire accounting cycle according to time series, outputting a performance rating reflecting the service provider's overall performance and the corresponding cost adjustment ratio. The cost adjustment ratio refers to the percentage reduction in funds or penalty conversion parameters triggered by service level achievement and quality defects, based on the basic maintenance pricing. This step establishes a connection between the underlying operational status of the business system and the upper-level contract management, transforming the objective quality and efficiency of each fault maintenance operation into a financial adjustment basis for calculating the supplier's annual or quarterly fees. Specifically, the process of receiving work order instructions, locating the virtual node and topology-related nodes of the faulty device in the digital twin space, and extracting the interface latency and service backlog of the topology-related nodes before the trigger is as follows: Parse the device identification code contained in the work order instruction, retrieve the three-dimensional spatial coordinates bound to the device identification code in the digital twin space to establish the virtual node; traverse the system operation topology diagram associated with the virtual node, extract the service nodes connected to the upstream and downstream of the service data flow to establish the topology-related nodes; extract the pre-observation time window based on the trigger time point of the work order instruction, capture the network packet round-trip time sequence of the topology-related nodes within the pre-observation time window to establish the interface latency, and aggregate the total number of queued tasks in the underlying database of the topology-related nodes to establish the service backlog.

[0025] In this implementation scheme, after receiving a work order instruction, the system extracts the device's physical address or network element logical number as the device identifier, inputs it into a pre-constructed 3D building information and network distribution model, and obtains the corresponding 3D spatial coordinates of the device through primary key mapping query, marking these coordinates as virtual nodes. The system reads the directed acyclic graph of device data interaction, uses the virtual node as the central vertex, and searches for directly connected vertices with both out-degree and in-degree greater than zero in the data packets through the adjacency matrix, establishing them as topology-related nodes. The system extracts the precise timestamp generated by the work order instruction, pushes back to the past according to the preset historical operation log backtracking step size, and defines a pre-observation time window. Within this window, network data packets used for handshake communication between topology-related nodes are captured through packet sniffing, and the interface latency is calculated based on the time difference between sending and receiving. Simultaneously, the amount of pending transactions is counted in the database cache pool where the topology-related nodes reside. The specific calculation formula for interface latency is expressed as follows: ; This represents the interface latency of the m-th topologically associated node; N represents the total number of network packets captured within the preceding observation time window; and n represents the sequence number of the data packet. Indicates the time when the request for the nth network data packet was sent; This represents the arrival time of the response for the nth network data packet. The formula for calculating the service backlog is: ; This represents the backlog of business at the m-th topological node; P represents the total number of queued task types in the database cache pool; p represents the task type number. This represents the backlog count value for task type p; This represents the baseline complexity coefficient for the p-th type of task. The baseline complexity coefficient is established by extracting the average number of processor cycles consumed by the same type of task during the stable operation period of the system over the past thirty days and performing maximum and minimum normalization processing. Specifically, the process of collecting the virtual node's operational load sequence and mapping interface latency and service backlog to the operational load sequence to extract the initial impact features of the fault is as follows: Collect the dynamic curve of physical resource consumption of the virtual node within the pre-observation time window to generate the operational load sequence; extract the time-series sampling points of the operational load sequence, align the interface latency and service backlog according to the execution timestamps of the time-series sampling points to generate a multi-dimensional service state cross-section; project the multi-dimensional service state cross-section onto the multi-dimensional tensor space where the operational load sequence resides, and extract the feature co-occurrence distribution density of the multi-dimensional service state cross-section and the operational load sequence in the multi-dimensional tensor space to generate the initial impact features of the fault.

