A method, system, equipment and medium for power metering and operation and maintenance resource scheduling
By collecting and analyzing multi-dimensional power metering and operation and maintenance data, and combining power metering and operation and maintenance knowledge graphs and pre-trained language models, a semantic description of scheduling strategies is generated and mapped to a multi-objective optimization model. This solves the problems of manual dependence and uneven resource allocation in power metering and operation and maintenance, and realizes an efficient and interpretable resource scheduling scheme.
Patent Information
- Application Number
- CN202611272097.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-20
- Publication Date
- 2026-11-10
AI Technical Summary
The existing power metering and operation and maintenance model relies on human experience, resulting in low operation and maintenance efficiency and uneven resource allocation, making it difficult to meet the unique needs of the power metering and operation and maintenance field. Moreover, the existing model lacks the ability to understand unstructured data processing and is unable to generate reasonable scheduling schemes.
By collecting multi-dimensional metering and operation and maintenance data, querying target knowledge fragments from the power metering and operation and maintenance knowledge graph, performing contextual analysis using the trained metering and operation and maintenance scheduling decision model, generating semantic descriptions of scheduling strategies, mapping them to dynamic parameters in a multi-objective optimization model, and generating resource scheduling schemes by combining multi-objective optimization algorithms.
It has achieved automated power metering and operation and maintenance resource scheduling, improved the accuracy of fault diagnosis and the efficiency of work order generation, reduced the time spent on repeated work assignments and on-site investigations, reduced operation and maintenance costs and average response time, and improved the rationality of resource scheduling and the interpretability of decisions.
Smart Images

Figure CN123070583A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of power system metering and operation and maintenance, and in particular to a method, system, equipment and medium for scheduling power metering and operation and maintenance resources. Background Technology
[0002] With the widespread adoption of smart meters and the continuous advancement of the ubiquitous power Internet of Things (IoT) construction, the number of power metering devices / equipment is exploding, leading to a sharp increase in the workload and complexity of metering operation and maintenance. Metering operation and maintenance services are increasingly characterized by large equipment scale, complex operating conditions, diverse fault types, and high real-time service demands. As a crucial link in ensuring the safe and stable operation of the power grid and improving the quality of power supply services, the efficiency and quality of power metering operation and maintenance directly impact the operating costs of power companies and customer satisfaction.
[0003] Traditional metering-based operation and maintenance (O&M) models primarily rely on manual experience to formulate O&M plans and assign work orders. O&M personnel determine fault types and allocate tasks based on experience, lacking scientific quantitative analysis and global optimization in the scheduling process. This model suffers from significant problems such as slow response times, low O&M efficiency, and uneven resource allocation. For example, inaccurate fault location leads to unnecessary travel for O&M personnel, and the failure to dynamically optimize and match different regions and types of O&M tasks based on personnel skills, geographical location, and urgency results in a serious waste of human and vehicle resources.
[0004] To address the aforementioned issues, two mainstream resource scheduling and optimization schemes have been proposed. The first scheme is resource collaborative scheduling based on multi-objective optimization algorithms. This type of scheme generates collaborative scheduling decisions by integrating multi-source structured data such as power generation, meteorology, and market data. This method has certain advantages in processing structured data and solving multi-objective optimization problems. However, the input of the optimization model of this type of scheme is highly dependent on predefined structured data fields. The flexibility and adaptability of the model are strictly limited by the data format. For example, when faced with a large amount of unstructured work order information in power metering and operation and maintenance scenarios, it lacks the ability to understand the complex semantics, implicit business needs, and risk information in text and speech. This may cause the generated optimal scheduling scheme to deviate from the actual business context, resulting in an imbalance between efficiency and user satisfaction. The second approach is to use a deep reinforcement learning-based model for state prediction and resource decision optimization. This model can learn the optimal strategy from a large amount of interactive data. However, the interpretability of this model is poor (it cannot meet the high requirements for safety and accountability in operation and maintenance management), it relies on massive interactive data (leading to difficulties in model training and slow convergence), and it is difficult to effectively and systematically integrate the expert experience and rule knowledge in the field of power metering (such as common fault modes of specific models of electricity meters and power supply rules for specific holidays). Its decisions may violate basic industry safety procedures or experience consensus.
[0005] In addition, the above solutions are mostly general models for general energy management or power plant operation and maintenance, lacking in-depth adaptation to the unique needs of the sub-sector of power metering operation and maintenance (such as massive terminals, specific fault types, and strong correlation with customer service). Power metering operation and maintenance has unique needs such as massive terminal equipment, specific fault types, and strong correlation with customer service, and general models often cannot meet these professional requirements. Summary of the Invention
[0006] In view of this, the purpose of this application is to provide a method, system, equipment, and medium for power metering operation and maintenance resource scheduling, which realizes automated power metering operation and maintenance resource scheduling, saves labor and time costs, greatly reduces the burden of manual coordination and decision-making, significantly improves the accuracy of fault diagnosis and work order generation efficiency, reduces duplicate work assignments and on-site troubleshooting time caused by misjudgments, and significantly reduces average response time and operation and maintenance costs, while simultaneously improving the accuracy of fault diagnosis, the rationality of resource scheduling, and the interpretability of decisions. The specific solution is as follows: Firstly, this application discloses a method for scheduling power metering and operation and maintenance resources, including: Collect multi-dimensional metering and operation and maintenance data for the target power system to obtain the current multi-dimensional power metering and operation and maintenance data; The system queries a preset power metering and maintenance knowledge graph to find the target knowledge fragment corresponding to the current multidimensional power metering and maintenance data; the target knowledge fragment includes power fault entities and associated fault paths; The current multidimensional power metering and maintenance data and the target knowledge fragment are input into the trained metering and maintenance scheduling decision model. The target knowledge fragment is used as a prompt word to perform contextual analysis on the current multidimensional power metering and maintenance data, generating a semantic description of the scheduling strategy for the target power system. The metering and maintenance scheduling decision model is a model obtained by fine-tuning a pre-trained language model using historical multidimensional power metering and maintenance data. The semantic description of the scheduling strategy includes fault diagnosis results and corresponding fault diagnosis work orders. The semantic description of the scheduling strategy is mapped to dynamic parameters in a multi-objective optimization model; the dynamic parameters include a set of objective function weights and a set of constraints. Using a multi-objective optimization algorithm and based on the semantic description of the scheduling strategy, the status information of schedulable operation and maintenance resources, and the dynamic parameters, multiple optimization objectives in the multi-objective optimization model are solved to generate an operation and maintenance resource scheduling scheme, and corresponding operation and maintenance tasks are executed according to the operation and maintenance resource scheduling scheme; the multiple optimization objectives include resource scheduling response time, resource scheduling response cost, and resource load balancing.
[0007] Optionally, the power metering and maintenance resource scheduling method further includes: Collect multi-dimensional historical operation and maintenance data for the target power system to obtain historical multi-dimensional power metering operation and maintenance data; The historical multidimensional power metering and operation and maintenance data are labeled to obtain labeled historical operation and maintenance corpus; the labeled historical operation and maintenance corpus includes first sample data and second sample data, the first sample data is paired sample data of natural language work order description and fault diagnosis conclusion, and the second sample data is paired sample data of natural language work order description and power resource demand list. The pre-trained language model is supervised and fine-tuned using LoRA technology and based on the labeled historical operation and maintenance corpus to obtain the metering operation and maintenance scheduling decision model.
[0008] Optionally, the step of querying the target knowledge fragment corresponding to the current multidimensional power metering and operation and maintenance data from a preset power metering and operation and maintenance knowledge graph includes: The current multidimensional power metering and maintenance data is preprocessed to obtain processed metering and maintenance data, and the processed metering and maintenance data is vectorized to obtain metering and maintenance feature data. Keywords are extracted from the metering and operation and maintenance feature data to obtain operation and maintenance feature keywords, and target knowledge fragments corresponding to the operation and maintenance feature keywords are retrieved from the preset power metering and operation and maintenance knowledge graph.
[0009] Optionally, the preprocessing of the current multidimensional power metering and maintenance data to obtain processed metering and maintenance data includes: The current multidimensional power metering operation and maintenance data is spatiotemporally aligned to obtain aligned operation and maintenance data; the current multidimensional power metering operation and maintenance data includes equipment operation status data, equipment business attribute data, equipment spatial topology data, and equipment abnormal work order data of different power metering devices in the target power system; The aligned operation and maintenance data is normalized to obtain the processed metering operation and maintenance data.
[0010] Optionally, the fault diagnosis results include the root cause of the fault, the location of the fault, and the confidence level of the fault. Accordingly, mapping the semantic description of the scheduling strategy to dynamic parameters in a multi-objective optimization model includes: The pre-created policy converter maps the qualitative description of work orders in the semantic description of the scheduling policy to the work order response time weight in the multi-objective optimization model; the policy converter is configured with a pre-set mapping rule base and / or a supervised learning model. The business requirements described in the semantic description of the scheduling strategy are mapped to hard matching constraints that need to be followed when solving the multi-objective optimization model; the hard matching constraints include fault handling skill matching constraints, fault handling qualification requirements constraints, and fault handling time and capacity constraints. Obtain real-time traffic data for the target area corresponding to the fault location, and based on the real-time traffic data, transform the traffic congestion information for the target area in the semantic description of the scheduling strategy into path travel cost weights in a multi-objective optimization model.
[0011] Optionally, the power metering and maintenance resource scheduling method further includes: When a new power metering and maintenance resource scheduling request is received, multi-dimensional metering and maintenance data for the target power system are collected to obtain real-time metering and maintenance data; the real-time metering and maintenance data includes equipment abnormality work order data. The real-time metering and maintenance data is used as the current multi-dimensional power metering and maintenance data, and the process jumps to the step of querying the target knowledge fragment corresponding to the current multi-dimensional power metering and maintenance data from the preset power metering and maintenance knowledge graph to generate a new resource scheduling scheme. The new resource scheduling scheme and the operation and maintenance resource scheduling scheme are sorted according to the priority of the equipment abnormality work order data to obtain the sorted resource scheduling scheme, and each scheduling scheme in the sorted resource scheduling scheme is executed in sequence.
[0012] Optionally, executing the corresponding operation and maintenance tasks according to the operation and maintenance resource scheduling scheme includes: An operation and maintenance task is generated based on the operation and maintenance resource scheduling scheme; the operation and maintenance task includes the target fault handling terminal, the travel path to the power metering equipment, and the semantic description of the scheduling strategy. The maintenance task is sent to the target fault handling terminal so that the maintenance personnel corresponding to the target fault handling terminal can execute the maintenance task.
[0013] Secondly, this application discloses a power metering and operation and maintenance resource scheduling system, comprising: The data acquisition module is used to collect multi-dimensional metering and operation and maintenance data for the target power system to obtain the current multi-dimensional power metering and operation and maintenance data. The query module is used to query the target knowledge fragments corresponding to the current multidimensional power metering and operation and maintenance data from the preset power metering and operation and maintenance knowledge graph; the target knowledge fragments include power fault entities and associated fault paths; The input module is used to input the current multidimensional power metering and maintenance data and the target knowledge fragment into the trained metering and maintenance scheduling decision model, so as to use the target knowledge fragment as prompt words to perform contextual analysis on the current multidimensional power metering and maintenance data and generate a semantic description of the scheduling strategy for the target power system; the metering and maintenance scheduling decision model is a model obtained by fine-tuning a pre-trained language model using historical multidimensional power metering and maintenance data, and the semantic description of the scheduling strategy includes fault diagnosis results and corresponding fault diagnosis work orders. The mapping module is used to map the semantic description of the scheduling strategy into dynamic parameters in a multi-objective optimization model; the dynamic parameters include a set of objective function weights and a set of constraints. The scheduling scheme generation module is used to solve multiple optimization objectives in the multi-objective optimization model using a multi-objective optimization algorithm and based on the semantic description of the scheduling strategy, the status information of schedulable operation and maintenance resources, and the dynamic parameters, to generate an operation and maintenance resource scheduling scheme, and to execute corresponding operation and maintenance tasks according to the operation and maintenance resource scheduling scheme; the multiple optimization objectives include resource scheduling response time, resource scheduling response cost, and resource load balancing degree.
[0014] Thirdly, this application discloses an electronic device, including a processor and a memory; wherein, when the processor executes a computer program stored in the memory, it implements the aforementioned power metering and maintenance resource scheduling method.
[0015] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned power metering and maintenance resource scheduling method.
[0016] As can be seen, this application first collects multi-dimensional metering and operation and maintenance data for the target power system to obtain current multi-dimensional power metering and operation and maintenance data. Then, it queries a preset power metering and operation and maintenance knowledge graph for target knowledge fragments (including power fault entities and associated fault paths) corresponding to the current multi-dimensional power metering and operation and maintenance data. Next, it inputs the current multi-dimensional power metering and operation and maintenance data and the target knowledge fragments into a trained metering and operation and maintenance scheduling decision model. The target knowledge fragments are used as prompt words to perform contextual analysis on the current multi-dimensional power metering and operation and maintenance data to generate a semantic description of the scheduling strategy for the target power system. The metering and operation and maintenance scheduling decision model utilizes historical multi-dimensional power metering and operation and maintenance data. Based on the model obtained after fine-tuning the pre-trained language model, the scheduling strategy semantic description includes fault diagnosis results and corresponding fault diagnosis work orders. Finally, the scheduling strategy semantic description is mapped to dynamic parameters (including the set of objective function weights and the set of constraints) in a multi-objective optimization model. Using a multi-objective optimization algorithm and based on the scheduling strategy semantic description, schedulable operation and maintenance resource status information, and the dynamic parameters, multiple optimization objectives in the multi-objective optimization model are solved to generate an operation and maintenance resource scheduling scheme. The corresponding operation and maintenance tasks are then executed according to the operation and maintenance resource scheduling scheme. The multiple optimization objectives include resource scheduling response time, resource scheduling response cost, and resource load balancing.