[0026] In this implementation scheme, the system invokes the resource monitoring probe of the underlying operating system of the virtual node to continuously collect processor core utilization, memory page resident ratio, and disk throughput within the pre-observation time window at a fixed sampling frequency. The monitoring data of these three dimensions are then concatenated in chronological order into a continuous three-dimensional vector set, forming a runtime load sequence. Using the sampling frequency of the runtime load sequence as a unified time reference, the previously acquired interface latency and service backlog are resampled using polynomial interpolation, ensuring that the timestamps of the data records at the service level completely coincide with the timestamps of the underlying hardware load. This results in a multi-dimensional service state cross-section containing hardware load and service obstruction data characteristics at a single discrete moment. All multi-dimensional service state cross-sections within the continuous time period are input into a predefined high-dimensional tensor space. A kernel density estimation model is used to calculate the degree of aggregation of software and hardware feature variables in a specific spatial coordinate system and generate initial fault impact features. The formula for calculating the feature co-occurrence distribution density is expressed as: ; H represents the extracted initial impact features of the fault; H represents the total number of aligned time-series sampling points within the preceding observation time window; h represents the sequence number of the time-series sampling point. This represents the running load sequence state value collected at the h-th time-series sampling point; This represents the mean of the baseline operating load sequence extracted from the corresponding virtual node under historical fault-free conditions; This represents the quantized value of the multi-dimensional service state cross-section generated by aligning at the h-th time-series sampling point; This represents the mean of the baseline multidimensional service state cross-section under historical fault-free conditions; Indicates the smoothing bandwidth parameter; This represents the load sensitivity weighting coefficient. This represents the business status sensitivity weighting coefficient. The runtime load sensitivity weighting coefficient... The method for determining the variance is as follows: construct an offline operation and maintenance status dataset containing samples of normal operation and various historical failures; calculate the ratio of the variance of the operating load to the variance of the business status in the sample set; and use the logarithmic transformation of this variance ratio as the weighting coefficient for the operating load sensitivity. And make the business status sensitivity weight coefficient This completes the adaptive establishment of the weight parameters; Specifically, the process of parsing the arrival and rejection times in the work order, extracting the location trajectory set of the maintenance terminal and the service recovery timestamp of the virtual node, and spatially matching the arrival time with the location trajectory set to calculate the position offset is as follows: Parse the work order text to extract the arrival and rejection times; retrieve the maintenance terminal movement coordinate sequence collected by the indoor positioning network node to generate the location trajectory set; extract the log record time point in the system operation log when the core service process status switches to normal response to establish the service recovery timestamp; traverse the location trajectory set and extract the terminal spatial coordinates that have a temporal mapping relationship with the arrival time; perform spatial geometric mapping comparison between the terminal spatial coordinates and the three-dimensional spatial coordinates of the virtual node, and extract the spatial physical distance divergence as the position offset.

[0027] In this implementation plan, the system parses the work order text data filled in by maintenance personnel through a structured form interface, accurately extracting the arrival time and exclusion time. Simultaneously, the system retrieves the indoor positioning base station network deployed in the physical environment, captures the movement coordinate sequence of the maintenance terminal based on its hardware physical address, and maps it to a three-dimensional coordinate system to generate a continuous set of positioning trajectories. At the system level, the system extracts the operation logs of the server where the corresponding virtual node is located, and uses regular expressions to match the identifier fields indicating that the service port is listening normally or has recovered its response status, establishing the precise record generation time of the corresponding log line as the service recovery timestamp. In the positioning trajectory set, the system performs interpolation retrieval along the time dimension, extracting the terminal spatial coordinates that are perfectly aligned with the manually entered arrival time. Subsequently, the system performs a spatial geometric mapping comparison between these terminal spatial coordinates and the inherent three-dimensional spatial coordinates bound to the corresponding virtual node in the digital twin space, quantifying the degree of distance divergence between the two in physical space. The formula for calculating the position offset is expressed as: ; Indicates the position offset; , , These represent the captured three-dimensional spatial coordinates of the maintenance terminal; , , These represent the three-dimensional spatial coordinates of the virtual nodes in the digital twin space; This represents the variance of the positioning signal fluctuation when the indoor positioning base station network captures this coordinate. , , These represent the spatial environment compensation coefficients in three dimensions of the three-dimensional coordinate system. The method for determining the spatial environment compensation coefficients is as follows: extract the thickness of the physical obstruction on the path connecting the maintenance terminal and the virtual node in the digital twin model, map the thickness of the physical obstruction to the signal penetration loss distribution curve, and output the attenuation compensation value of the corresponding dimension as the spatial environment compensation coefficient of that dimension. Specifically, the process of aligning the exclusion time with the service recovery timestamp to calculate the repair lag, and constructing the maintenance execution deviation feature by combining the location offset with the repair lag, is as follows: Map the exclusion time and service recovery timestamp to a unified reference time axis, and extract the absolute time span of the exclusion time and service recovery timestamp on the unified reference time axis as the repair lag; construct a two-dimensional feature matrix containing spatial and temporal dimensions, fill the spatial dimension of the two-dimensional feature matrix with the location offset, fill the temporal dimension of the two-dimensional feature matrix with the repair lag, and perform feature flattening and dimensionality reduction splicing of the two-dimensional feature matrix to generate the maintenance execution deviation feature.

[0028] In this implementation plan, the system converts the manually excluded time and the service recovery timestamp extracted from the underlying layer into a unified standard machine timestamp format, mapping them onto the same linear baseline time axis. On this unified baseline time axis, the system calculates the absolute value of the time difference between the two to quantify the objective time span between the manually recorded time and the actual system recovery time. The formula for calculating the repair lag is expressed as follows: ; Indicates the repair lag; This indicates the excluded time mapped to the unified reference time axis; This represents the service recovery timestamp mapped to a unified baseline time axis; This represents a mapping function that takes the maximum value. The penalty coefficient for malicious falsification is determined by retrieving a sample set of work orders from the historical maintenance database that have been classified as falsely completed. The expected repair time difference for this sample set is calculated, and the absolute value of the current time series span is divided by this expected value. This result is then input into a Sigmoid activation function, and the output activation value is used as the penalty coefficient. After obtaining the basic deviation parameters, the system initializes a two-dimensional feature matrix. The positional offset is written into the first column of the feature matrix to construct a spatial dimension mapping feature, and the repair lag is written into the second column to construct a temporal dimension mapping feature. Subsequently, a projection mapping algorithm is used to perform feature flattening and dimensionality reduction stitching operations on the two-dimensional feature matrix, outputting a scalar result that integrates spatiotemporal correlation attributes. The calculation formula for the maintenance execution deviation feature is expressed as: ; Indicates the characteristics of maintenance execution deviations; Indicates the position offset; G represents the total number of dimension reduction mapping channels; g represents the mapping channel number; The fusion dimensionality reduction weight parameter of the g-th channel is established by performing principal component analysis on the spatiotemporal deviation data under historical normal maintenance conditions, extracting the eigenvalues ​​of the first G principal components and normalizing them. This represents the angle of the projection distribution of the g-th channel between spatial and temporal features, according to... The linear equal-partition mapping rule is established; Specifically, the process of extracting the service dissipation rate of topology-related nodes after the service recovery timestamp and performing correlation mapping between the service dissipation rate and the initial impact features of the fault to extract service recovery hysteresis features is as follows: A feature observation time window is set after the service recovery timestamp, and the service queue backlog decay sequence of topology-related nodes is collected within the feature observation time window; the first-order difference features of the service queue backlog decay sequence at discrete time points are extracted to constitute the service dissipation rate; a multi-dimensional correlation mapping space is constructed, mapping the service dissipation rate to the dynamic response dimension within the multi-dimensional correlation mapping space, mapping the initial impact features of the fault to the initial state dimension within the multi-dimensional correlation mapping space, and extracting the decay deviation of the dynamic response dimension relative to the initial state dimension to generate service recovery hysteresis features.

[0029] In this implementation scheme, the system uses the extracted service recovery timestamp as a starting point and divides a feature observation time window along the positive time axis for continuous monitoring of system uptime. Within this window, the system continuously reads the queued task backlog in the underlying database of the topology-related nodes at a fixed acquisition step size, forming a service queue backlog decay sequence that decays over time. For discrete time points in this sequence, the system uses first-order difference to calculate the queued task reduction of adjacent sampling points, thereby obtaining the service dissipation rate, which characterizes the system's true throughput capacity after troubleshooting. The formula for calculating the service dissipation rate is expressed as: ; K represents the rate of dissipation of traffic; K represents the total number of discrete-time sampling points within the characteristic observation time window; k represents the sampling point number. This represents the backlog of business queues for topologically related nodes extracted at the kth sampling point; This represents the backlog of business queues extracted at the (k+1)th sampling point; This represents the function that takes the maximum value, used to shield against occasional sudden increases in task load. This represents the time step between adjacent discrete-time sampling points. After obtaining the service dissipation rate, the system constructs a multi-dimensional correlation mapping space in memory, containing dynamic response and initial state dimensions. The service dissipation rate is mapped to the dynamic response dimension, and the previously extracted initial fault impact features are mapped to the initial state dimension. Service recovery hysteresis features are extracted by calculating the degree of spatial coordinate deviation between the two in the mapping space. The formula for calculating the service recovery hysteresis features is expressed as: ; This indicates a sluggish business recovery. This indicates the initial impact characteristics of the fault generated in the preceding sequence; This represents the smallest non-zero floating-point constant supported by the system processor. This constant is introduced to prevent denominator overflow errors during calculation. Represents the initial state mapping coefficients; This represents the response hysteresis mapping coefficient. The method for determining the above mapping coefficient is as follows: extract the baseline recovery data set from the historical operation and maintenance database under the same business pressure, assuming hardware self-healing without human intervention; calculate the state variance of the baseline recovery data set in the initial impact dimension as the initial state mapping coefficient; and calculate the reciprocal of the rate variance of the baseline recovery data set in the dissipation response dimension as the response hysteresis mapping coefficient. Please refer to [link to relevant documentation]. Figure 2 In the figure, the horizontal axis corresponds to the characteristic observation time window defined by the system, and the vertical axis corresponds to the backlog of the business queue in the underlying database. The solid lines and discrete nodes in the figure represent the decay sequence of the business queue backlog formed by continuous monitoring within the time window. By extracting the numerical values ​​of adjacent discrete time sampling points in the sequence and calculating the drop (i.e., the dashed triangular area marked in the figure), a first-order difference feature is constructed. The slope represented by this feature is the business dissipation rate. This figure confirms the actual business throughput recovery status of the system after troubleshooting from the data level. Specifically, the process of cross-validating maintenance execution deviation features and business recovery hysteresis features to generate a single work order quality score is as follows: The maintenance execution deviation features are converted into a deviation polarity matrix, and the business recovery hysteresis features are converted into a hysteresis weight matrix; a matrix inner product transformation is performed on the deviation polarity matrix and the hysteresis weight matrix to extract a joint negative feature map; a preset full-score benchmark feature scalar is extracted, and the full-score benchmark feature scalar and the joint negative feature map are input into a nonlinear weighted dimensionality reduction network to extract the dimensionality reduction output scalar value of the full-score benchmark feature scalar under the feature constraints of the joint negative feature map, thereby generating a single work order quality score.