[0017] This application first queries the pre-defined power metering and operation and maintenance knowledge graph to find the target knowledge fragments (including power fault entities and associated fault paths) corresponding to the current multi-dimensional power metering and operation and maintenance data. Then, it inputs the metering and operation and maintenance feature data and the target knowledge fragments into a metering and operation and maintenance scheduling decision model based on a pre-trained language model, thereby generating a scheduling strategy semantic description that includes fault diagnosis results and corresponding fault diagnosis work orders. The scheduling strategy semantic description is then mapped to dynamic parameters in a multi-objective optimization model. The multi-objective optimization algorithm is then used to solve multiple optimization objectives in the multi-objective optimization model based on the scheduling strategy semantic description, schedulable operation and maintenance resource status information, and dynamic parameters, thereby generating an operation and maintenance resource scheduling scheme. Through the above methods, automated power metering and maintenance resource scheduling is achieved, saving labor and time costs while significantly reducing the burden of manual coordination and decision-making. By combining multi-dimensional power metering and maintenance data, power metering and maintenance knowledge graphs, and pre-trained language models, the accuracy of fault diagnosis and work order generation efficiency can be greatly improved. The deep understanding and reasoning capabilities of the large language model for multi-source information make fault root cause analysis more accurate, thereby reducing the time spent on repeated dispatching and on-site investigation due to misjudgment and improving the success rate of initial fault handling. Secondly, by deeply integrating the semantic understanding capabilities of the large language model with multi-objective optimization algorithms, it is possible to consider multiple dimensions such as efficiency, cost, load, and customer experience simultaneously, enabling the generation of more reasonable and economical scheduling schemes under complex real-time constraints, significantly reducing average response time and maintenance costs, thereby achieving a simultaneous improvement in fault diagnosis accuracy, resource scheduling rationality, and decision interpretability. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0019] Figure 1 This application discloses a flowchart of a power metering and operation and maintenance resource scheduling method. Figure 2 This application discloses a specific flowchart of a power metering and operation and maintenance resource scheduling method. Figure 3 This is a schematic diagram of the structure of a power metering and operation and maintenance resource scheduling system disclosed in this application; Figure 4 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation
[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0021] This application discloses a method for scheduling power metering and maintenance resources. (See also...) Figure 1 As shown, the method includes: Step S11: Collect multi-dimensional metering and operation data for the target power system to obtain the current multi-dimensional power metering and operation data.
[0022] In this embodiment, multi-dimensional metering and maintenance data for the target power system are first collected from multiple data sources, including the power automation metering system, the power customer service system, the power asset management system, and the mobile power operation and maintenance platform, to obtain the current multi-dimensional power metering and maintenance data. This current multi-dimensional power metering and maintenance data includes, but is not limited to, equipment operating status data, equipment business attribute data, equipment spatial topology data, and equipment abnormal work order data for different power metering devices within the target power system.
[0023] Specifically, equipment operation status data refers to the dynamic operating indicators of all power metering equipment within a preset time window before the current moment, including but not limited to voltage / current values, active / reactive power, power factor, clock deviation, and online / offline status. Equipment business attribute data refers to the static business attribute information of power metering equipment, used to distinguish the business characteristics of different equipment, including but not limited to equipment type identification, installation location type, user level, and equipment asset commissioning date. Equipment spatial topology data refers to the static geographical location information of power metering equipment, including but not limited to longitude, latitude, altitude, and the power supply substation number and line topology relationship to which the equipment belongs. Additionally, it may include user description information (such as smart meter malfunctions), historical maintenance records, and equipment spatial location information, among other data types.
[0024] Step S12: Query the target knowledge fragment corresponding to the current multidimensional power metering and maintenance data from the preset power metering and maintenance knowledge graph; the target knowledge fragment includes power fault entities and associated fault paths.
[0025] In this embodiment, after collecting the current multidimensional power metering and maintenance data, the system further retrieves the preset power metering and maintenance knowledge graph based on the current multidimensional power metering and maintenance data to query the target knowledge fragments corresponding to the current multidimensional power metering and maintenance data; wherein the target knowledge fragments include, but are not limited to, power fault entities and associated fault paths.
[0026] In this embodiment, the step of querying the target knowledge fragment corresponding to the current multidimensional power metering and operation and maintenance data from a preset power metering and operation and maintenance knowledge graph can specifically include: preprocessing the current multidimensional power metering and operation and maintenance data to obtain processed metering and operation and maintenance data, and vectorizing the processed metering and operation and maintenance data to obtain metering and operation and maintenance feature data; extracting keywords from the metering and operation and maintenance feature data to obtain operation and maintenance feature keywords, and retrieving the target knowledge fragment corresponding to the operation and maintenance feature keywords from the preset power metering and operation and maintenance knowledge graph. In this embodiment, considering that multidimensional metering and operation and maintenance data from multiple data sources have diverse formats, inconsistent sampling frequencies, scattered data sources, and large differences in dimensions, corresponding preprocessing operations can be performed on them, such as spatiotemporal alignment, normalization, and mask construction, to generate a standard data format that meets the requirements of subsequent keyword extraction and model input. Furthermore, after preprocessing the current multidimensional power metering and operation data, the processed metering and operation data can be vectorized using a vectorization model (such as Word2Vec) to obtain metering and operation feature data. Then, keyword extraction algorithms, such as the TF-IDF (Term Frequency-Inverse Document Frequency) algorithm based on word frequency and the TextRank (a graph-based ranking algorithm for text) algorithm based on text importance, are used to extract keywords from the metering and operation feature data to obtain operation and operation feature keywords. Finally, the target knowledge fragment corresponding to the operation and operation feature keyword is retrieved from the preset power metering and operation knowledge graph.
[0027] It should be noted that this application pre-maintains a lightweight power metering and operation knowledge graph based on a Knowledge Graph-Retrieval-Augmented Generation (KG-RAG) architecture. Node E can specifically include information such as the faulty equipment model, fault symptoms, fault handling measures, professional skills required to resolve the fault, and spare parts model. Edge R represents the relationships between different nodes. Specifically, when processing a new equipment fault work order, keywords in the work order text can be used as the search entry point to find relevant power fault entities and associated fault paths in the knowledge graph G, and the retrieved structured knowledge fragments can be processed. The data is converted into natural language text and used as part of the prompt words, which are then input along with the original equipment fault work order into the fine-tuned large language model. By dynamically injecting static domain knowledge into the large language model for reasoning, the professionalism and accuracy of fault diagnosis can be significantly improved.