[0030] In this implementation scheme, to quantify the matching degree between the physical execution trajectory of maintenance personnel and the system's underlying business recovery efficiency, the system converts the pre-generated one-dimensional maintenance execution deviation feature into a two-dimensional deviation polarity matrix through orthogonal basis expansion, and simultaneously converts the one-dimensional business recovery hysteresis feature into a hysteresis weight matrix of the same dimension through gradient expansion. Subsequently, the system calls the tensor computation unit to perform matrix inner product transformation on the deviation polarity matrix and the hysteresis weight matrix, fusing their feature information and outputting a joint feature map containing cross-validation correlation information. The calculation formula for the joint feature map is expressed as: ; This represents the feature element in the u-th row and v-th column of the joint feature map; This indicates the maintenance execution deviation characteristics generated in the preceding steps; Let represent the basis vectors of the polarity expansion of the deviation polarity matrix in the u-th row; This represents the hysteresis mapping basis vector in the v-th column of the hysteresis weight matrix; This represents a nonlinear activation function. Next, the system extracts a preset benchmark feature scalar representing the service's perfect score standard from the read-only memory. This benchmark feature scalar, along with the joint feature map, is fed into a nonlinear weighted dimensionality reduction network as input features. The network uses hidden layer neurons to perform feature aggregation and nonlinear decay calculations on the negative blocking elements in the joint feature map, outputting the final score after feature-constrained dimensionality reduction. The formula for calculating the quality score of a single work order is expressed as: ; This indicates the quality score for a single work order; R represents the preset full-score benchmark feature scalar; R represents the total number of one-dimensional feature elements after the joint feature map is flattened; r represents the feature element index; This represents the r-th feature element after flattening; This represents the network connection weights of the nonlinear weighted dimensionality reduction network with respect to the r-th feature element; Represents the global bias parameters of a nonlinear weighted dimensionality reduction network. Network connection weights. and global bias parameters The determination method is as follows: extract massive operation and maintenance execution records that have been de-identified from the historical business system and construct an offline training set with the corresponding refined review and scoring results of management personnel. Use the review and scoring results of management personnel as supervision labels, construct a loss function using the mean square error algorithm, and iteratively update the network node parameters through gradient descent via the backpropagation mechanism. When the feedback calculation value of the loss function decreases and stabilizes within the preset convergence limit, the network connection weights and global bias parameters at this time are extracted and solidified and deployed to the online network model. Specifically, the process of reading the billing parameters from the contract database and cross-calculating the single work order quality score with the billing parameters to generate single performance data is as follows: Parse the operation and maintenance service agreement data entity in the contract database, extract the single maintenance basic pricing vector and service level threshold parameters to generate billing parameters; perform boundary limit comparison between the single work order quality score and the service level threshold parameters to extract the quality assessment deviation; perform multi-dimensional spatial mapping calculation between the quality assessment deviation and the single maintenance basic pricing vector to generate single performance data containing resource reduction coefficients and quality anomaly classification labels.