[0028] Specifically, the preprocessing of the current multi-dimensional power metering and maintenance data to obtain processed metering and maintenance data may include: spatiotemporally aligning the current multi-dimensional power metering and maintenance data to obtain aligned maintenance data; and normalizing the aligned maintenance data to obtain processed metering and maintenance data. In this embodiment, spatiotemporal alignment can be performed on multi-dimensional power metering and maintenance data from multiple different data sources with different sampling frequencies (e.g., electrical quantity data at 15 minutes / time, work order data triggered by events) according to timestamps. For example, according to a preset unified time window. (e.g., 15 minutes) The sampling frequency in multi-dimensional power metering and maintenance data is higher than The data is aggregated using average values (for example, for time-series data such as electrical quantities and equipment monitoring with sampling frequencies higher than a unified time window, when aligning to a unified time window, the average of multiple continuously collected raw values within that window is taken as the unique representative data for that period, thereby compressing the high-frequency data volume, unifying the time-series dimension, and enabling metering and maintenance data with different collection frequencies to match the same time axis, facilitating subsequent input into large models and optimization scheduling engines for calculations); for data with sampling frequencies lower than... The data can be filled using the forward fill method, thereby mapping the data from all power metering devices onto the time axis in a unified manner. Next, to eliminate the dimensional differences between different data such as voltage (V), current (A), and power (kW) and accelerate model convergence, the Z-Score normalization method can be used to normalize the aligned operation and maintenance data for each feature dimension. Specifically, for the j-th feature value of the i-th device collected... Its normalized calculation formula is: ; In the formula, Let be the mean of the j-th feature value of the i-th device collected so far across the full dataset in the corresponding dimension. The currently collected feature values standard deviation These are the normalized eigenvalues. After the above normalization process, the distribution of all eigenvalues is mapped to the standard normal distribution interval with a mean of 0 and a variance of 1.
[0029] Step S13: Input the current multidimensional power metering and maintenance data and the target knowledge fragment into the trained metering and maintenance scheduling decision model, so as to use the target knowledge fragment as a prompt word to perform contextual analysis on the current multidimensional power metering and maintenance data and generate a scheduling strategy semantic description for the target power system; the metering and maintenance scheduling decision model is a model obtained by fine-tuning a pre-trained language model using historical multidimensional power metering and maintenance data, and the scheduling strategy semantic description includes fault diagnosis results and corresponding fault diagnosis work orders.
[0030] In this embodiment, after preprocessing the multidimensional power metering and operation data, the processed metering and operation data and the target knowledge fragment can be input into the trained metering and operation scheduling decision model. The model will use the target knowledge fragment as a prompt to perform contextual analysis on the processed metering and operation data, thereby generating a structured fault diagnosis work order i for the target power system. Then, it will further analyze the context of the fault diagnosis work order i to generate a semantic description of the scheduling strategy. The description should include at least the fault diagnosis results and the corresponding structured fault diagnosis work order i. This description can summarize scheduling preferences in natural language. For example, this work order involves an important industrial customer, and the fault is a plant-wide power outage with extremely high risk; therefore, it should be prioritized and assigned to a team with experience in handling high-voltage side faults and special operation qualifications. The fault diagnosis results include, but are not limited to, information such as the root cause of the fault, the location of the fault, and the confidence level of the fault.
[0031] The metering operation and maintenance scheduling decision model is a model obtained by fine-tuning a pre-trained language model (i.e., Large Language Model, LLM, as a base model) using historical multi-dimensional power metering operation and maintenance data.
[0032] It should be noted that before scheduling operation and maintenance resources, a metering operation and maintenance scheduling decision model needs to be created. The specific creation process includes: collecting multi-dimensional historical operation and maintenance data for the target power system to obtain historical multi-dimensional power metering operation and maintenance data; labeling the historical multi-dimensional power metering operation and maintenance data to obtain labeled historical operation and maintenance corpus; the labeled historical operation and maintenance corpus includes first sample data and second sample data, the first sample data being paired sample data of natural language work order description and fault diagnosis conclusion, and the second sample data being paired sample data of natural language work order description and power resource demand list; using LoRA technology and based on the labeled historical operation and maintenance corpus, the pre-trained language model is supervised and fine-tuned to obtain the metering operation and maintenance scheduling decision model. In this embodiment, firstly, multi-dimensional historical operation and maintenance data for the target power system is collected, including but not limited to historical equipment operating status data, historical equipment business attribute data, historical equipment spatial topology data, and historical equipment abnormal work order data for different power metering devices within the target power system, to obtain historical multi-dimensional power metering operation and maintenance data. Then, corresponding preprocessing operations are performed on the structured and unstructured data in the historical multi-dimensional power metering operation and maintenance data. For example, for structured data such as equipment files and real-time equipment monitoring data obtained from the metering asset management system, they are standardized and cleaned to form structured feature vectors. For unstructured device malfunction work order data input by users via voice, text, etc., it can be converted into text using an ASR (Automatic Speech Recognition) model to obtain a text work order. This text work order is then input into a pre-trained text encoding model to encode it into a fixed-dimensional semantic vector. For unstructured multimedia data collected on-site (such as equipment images and videos), key visual feature vectors can be extracted using convolutional neural networks (CNNs). All vectorized data can be organized into a unified multimodal context data package, including its corresponding data source, timestamp, and other metadata. Furthermore, the aforementioned multimodal context data packets can be annotated to obtain an annotated historical operation and maintenance corpus. This annotated historical operation and maintenance corpus contains two types of sample data: a first sample data set and a second sample data set. The first sample data set consists of paired sample data of natural language work order descriptions and structured fault diagnosis conclusions, while the second sample data set consists of paired sample data of natural language work order descriptions and standard power resource demand lists. Finally, using LoRA (Low-Rank Adaptation) technology and based on the annotated historical operation and maintenance corpus, the pre-trained language model is supervised fine-tuned (SFT) to obtain a metering operation and maintenance scheduling decision model. The goal of the fine-tuning is to enable the model to learn the following two core tasks: first, to identify fault entities from free text, infer root causes, and output confidence scores; second, to generate structured diagnostic work orders containing fault types, required skills, estimated working hours, necessary tools, and spare parts based on user-input fault descriptions.
[0033] It should be noted that this application uses the LoRA method in the process of fine-tuning the model. This method only requires training a small number of parameters, which significantly reduces the memory usage, consumes little resources, trains quickly, and facilitates rapid iteration. In addition, it can retain the general capabilities of the base model and avoid catastrophic forgetting.
[0034] In addition, for the metering operation and maintenance scheduling decision model obtained after training, in order to meet the diverse application needs in the power metering operation and maintenance scenario, its performance can be quantitatively evaluated from multiple dimensions to obtain the corresponding performance evaluation results, and to determine whether the evaluation results meet the preset requirements. For example, it should have the ability to understand a large amount of work order text (including professional terms, colloquial expressions and implicit semantics), be able to infer the cause of the fault from the fuzzy problem description input by the user, and generate structured fault diagnosis results. Furthermore, the final generated scheduling decision should provide natural language explanation and meet the traceability requirements of operation and maintenance management. At the same time, it should meet the feasibility deployment requirements, such as the computing power constraints and response latency requirements in the power intranet environment.