[0031] In this implementation plan, the system accesses the management agency's contract management database through an encrypted secure interface, parses the structured operation and maintenance service agreement data entities, extracts the resource unit price of each maintenance item to construct a basic pricing vector for a single maintenance service, and simultaneously extracts the service level threshold parameter agreed upon in the agreement. These two are combined to form the billing parameters. The method for determining the service level threshold parameter is as follows: under the premise of legal authorization and shielding the privacy information of the supplier entity, all maintenance work orders that have passed acceptance in historical years are extracted, and the lower quartile of their quality score distribution is calculated as the service level threshold parameter to objectively filter out extreme value interference. Subsequently, the system retrieves the quality score of the previously generated single work order, compares it with the service level threshold parameter execution boundary limit, and extracts the quality assessment deviation degree, which represents the degree of non-compliance. The formula for calculating the quality assessment deviation degree is expressed as: ; Indicates the deviation from the quality assessment; This represents the service level threshold parameter; This indicates the quality score of a single work order generated previously. This represents the maximum value function, used to mask negative deviations when service standards are exceeded. This represents the threshold sensitivity control parameter. The method for determining the threshold sensitivity control parameter is as follows: Based on anonymized historical performance dispute sample data, calculate the proportion of work orders with complaints arising from service quality to the total number of work orders. The product of the reciprocal of this proportion and a constant logarithmic function is set as the threshold sensitivity control parameter. Next, the system performs multi-dimensional spatial mapping calculations on the quality assessment deviation and the single maintenance basic pricing vector to generate single performance efficiency data. The formula for calculating the resource reduction coefficient is expressed as: ; Z represents the extracted resource reduction coefficient; Z represents the total number of dimensions in the basic pricing vector for a single maintenance service; z represents the time sequence number of the pricing dimension. This represents the feature component value of the basic pricing vector for a single maintenance service in the z-th dimension; This represents the penalty weight constant for the z-th dimension. Simultaneously, the system inputs the quality assessment deviation into a preset decision tree classifier, outputs the corresponding quality anomaly classification label, and encapsulates the resource reduction coefficient and the quality anomaly classification label to generate single-time performance data. Specifically, the process of performing trend extrapolation on single-time performance data within the aggregation period and outputting an operation and maintenance service quality assessment result including performance rating and cost adjustment ratio is as follows: Extract all single-time performance data arranged in a timestamp sequence within the accounting period; extract resource reduction coefficients and quality anomaly classification labels from the single-time performance data to construct a performance time-series feature matrix; extract the feature decay slope of the performance time-series feature matrix on the accounting period time axis; perform distribution similarity matching between the feature decay slope and the standard rating feature library to establish the performance rating; aggregate and extract the mean distribution features of resource reduction coefficients within the performance time-series feature matrix; map the mean distribution features to the billing compensation rule matrix to extract the cost adjustment ratio; combine the performance rating and cost adjustment ratio to generate the operation and maintenance service quality assessment result.

[0032] In this implementation plan, at the end of a set quarterly or annual accounting cycle, the system reads all single-transaction performance data arranged in a timestamp sequence within that cycle. The system extracts resource reduction coefficients and quality anomaly classification labels at discrete time points, constructing a multi-dimensional performance time-series feature matrix with the timestamp sequence as the horizontal axis and the resource reduction coefficients and label quantification values ​​as the vertical axis features. To assess the service provider's quality change trend throughout the accounting cycle, the system extracts the feature decay slope of the performance time-series feature matrix on the accounting cycle time axis. The formula for calculating the feature decay slope is as follows: ; The slope represents the characteristic decay rate; F represents the total number of single performance data extracted within the accounting period; f represents the time sequence number of the performance data. This represents the relative value of the timestamp corresponding to the f-th performance data point; This represents the average of all timestamp values ​​within the accounting period. This represents the resource reduction coefficient in the f-th performance data point; This represents the average of all resource reduction coefficients within the accounting period; This represents a constant adjustment to prevent the denominator from becoming zero. After obtaining the feature decay slope, the system performs Gaussian kernel feature similarity matching with a pre-built standard rating feature library, selecting the rating level corresponding to the highest similarity to establish the performance rating. Subsequently, the system aggregates and extracts the mean distribution features of resource reduction coefficients within the performance time-series feature matrix, mapping them to the operation and maintenance billing compensation rule matrix to extract the corresponding cost adjustment ratio. The formula for calculating the cost adjustment ratio is expressed as: ; Indicates the percentage of fees to be adjusted. This represents the hyperbolic tangent mapping function, used to constrain the penalty amplification effect of the decay slope within a fixed numerical boundary; This indicates the influence of the trend on the gain parameter; This represents the baseline adjustment constant obtained by matching from the billing compensation rule matrix. The method for determining the trend impact gain parameter is as follows: extract anonymized historical financial statement records, calculate the median ratio of the historical average service provider quality downgrade penalties to resource reductions, and set this as the gain parameter. Finally, the system encapsulates the determined performance rating and cost adjustment ratio using a multimodal combination, outputting a tamper-proof operation and maintenance service quality assessment result. Please refer to [link / reference]. Figure 3 In the graph, the horizontal axis represents the time series of the accounting cycle, and the vertical axis represents the resource reduction coefficient. The discrete array of dots in the graph represents the single-time performance data extracted throughout the entire accounting cycle. Based on this performance time-series feature matrix, the system uses a Gaussian fitting algorithm to output a trend-following fitted line that runs through the data distribution. The slope of this fitted line is the feature decay slope. The system uses this feature decay slope to determine the direction of subsequent billing compensation, thus transforming the microscopic single-time obstacle-clearing score into an intuitive and objective macroscopic periodic performance rating basis.