[0035] In one specific implementation, the formula for calculating the performance evaluation result is as follows: ; In the formula, This represents the semantic understanding accuracy of the model on the historical work order dataset; Indicates the accuracy of fault diagnosis / identification; The BLEU (Bilingual Evaluation Understudy) score is used to evaluate the quality of the structured text generated by the model. For example, the standardized diagnostic content produced by the model is compared with the standard reference text, and the score is calculated by statistically analyzing the overlap of consecutive word segments. At the same time, a shortness penalty mechanism can be set to prevent the model from outputting too short text. The higher the score, the higher the matching degree between the structured work orders and fault conclusions generated by the model and the standard reference text. This score is used to measure the model's ability to output standardized and structured operation and maintenance text. Score the inference speed of the model in the target hardware environment; Cost scoring for commercial use of the model; to The weights for each dimension can be dynamically adjusted based on the actual business scenario.
[0036] Specifically, the BLEU score is obtained by calculating the overlap of n consecutive words (n-grams), and is used to evaluate the similarity between the model-generated text and the human-generated reference text. The higher the similarity of word sequences between the model-generated text and the human-generated reference text (e.g., the more identical word sequences), the higher the quality of the model's generation. The formula for calculating the BLEU score is: ; In the formula, The precision of n-grams represents the proportion of n-grams in the generated text that appear in the reference text to the total number of n-grams in the generated text. The weights of each n-gram are usually taken as the average, i.e. Additionally, the Brevity Penalty (BP) mechanism can be used to penalize excessively short generated text, which can be specifically represented as follows: ; In the formula, c is the length of the text generated by the model, and r is the length of the reference text.
[0037] After obtaining the performance evaluation results Then, it can be compared with a preset threshold. If it is greater than the threshold, it indicates that the model meets the requirements and can be applied to actual resource scheduling tasks.
[0038] Furthermore, the model can be continuously optimized and adjusted according to a preset fine-tuning cycle. For example, based on the processing results reported by on-site personnel via a mobile app, the model's diagnostic results, scheduling plans, and actual resource scheduling and fault handling results are automatically compared. High-quality feedback data is then selected, which can be used for incremental fine-tuning of the metering and maintenance scheduling decision model. For instance, samples with diagnostic errors or large deviations are selected, manually verified, and used to form new high-quality paired samples. Parameter-Efficient Fine-Tuning (PEFT) is then performed periodically to correct model errors and enrich its experience. Through this feedback mechanism, the model can continuously absorb new experience, thereby adapting to changes in operation and maintenance models and the external environment.
[0039] Step S14: Map the semantic description of the scheduling strategy to dynamic parameters in a multi-objective optimization model; the dynamic parameters include a set of objective function weights and a set of constraints.
[0040] In this embodiment, after generating the semantic description of the scheduling strategy, it can be mapped to dynamic parameters in a multi-objective optimization model, specifically including the set of objective function weights and the set of constraints.
[0041] In one specific implementation, mapping the semantic description of the scheduling strategy to dynamic parameters in a multi-objective optimization model may specifically include: mapping the qualitative description of work orders in the semantic description of the scheduling strategy to work order response time weights in the multi-objective optimization model through a pre-created strategy converter; the strategy converter is configured with a preset mapping rule base and / or a supervised learning model; mapping the business requirement description in the semantic description of the scheduling strategy to hard matching constraints that must be followed when solving the multi-objective optimization model; the hard matching constraints include fault handling skill matching constraints. Fault handling qualification requirements and constraints Fault handling time and capacity constraints The system acquires real-time traffic data for the target area corresponding to the fault location and, based on this data, transforms the traffic congestion information for the target area in the semantic description of the scheduling strategy into path travel cost weights in a multi-objective optimization model. In this embodiment, the scheduling strategy semantic description can be transformed using a pre-created strategy converter (embedded with a configurable rule mapping library and / or a lightweight supervised learning model). The qualitative description of a work order in the algorithm is mapped to a quantifiable work order response time weight in a multi-objective optimization model. For example, the rule definition in the rule mapping library maps the highest priority processing level (such as P0 level) to the weight of the average response time sub-objective of work order i in the multi-objective optimization function. Add a dynamic increment to the global baseline value Mathematically, this is represented as: .
[0042] At the same time, the scheduling policy semantic description can be... The requirement description is mapped to optimization constraints. For example, the description of needing high-voltage side fault handling experience is mapped to a hard constraint: the cosine similarity between the skill tag vector of assigned personnel j and the standard skill vector of high-voltage side fault handling must be greater than or equal to a preset threshold. (e.g., 0.9); or, mapping the description of special operation qualifications to another hard constraint. The personnel qualification list must include specific qualifications. ,Right now .
[0043] In addition, the scheduling strategy semantic description can also be included. The contextual information is mapped to model parameters. For example, real-time traffic data of the target area corresponding to the fault location is obtained, and the congestion information of the target area during the morning rush hour is converted into weights of the path travel time cost sub-objective in the path planning sub-model. It should be noted that the travel time cost coefficient of a specific path / segment... Multiply by the penalty factor .
[0044] Finally, the policy converter outputs a set of dynamically adjusted objective function weights W (containing...). and the newly added set of constraints .
[0045] It should be noted that this can be achieved by analyzing the effects of historical decisions (such as the semantic description of scheduling strategies). The weights actually selected The relationship between natural language keywords and quantitative parameters is continuously optimized by using Bayesian optimization or regression analysis methods to automatically adjust or learn the optimal mapping relationship between natural language keywords and quantitative parameters.
[0046] Step S15: Using a multi-objective optimization algorithm and based on the semantic description of the scheduling strategy, the status information of schedulable operation and maintenance resources, and the dynamic parameters, solve multiple optimization objectives in the multi-objective optimization model to generate an operation and maintenance resource scheduling scheme, and execute corresponding operation and maintenance tasks according to the operation and maintenance resource scheduling scheme; the multiple optimization objectives include resource scheduling response time, resource scheduling response cost, and resource load balancing.
[0047] In this embodiment, the set of all structured diagnostic work orders T to be scheduled in the semantic description of the scheduling strategy, the status information R of the schedulable operation and maintenance resources, and the dynamic parameters (including the set of objective function weights W output by the strategy converter and the set of newly added constraints C) are input into a pre-created multi-objective dynamic optimization scheduling engine. The multi-objective optimization algorithm (such as the NSGA-III algorithm) is then used to solve multiple optimization objectives in the multi-objective optimization model to generate an operation and maintenance resource scheduling scheme. Then, the corresponding operation and maintenance tasks are executed according to the operation and maintenance resource scheduling scheme.
[0048] The optimization objectives include, but are not limited to, resource scheduling response time (reflecting customer service quality), resource scheduling response cost (such as the total driving cost of vehicles, reflecting operational economy), and resource load balancing (reflecting team fairness). These three objectives are interdependent and require finding an optimal balance while satisfying various business constraints. If the set of work orders to be scheduled is T, and the set of available vehicles is Vehicles, then the objective optimization model can be expressed as: ; in, This indicates the resource scheduling response time. For dynamic weights generated by model semantic understanding, For the estimated arrival time, The time when the work order was created.