[0033] Example 2; please refer to Figure 4 A digital twin-based intelligent assessment system for operation and maintenance service quality, used to execute the digital twin-based intelligent assessment method for operation and maintenance service quality described in the embodiments, includes: a status awareness module, used to receive work order instructions, locate the virtual node and topology-related nodes of the faulty device in the digital twin space, extract the interface latency and service backlog of the topology-related nodes before triggering, collect the operating load sequence of the virtual node, and map the interface latency and service backlog to the operating load sequence to extract the initial impact features of the fault; and a deviation comparison module, used to parse the arrival time and troubleshooting time of the work order, extract the positioning trajectory set of the maintenance terminal and the service recovery timestamp of the virtual node, spatially match the arrival time with the positioning trajectory set to calculate the position offset, and sort the fault. In addition to aligning the repair lag with the service recovery timestamp to calculate the repair lag, the system constructs maintenance execution deviation features by combining the location offset with the repair lag. The quality scoring module extracts the business dissipation rate of topology-related nodes after the service recovery timestamp, performs correlation mapping between the business dissipation rate and the initial impact features of the fault to extract business recovery lag features, and cross-validates the maintenance execution deviation features with the business recovery lag features to generate a single work order quality score. The performance accounting module reads the billing parameters from the contract database, cross-calculates the single work order quality score with the billing parameters to generate single performance performance data, aggregates the single performance performance data within the cycle to perform trend extrapolation, and outputs an operation and maintenance service quality assessment result including performance rating and cost adjustment ratio.

[0034] In this implementation scheme, the state awareness module, serving as the system's data access and initial analysis unit, directly parses and locates the virtual node of the faulty device and the topologically related nodes with which it interacts in the digital twin space upon receiving a work order instruction. This module extracts the interface latency and service backlog of the topologically related nodes before the work order is triggered by calling the underlying communication probe, and simultaneously collects the hardware operating load sequence of the virtual node. Next, the module's internal processing engine maps and aligns the extracted interface latency and service backlog to the operating load sequence according to timestamps, performs multi-dimensional feature concatenation calculations, and outputs the initial impact features of the fault, which are then used as the baseline state data flow to subsequent processing nodes at the initial stage of the fault.

[0035] The deviation comparison module, acting as the system's spatiotemporal status verification unit, is responsible for parsing the arrival and troubleshooting times reported by maintenance personnel from the business form interface. This module extracts the location trajectory set of the maintenance terminal by calling the location network interface and reads the device's operating system logs to establish the service recovery timestamp of the virtual node. During the actual calculation, the module extracts the terminal's physical coordinates aligned with the arrival time from the location trajectory set and calculates the spatial divergence with the virtual node's three-dimensional coordinates to obtain the position offset. Simultaneously, it places the reported troubleshooting time and the system's underlying service recovery timestamp on the same reference time axis, calculates their absolute temporal span to obtain the repair lag, and finally concatenates the position offset and repair lag to construct the maintenance execution deviation characteristics.

[0036] The quality scoring module is used to quantitatively verify the objective effectiveness of troubleshooting actions. This module defines a monitoring window based on the established service recovery timestamp and continuously extracts the service dissipation rate of backlogged tasks within the topology-related nodes. The module's built-in feature mapper correlates the service dissipation rate with the initial fault impact features transmitted from the state awareness module in a multi-dimensional space, extracting service recovery hysteresis features reflecting the system's service throughput recovery status. Based on this, the module receives maintenance execution deviation features output by the deviation comparison module, inputs these features and service recovery hysteresis features into a built-in weighted dimensionality reduction network for cross-validation of execution features, and outputs a single work order quality score after dimensionality reduction mapping calculation.

[0037] The performance accounting module is the terminal unit of the system for completing periodic settlements and outputting auxiliary decision-making data. This module reads billing parameters such as the agreed-upon basic maintenance pricing and service level thresholds from the contract database through a secure interface. It then compares and cross-calculates the received single-work order quality score with the execution limits of the billing parameters to generate single-work performance data including resource reduction coefficients. At the end of the preset financial accounting cycle, this module automatically aggregates all single-work performance data over the time series within the cycle, constructs a performance time-series feature matrix, extracts its feature decay slope, and deduces the evolution trend of the execution status. Finally, it combines the performance rating established by similarity matching with the cost adjustment ratio calculated by rule mapping to generate and issue the operation and maintenance service quality assessment results.