[0049] in, , representing the total cost of travel. Let K be the operating path for vehicle K. The fuel cost function for traveling along this path; This represents the resource load imbalance penalty (i.e., resource load balance), which is used to promote a balanced distribution of tasks among vehicles.
[0050] The optimization objective is: This indicates the various constraints that need to be met, including the fault handling skill matching constraint. Fault handling qualification requirements and constraints Fault handling time and capacity constraints wait.
[0051] It should be noted that the entire resource scheduling process may also include: when a new power metering and maintenance resource scheduling request is received, collecting the current multi-dimensional metering and maintenance data for the target power system to obtain real-time metering and maintenance data; the real-time metering and maintenance data includes equipment abnormal work order data; using the real-time metering and maintenance data as the current multi-dimensional power metering and maintenance data, and jumping to the step of querying the target knowledge fragment corresponding to the current multi-dimensional power metering and maintenance data from the preset power metering and maintenance knowledge graph to generate a new resource scheduling scheme; sorting the new resource scheduling scheme and the maintenance resource scheduling scheme based on the priority corresponding to the equipment abnormal work order data to obtain a sorted resource scheduling scheme, and executing each scheduling scheme in the sorted resource scheduling scheme in sequence. In this embodiment, during the current resource scheduling process, the execution progress of the current scheduling task and the external environment are continuously monitored. When the triggering conditions are met (such as a new emergency work order, resource delay (e.g., failure to resolve the fault within the estimated time of the fault diagnosis work order), sudden weather changes, etc.), the rescheduling process can be started immediately. This process employs a local re-optimization strategy, using the current state of the executing scheduling task as a new initial condition to regenerate the scheduling strategy / scheme. It only adjusts the allocation of tasks that haven't started or are directly affected, and pushes the adjustment instructions to the corresponding mobile terminals in real time. For example, when a new power metering and maintenance resource scheduling request is received, the process first collects multi-dimensional metering and maintenance data of the current target power system to obtain real-time metering and maintenance data including equipment anomaly work order data. This real-time metering and maintenance data is then used as the current multi-dimensional power metering and maintenance data, and the process re-queries the target knowledge fragments corresponding to the current multi-dimensional power metering and maintenance data from a preset power metering and maintenance knowledge graph until a new resource scheduling scheme is generated. Then, based on the priority corresponding to the equipment anomaly work order data, the new resource scheduling scheme and the maintenance resource scheduling scheme are sorted to obtain a sorted resource scheduling scheme, and each scheduling scheme in the sorted resource scheduling scheme is executed sequentially. For example, it automatically infers "high risk of fire" from the description of "burnt smell" and converts it into a higher scheduling priority; similarly, it automatically understands the humanistic care needs represented by the description of "elderly person living alone" and adjusts the scheduling strategy accordingly.
[0052] For example, when a user's smart meter in a certain community is detected to have lost communication during the morning session, and a voice work order from the customer service system reporting a sudden power outage is automatically received, the system immediately integrates the meter's disconnection data, the user's voice-to-text message, the normal electricity consumption data of other users in the area, and the meter's asset information to obtain integrated information. Then, this integrated information is analyzed using a large language model, and a fault diagnosis result is output: suspected circuit breaker tripping or internal fault in the electricity meter, with a confidence level of 88%. A standardized fault diagnosis work order is automatically generated, indicating that the circuit breaker and meter need to be checked. Next, the optimization scheduling engine, based on the fault location and skill requirements indicated in the work order, and combined with the real-time resource pool status information (showing that engineer A has just completed a task nearby with matching skills), initiates optimization calculations. At the same time, the large model-assisted analysis indicates that this period is the morning rush hour, and the area around the target community is congested. It is recommended to increase the weight of the travel time sub-objective in the multi-objective optimization model. After comprehensive calculation, the scheduling engine decides to assign the work order to engineer A and plans the optimal travel route to avoid congested sections. Next, the fault diagnosis work order and navigation information are pushed to Engineer A's mobile app in real time. Upon arriving at the site, Engineer A confirms that the user's circuit breaker has tripped, restores power after closing the breaker, takes a photo via the app, selects the "resolved" option, and reports the actual fault as a tripped circuit breaker on the user's side. Simultaneously, all information related to this task is recorded, and the actual processing result is correlated with the initial fault diagnosis result, used as positive samples in the subsequent training data of the large model. Meanwhile, when a high-priority work order with abnormal power data from a critical industrial user is received, a dynamic rescheduling process is immediately triggered, re-allocating other available resources to Engineer A's subsequent low-priority tasks, thereby ensuring rapid response to key customers.
[0053] As can be seen, in this embodiment, the target knowledge fragments (including power fault entities and associated fault paths) corresponding to the current multi-dimensional power metering and maintenance data are first queried from the preset power metering and maintenance knowledge graph. Then, the metering and maintenance feature data and the target knowledge fragments are input into the metering and maintenance scheduling decision model based on the pre-trained language model, thereby generating a scheduling strategy semantic description containing fault diagnosis results and corresponding fault diagnosis work orders. The scheduling strategy semantic description is then mapped to dynamic parameters in the multi-objective optimization model. The multi-objective optimization algorithm is then used to solve multiple optimization objectives in the multi-objective optimization model based on the scheduling strategy semantic description, schedulable maintenance resource status information, and dynamic parameters, thereby generating a maintenance resource scheduling scheme. Through the above methods, automated power metering and maintenance resource scheduling is achieved, saving labor and time costs while significantly reducing the burden of manual coordination and decision-making. By combining multi-dimensional power metering and maintenance data, power metering and maintenance knowledge graphs, and pre-trained language models, the accuracy of fault diagnosis and work order generation efficiency can be greatly improved. The deep understanding and reasoning capabilities of the large language model for multi-source information make fault root cause analysis more accurate, thereby reducing the time spent on repeated dispatching and on-site investigation due to misjudgment and improving the success rate of initial fault handling. Secondly, by deeply integrating the semantic understanding capabilities of the large language model with multi-objective optimization algorithms, it is possible to consider multiple dimensions such as efficiency, cost, load, and customer experience simultaneously, enabling the generation of more reasonable and economical scheduling schemes under complex real-time constraints, significantly reducing average response time and maintenance costs, thereby achieving a simultaneous improvement in fault diagnosis accuracy, resource scheduling rationality, and decision interpretability.
[0054] This application discloses a specific method for scheduling power metering and operation and maintenance resources. (See also...) Figure 2 As shown, the method includes: Step S21: Collect multi-dimensional metering and operation data for the target power system to obtain the current multi-dimensional power metering and operation data.
[0055] Step S22: Query the target knowledge fragment corresponding to the current multidimensional power metering and maintenance data from the preset power metering and maintenance knowledge graph; the target knowledge fragment includes power fault entities and associated fault paths.
[0056] Step S23: Input the current multidimensional power metering and maintenance data and the target knowledge fragment into the trained metering and maintenance scheduling decision model, so as to use the target knowledge fragment as a prompt word to perform contextual analysis on the current multidimensional power metering and maintenance data and generate a scheduling strategy semantic description for the target power system; the metering and maintenance scheduling decision model is a model obtained by fine-tuning a pre-trained language model using historical multidimensional power metering and maintenance data, and the scheduling strategy semantic description includes fault diagnosis results and corresponding fault diagnosis work orders.
[0057] Step S24: Map the semantic description of the scheduling strategy to dynamic parameters in a multi-objective optimization model; the dynamic parameters include a set of objective function weights and a set of constraints.
[0058] Step S25: Using a multi-objective optimization algorithm and based on the scheduling strategy semantic description, schedulable operation and maintenance resource status information, and the dynamic parameters, solve multiple optimization objectives in the multi-objective optimization model to generate an operation and maintenance resource scheduling scheme, and generate operation and maintenance tasks based on the operation and maintenance resource scheduling scheme; the operation and maintenance tasks include the target fault handling terminal, the travel path to the power metering equipment, and the scheduling strategy semantic description, and the multiple optimization objectives include resource scheduling response time, resource scheduling response cost, and resource load balancing degree.
[0059] In this embodiment, after solving multiple optimization objectives in the multi-objective optimization model to generate an operation and maintenance resource scheduling scheme, an operation and maintenance task can be generated based on the operation and maintenance resource scheduling scheme. This task includes a target fault handling terminal (such as a matched A-type mobile terminal), a travel path to the power metering equipment (such as a navigation route to the faulty equipment), and a scheduling strategy semantic description.
[0060] Step S26: Send the maintenance task to the target fault handling terminal so that the maintenance personnel corresponding to the target fault handling terminal can execute the maintenance task.
[0061] In this embodiment, the above-mentioned maintenance task is sent to the target fault handling terminal (i.e., worker A's mobile terminal) so that the maintenance personnel (i.e. worker A) corresponding to the target fault handling terminal can carry the corresponding tools according to the semantic description of the scheduling strategy and handle the faulty equipment in a timely manner according to the planned navigation route.
[0062] For more detailed processing procedures of steps S21 to S24, please refer to the relevant content disclosed in the foregoing embodiments, which will not be repeated here.
[0063] As can be seen, this embodiment utilizes a multi-objective optimization algorithm and, based on the semantic description of the scheduling strategy, the status information of schedulable operation and maintenance resources, and dynamic parameters, solves multiple optimization objectives in the multi-objective optimization model to generate an operation and maintenance resource scheduling scheme. Based on this scheme, operation and maintenance tasks are generated and then sent to the target fault handling terminal so that the corresponding operation and maintenance personnel can execute them. This significantly improves the accuracy of fault diagnosis and the efficiency of fault diagnosis work order generation. The deep understanding and reasoning capabilities of the large language model for multi-source metering operation and maintenance data make fault root cause analysis more accurate, reducing duplicate dispatching and on-site troubleshooting time due to misjudgments, and increasing the success rate of initial fault handling. Secondly, it achieves global, dynamic, and refined optimal allocation of operation and maintenance resources. By deeply integrating the semantic understanding capabilities of the large language model with the multi-objective optimization algorithm, it can simultaneously consider multiple dimensions such as efficiency, cost, load, and customer experience, generating more reasonable and economical scheduling schemes under complex real-time constraints, significantly reducing average response time and operation and maintenance costs. Furthermore, it enhances the interpretability of decision-making and the adaptive capabilities of the system, and realizes full-process automation from anomaly detection, intelligent diagnosis, automatic dispatching, path planning to feedback learning, which greatly reduces the burden of manual coordination and decision-making, thereby improving the accuracy of electricity metering, the reliability of power grid operation, and the quality of electricity services for users.
[0064] Accordingly, this application also discloses a power metering and operation and maintenance resource scheduling system, see [link to relevant documentation]. Figure 3 As shown, the system includes: The acquisition module 11 is used to acquire multi-dimensional metering and operation and maintenance data for the target power system and obtain the current multi-dimensional power metering and operation and maintenance data. The query module 12 is used to query the target knowledge fragment corresponding to the current multidimensional power metering and operation and maintenance data from the preset power metering and operation and maintenance knowledge graph; the target knowledge fragment includes power fault entities and associated fault paths; Input module 13 is used to input the current multidimensional power metering operation and maintenance data and the target knowledge fragment into the trained metering operation and maintenance scheduling decision model, so as to use the target knowledge fragment as prompt words to perform contextual analysis on the current multidimensional power metering operation and maintenance data and generate a scheduling strategy semantic description for the target power system; the metering operation and maintenance scheduling decision model is a model obtained by fine-tuning a pre-trained language model using historical multidimensional power metering operation and maintenance data, and the scheduling strategy semantic description includes fault diagnosis results and corresponding fault diagnosis work orders; Mapping module 14 is used to map the semantic description of the scheduling strategy into dynamic parameters in a multi-objective optimization model; the dynamic parameters include a set of objective function weights and a set of constraints. The scheduling scheme generation module 15 is used to solve multiple optimization objectives in the multi-objective optimization model using a multi-objective optimization algorithm and based on the semantic description of the scheduling strategy, the status information of schedulable operation and maintenance resources, and the dynamic parameters, to generate an operation and maintenance resource scheduling scheme, and to execute corresponding operation and maintenance tasks according to the operation and maintenance resource scheduling scheme; the multiple optimization objectives include resource scheduling response time, resource scheduling response cost, and resource load balancing degree.
[0065] The specific workflow of each of the above modules can be found in the relevant content disclosed in the foregoing embodiments, and will not be repeated here.
[0066] Furthermore, embodiments of this application also disclose an electronic device, Figure 4 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.
[0067] Figure 4 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the power metering and operation and maintenance resource scheduling method disclosed in any of the foregoing embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.
[0068] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.
[0069] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.
[0070] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the power metering and maintenance resource scheduling method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include a computer program capable of performing other specific tasks.
[0071] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned disclosed power metering and operation and maintenance resource scheduling method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.
[0072] Furthermore, embodiments of this application also disclose a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the power metering operation and maintenance resource scheduling method disclosed above.
[0073] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.
[0074] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0075] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.
[0076] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0077] The above provides a detailed description of a power metering and maintenance resource scheduling method, system, equipment, and medium provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A method for scheduling power metering and operation and maintenance resources, characterized in that, include: Collect multi-dimensional metering and operation and maintenance data for the target power system to obtain the current multi-dimensional power metering and operation and maintenance data; Query the target knowledge fragments corresponding to the current multidimensional power metering and operation and maintenance data from the preset power metering and operation and maintenance knowledge graph; The target knowledge fragment includes power fault entities and associated fault paths; The current multidimensional power metering and maintenance data and the target knowledge fragment are input into the trained metering and maintenance scheduling decision model. The target knowledge fragment is used as a prompt word to perform contextual analysis on the current multidimensional power metering and maintenance data, generating a semantic description of the scheduling strategy for the target power system. The metering and maintenance scheduling decision model is a model obtained by fine-tuning a pre-trained language model using historical multidimensional power metering and maintenance data. The semantic description of the scheduling strategy includes fault diagnosis results and corresponding fault diagnosis work orders. The semantic description of the scheduling strategy is mapped to dynamic parameters in a multi-objective optimization model; the dynamic parameters include a set of objective function weights and a set of constraints. Using a multi-objective optimization algorithm and based on the semantic description of the scheduling strategy, the status information of schedulable operation and maintenance resources, and the dynamic parameters, multiple optimization objectives in the multi-objective optimization model are solved to generate an operation and maintenance resource scheduling scheme, and corresponding operation and maintenance tasks are executed according to the operation and maintenance resource scheduling scheme; the multiple optimization objectives include resource scheduling response time, resource scheduling response cost, and resource load balancing.