[0038] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A digital twin-based intelligent assessment method for operation and maintenance service quality, characterized in that, Includes the following steps: S1. Receive work order instructions, locate the virtual node and topology associated node of the faulty device in the digital twin space, extract the interface latency and business backlog of the topology associated node before the trigger, collect the running load sequence of the virtual node, and map the interface latency and business backlog to the running load sequence to extract the initial impact features of the fault. S2. Analyze the arrival time and exclusion time reported in the work order, extract the location trajectory set of the maintenance terminal and the service recovery timestamp of the virtual node, spatially match the arrival time with the location trajectory set to calculate the position offset, align the exclusion time with the service recovery timestamp to calculate the repair lag, and construct the maintenance execution deviation feature by combining the position offset and the repair lag. S3. Extract the service dissipation rate of the topology-related nodes after the service recovery timestamp, perform association mapping between the service dissipation rate and the initial impact features of the fault to extract the service recovery hysteresis features, cross-validate the maintenance execution deviation features with the service recovery hysteresis features, and generate a single work order quality score. S4. Read the billing parameters from the contract database, cross-calculate the single work order quality score with the billing parameters to generate single performance performance data, aggregate the execution trend of single performance performance data within the cycle, and output the operation and maintenance service quality assessment results including performance rating and cost adjustment ratio.

2. The intelligent evaluation method for operation and maintenance service quality based on digital twins according to claim 1, characterized in that: The specific process of receiving work order instructions, locating the virtual node and topology-related nodes of the faulty device in the digital twin space, and extracting the interface latency and service backlog of the topology-related nodes before the trigger is as follows: Parse the device identification code contained in the work order instruction, and retrieve the three-dimensional spatial coordinates bound to the device identification code in the digital twin space to establish a virtual node; Traverse the system operation topology graph associated with virtual nodes, extract business nodes connected to the upstream and downstream of business data flows, and establish topology-related nodes. Based on the trigger time of the work order instruction, an advance observation time window is extracted. Within the advance observation time window, the network packet round-trip time sequence of the topology-related nodes is captured to establish the interface latency. The total number of queued tasks in the underlying database of the topology-related nodes is aggregated to establish the business backlog.

3. The intelligent evaluation method for operation and maintenance service quality based on digital twins according to claim 1, characterized in that: The specific process of collecting the runtime load sequence of virtual nodes and mapping interface latency and service backlog to the runtime load sequence to extract the initial impact features of faults is as follows: The dynamic curves of physical resource consumption of virtual nodes within the pre-observation time window are collected to generate a running load sequence. Extract the time-series sampling points of the running load sequence, align the interface latency and business backlog according to the execution timestamp of the time-series sampling points, and generate a multi-dimensional business status cross-section; The multidimensional business state section is projected onto the multidimensional tensor space where the running load sequence is located. The feature co-occurrence distribution density of the multidimensional business state section and the running load sequence in the multidimensional tensor space is extracted to generate the initial impact features of the fault.

4. The intelligent evaluation method for operation and maintenance service quality based on digital twins according to claim 1, characterized in that: The specific process of parsing the arrival time and exclusion time reported in the work order, extracting the location trajectory set of the maintenance terminal and the service recovery timestamp of the virtual node, and spatially matching the arrival time with the location trajectory set to calculate the location offset is as follows: Parse the work order text to extract the arrival time and exclusion time; retrieve the maintenance terminal movement coordinate sequence collected by indoor positioning network nodes to generate a positioning trajectory set; Extract the log records from the system operation logs showing the time when the core service process status switched to normal response to establish the service recovery timestamp. Traverse the set of positioning trajectories and extract the terminal spatial coordinates that have a temporal mapping relationship with the arrival time. Perform spatial geometric mapping comparison between the terminal spatial coordinates and the three-dimensional spatial coordinates of the virtual nodes, and extract the spatial physical distance divergence as the position offset.

5. The intelligent evaluation method for operation and maintenance service quality based on digital twins according to claim 1, characterized in that: The specific process of aligning the exclusion time with the service recovery timestamp to calculate the repair lag, and constructing the maintenance execution deviation characteristics by combining the location offset with the repair lag, is as follows: Map the exclusion time and service recovery timestamp to a unified baseline time axis, and extract the absolute time span of the exclusion time and service recovery timestamp on the unified baseline time axis as the repair lag. Construct a two-dimensional feature matrix containing spatial and temporal dimensions. Fill the spatial dimension of the two-dimensional feature matrix with the position offset and the temporal dimension with the repair lag. Perform feature flattening and dimensionality reduction splicing of the two-dimensional feature matrix to generate maintenance execution deviation features.