2. The power metering and maintenance resource scheduling method according to claim 1, characterized in that, Also includes: Collect multi-dimensional historical operation and maintenance data for the target power system to obtain historical multi-dimensional power metering operation and maintenance data; The historical multidimensional power metering and operation and maintenance data are labeled to obtain labeled historical operation and maintenance corpus; the labeled historical operation and maintenance corpus includes first sample data and second sample data, the first sample data is paired sample data of natural language work order description and fault diagnosis conclusion, and the second sample data is paired sample data of natural language work order description and power resource demand list. The pre-trained language model is supervised and fine-tuned using LoRA technology and based on the labeled historical operation and maintenance corpus to obtain the metering operation and maintenance scheduling decision model.
3. The power metering and maintenance resource scheduling method according to claim 1, characterized in that, The step of querying the target knowledge fragment corresponding to the current multidimensional power metering and operation data from the preset power metering and operation knowledge graph includes: The current multidimensional power metering and maintenance data is preprocessed to obtain processed metering and maintenance data, and the processed metering and maintenance data is vectorized to obtain metering and maintenance feature data. Keywords are extracted from the metering and operation and maintenance feature data to obtain operation and maintenance feature keywords, and target knowledge fragments corresponding to the operation and maintenance feature keywords are retrieved from the preset power metering and operation and maintenance knowledge graph.
4. The power metering and maintenance resource scheduling method according to claim 3, characterized in that, The preprocessing of the current multidimensional power metering and maintenance data to obtain processed metering and maintenance data includes: The current multidimensional power metering operation and maintenance data is spatiotemporally aligned to obtain aligned operation and maintenance data; the current multidimensional power metering operation and maintenance data includes equipment operation status data, equipment business attribute data, equipment spatial topology data, and equipment abnormal work order data of different power metering devices in the target power system; The aligned operation and maintenance data is normalized to obtain the processed metering operation and maintenance data.
5. The power metering and maintenance resource scheduling method according to claim 1, characterized in that, The fault diagnosis results include the root cause of the fault, the location of the fault, and the confidence level of the fault. Accordingly, mapping the semantic description of the scheduling strategy to dynamic parameters in a multi-objective optimization model includes: The pre-created policy converter maps the qualitative description of work orders in the semantic description of the scheduling policy to the work order response time weight in the multi-objective optimization model; the policy converter is configured with a pre-set mapping rule base and / or a supervised learning model. The business requirements described in the semantic description of the scheduling strategy are mapped to hard matching constraints that need to be followed when solving the multi-objective optimization model; the hard matching constraints include fault handling skill matching constraints, fault handling qualification requirements constraints, and fault handling time and capacity constraints. Obtain real-time traffic data for the target area corresponding to the fault location, and based on the real-time traffic data, transform the traffic congestion information for the target area in the semantic description of the scheduling strategy into path travel cost weights in a multi-objective optimization model.
6. The power metering and maintenance resource scheduling method according to claim 3, characterized in that, Also includes: When a new power metering and maintenance resource scheduling request is received, multi-dimensional metering and maintenance data for the target power system are collected to obtain real-time metering and maintenance data. The real-time metering and maintenance data includes equipment malfunction work order data. The real-time metering and maintenance data is used as the current multi-dimensional power metering and maintenance data, and the process jumps to the step of querying the target knowledge fragment corresponding to the current multi-dimensional power metering and maintenance data from the preset power metering and maintenance knowledge graph to generate a new resource scheduling scheme. The new resource scheduling scheme and the operation and maintenance resource scheduling scheme are sorted according to the priority of the equipment abnormality work order data to obtain the sorted resource scheduling scheme, and each scheduling scheme in the sorted resource scheduling scheme is executed in sequence.
7. The power metering and operation and maintenance resource scheduling method according to any one of claims 1 to 6, characterized in that, The step of executing the corresponding operation and maintenance tasks according to the operation and maintenance resource scheduling scheme includes: An operation and maintenance task is generated based on the operation and maintenance resource scheduling scheme; the operation and maintenance task includes the target fault handling terminal, the travel path to the power metering equipment, and the semantic description of the scheduling strategy. The maintenance task is sent to the target fault handling terminal so that the maintenance personnel corresponding to the target fault handling terminal can execute the maintenance task.
8. A power metering and operation and maintenance resource scheduling system, characterized in that, include: The data acquisition module is used to collect multi-dimensional metering and operation and maintenance data for the target power system to obtain the current multi-dimensional power metering and operation and maintenance data. The query module is used to query the target knowledge fragments corresponding to the current multidimensional power metering and operation and maintenance data from the preset power metering and operation and maintenance knowledge graph. The target knowledge fragment includes power fault entities and associated fault paths; The input module is used to input the current multidimensional power metering and maintenance data and the target knowledge fragment into the trained metering and maintenance scheduling decision model, so as to use the target knowledge fragment as prompt words to perform contextual analysis on the current multidimensional power metering and maintenance data and generate a semantic description of the scheduling strategy for the target power system; the metering and maintenance scheduling decision model is a model obtained by fine-tuning a pre-trained language model using historical multidimensional power metering and maintenance data, and the semantic description of the scheduling strategy includes fault diagnosis results and corresponding fault diagnosis work orders. The mapping module is used to map the semantic description of the scheduling strategy into dynamic parameters in a multi-objective optimization model; the dynamic parameters include a set of objective function weights and a set of constraints. The scheduling scheme generation module is used to solve multiple optimization objectives in the multi-objective optimization model using a multi-objective optimization algorithm and based on the semantic description of the scheduling strategy, the status information of schedulable operation and maintenance resources, and the dynamic parameters, to generate an operation and maintenance resource scheduling scheme, and to execute corresponding operation and maintenance tasks according to the operation and maintenance resource scheduling scheme; the multiple optimization objectives include resource scheduling response time, resource scheduling response cost, and resource load balancing degree.
9. An electronic device, characterized in that, It includes a processor and a memory; wherein, when the processor executes a computer program stored in the memory, it implements the power metering operation and maintenance resource scheduling method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, Used to store computer programs; wherein, when the computer programs are executed by a processor, they implement the power metering and maintenance resource scheduling method as described in any one of claims 1 to 7.