6. The intelligent evaluation method for operation and maintenance service quality based on digital twins according to claim 1, characterized in that: The specific process of extracting the service dissipation rate of topology-related nodes after the service recovery timestamp, and performing an association mapping between the service dissipation rate and the initial impact features of the fault to extract service recovery hysteresis features is as follows: Set a feature observation time window after the service recovery timestamp, and collect the service queue backlog decay sequence of topologically related nodes within the feature observation time window; extract the first-order difference features of the service queue backlog decay sequence at discrete time points to form the service dissipation rate; A multidimensional correlation mapping space is constructed, the business dissipation rate is mapped to the dynamic response dimension within the multidimensional correlation mapping space, the initial impact features of the fault are mapped to the initial state dimension within the multidimensional correlation mapping space, the attenuation deviation of the dynamic response dimension relative to the initial state dimension is extracted, and the business recovery hysteresis features are generated.

7. The intelligent evaluation method for operation and maintenance service quality based on digital twins according to claim 1, characterized in that: The specific process of cross-validating maintenance execution deviation characteristics with business recovery lag characteristics to generate a single work order quality score is as follows: The maintenance execution deviation features are converted into a deviation polarity matrix, and the business recovery hysteresis features are converted into a hysteresis weight matrix; the inner product transformation of the deviation polarity matrix and the hysteresis weight matrix is ​​performed to extract the joint negative feature map. Extract the preset full-score benchmark feature scalar, input the full-score benchmark feature scalar and the joint negative feature map into the nonlinear weighted dimensionality reduction network, extract the dimensionality reduction output scalar value of the full-score benchmark feature scalar under the feature constraints of the joint negative feature map, and generate a single work order quality score.

8. The intelligent evaluation method for operation and maintenance service quality based on digital twins according to claim 1, characterized in that: The specific process of reading the billing parameters from the contract database and cross-calculating the single work order quality score with the billing parameters to generate single performance data is as follows: Parse the operation and maintenance service agreement data entities in the contract database, extract the basic pricing vector for a single maintenance service and combine it with the service level threshold parameters to generate billing parameters; Compare the quality score of a single work order with the boundary limit values ​​of the service level threshold parameter to extract the deviation of the quality assessment. The deviation of quality assessment is mapped and calculated in a multi-dimensional space to the basic pricing vector of a single maintenance service, generating single performance data that includes resource reduction coefficients and quality anomaly classification labels.

9. The intelligent evaluation method for operation and maintenance service quality based on digital twins according to claim 1, characterized in that: The specific process of extrapolating the execution trend of single-time performance data within the aggregation cycle and outputting the operation and maintenance service quality assessment results, including performance rating and cost adjustment ratio, is as follows: Extract all single performance performance data arranged in timestamp sequence within the accounting period, and extract the resource reduction coefficient and quality anomaly classification label from the single performance performance data to construct a performance time series feature matrix; Extract the feature decay slope of the performance time series feature matrix on the accounting cycle time axis, and perform performance rating by matching the distribution similarity between the feature decay slope and the standard rating feature library; Aggregate and extract the mean distribution features of resource reduction coefficients within the performance time-series feature matrix, map the mean distribution features to the billing compensation rule matrix to extract the cost adjustment ratio, and combine the performance rating and cost adjustment ratio to generate the operation and maintenance service quality assessment results.

10. A digital twin-based intelligent assessment system for operation and maintenance service quality, used to execute the digital twin-based intelligent assessment method for operation and maintenance service quality as described in any one of claims 1-9, characterized in that, include: The status awareness module is used to receive work order instructions, locate the virtual nodes and topology-related nodes of the faulty equipment in the digital twin space, extract the interface latency and business backlog of the topology-related nodes before the trigger, collect the running load sequence of the virtual nodes, and map the interface latency and business backlog to the running load sequence to extract the initial impact features of the fault. The deviation comparison module is used to parse the arrival time and exclusion time of the work order, extract the positioning trajectory set of the maintenance terminal and the service recovery timestamp of the virtual node, spatially match the arrival time with the positioning trajectory set to calculate the position offset, chronologically align the exclusion time with the service recovery timestamp to calculate the repair lag, and construct the maintenance execution deviation feature by combining the position offset and the repair lag. The quality scoring module is used to extract the service dissipation rate of topology-related nodes after the service recovery timestamp, perform correlation mapping between the service dissipation rate and the initial impact features of the fault to extract the service recovery hysteresis features, cross-validate the maintenance execution deviation features with the service recovery hysteresis features, and generate a single work order quality score. The performance accounting module is used to read the billing parameters from the contract database, cross-calculate the quality score of a single work order with the billing parameters to generate single performance performance data, aggregate the execution trend of single performance performance data within the cycle, and output the operation and maintenance service quality assessment results including performance rating and cost adjustment ratio.

Images