Power supply enterprise dynamic precise staff determination method and system based on multi-dimensional data association

Abstract

The present application relates to the technical field of power supply enterprise human resource management, in particular to a dynamic and accurate staff allocation method and system for power supply enterprises based on multi-dimensional data association, which comprises the following steps: collecting multi-source heterogeneous data, covering six types of data sources such as equipment, business and work order, and completing ETL processing; generating dynamic association coefficients by fusing linear / nonlinear association, knowledge graph calibration and structure dimension adjustment; calculating accurate working hours based on machine learning, completing quota and non-quota time calculation and verification; calculating the number of staff by accounting for total working hours and combining with comprehensive efficiency coefficient; and implementing dynamic iteration optimization. The system of the present application comprises seven modules of data collection, processing, association modeling, working hour calculation, staff calculation, dynamic optimization and output display. The present application realizes the association of the whole chain of "equipment-business-personnel", has strong dynamic adaptability, is in line with the actual business of power supply enterprises, and greatly improves the efficiency of human resource allocation.

Description

Technical Field

[0001] This invention relates to the field of human resource management technology for power supply enterprises, and in particular to a method and system for dynamic and accurate staffing of power supply enterprises based on multi-dimensional data correlation. Background Technology

[0002] As a key energy security industry, power supply companies have businesses covering multiple scenarios such as transmission line operation and maintenance, substation duty, distribution emergency repair, and customer service. The production process is significantly complex due to factors such as equipment distribution, power grid structure, and load fluctuations, while the service process is highly random due to factors such as customer demand and sudden failures. Therefore, labor staffing analysis is a core aspect of enterprise operation and management.

[0003] For a long time, power supply companies have mainly adopted the traditional staffing method based on direct calculation of equipment quantity, such as the "Power Industry Power Supply Labor Staffing" (LD / T46-2016), which directly provides staffing data tables based on core data such as transformer capacity, transmission line length, and number of customers, providing guidance for standardized staffing management. However, with technological and industry changes such as the grid connection of new energy sources and the construction of smart grids, the traditional method has gradually exposed many defects: it is highly subjective and lacks an intermediary mechanism, failing to establish a complete correlation chain between equipment, business, and personnel, resulting in a disconnect between staffing results and actual workload; it is static, rigid, and lacks adaptability, unable to dynamically respond to efficiency changes brought about by the popularization of smart equipment and business process optimization, resulting in lagging parameter updates; it does not fully utilize data and has a single dimension, relying only on basic data on the number of equipment while ignoring key factors such as working hours, worker skill levels, and working environment; it lacks consideration for sudden business, either ignoring or subjectively estimating temporary business such as emergency repairs and power supply, easily leading to personnel redundancy or insufficient emergency response; and its optimization methods are crude, failing to break through the direct mapping logic of "equipment-personnel," resulting in limited optimization effects that are difficult to replicate and promote.

[0004] Therefore, there is an urgent need for a precise staffing technology solution that fits the business characteristics of power supply enterprises, integrates multi-dimensional data, and supports dynamic adjustments, so as to overcome the limitations of traditional methods and achieve precise, dynamic, and intelligent staffing analysis. Summary of the Invention

[0005] To overcome the problems existing in the prior art, the present invention provides a method and system for dynamic and accurate staffing of power supply enterprises based on multi-dimensional data correlation.

[0006] The technical solution adopted by this invention to achieve the above objectives is: a dynamic and accurate staffing method for power supply enterprises based on multi-dimensional data correlation, comprising the following steps:

[0007] S1. Collect multi-source heterogeneous data from power supply enterprises, including equipment basic data, business operation data, work order process data, work record data, working hours and personnel data, and procedures and guidelines data;

[0008] S2. Perform differentiated cleaning and structuring processing on multi-source heterogeneous data according to data type, and convert the data into a unified structured format to obtain a standardized dataset;

[0009] S3. Based on standardized datasets, feature data is extracted, and through linear and nonlinear association mining, knowledge graph construction and structural dimension adjustment, and DS evidence theory fusion, a dynamic association model between equipment and business is obtained.

[0010] S4. Based on standardized datasets and equipment-business dynamic correlation models, construct a multi-dimensional working time calculation model to calculate the standard time and non-standard time of working hours respectively.

[0011] S5. Based on GIS technology and multi-source heterogeneous data, calculate the single trip and annual total travel time to obtain the travel time.

[0012] S6. Based on the quota time, non-quota time and travel time, calculate the total annual working hours, and combine the annual average effective working time per person to calculate the theoretical number of employees and obtain the staffing plan for the power supply company.

[0013] Preferably, the method further includes step S7, establishing a dynamic and precise staffing plan for the power supply company: updating multi-source heterogeneous data according to actual conditions, incrementally training the equipment-business dynamic correlation model and the multi-dimensional working hour calculation model, monitoring the load deviation between the power supply company's staffing plan and the actual configuration, and triggering dynamic optimization adjustment when the deviation exceeds a set threshold.

[0014] Preferably, in step S1, the extraction, transformation and loading of multi-source heterogeneous data are completed by ETL tools. The transformation specifically includes data type conversion, missing value filling, data standardization and data association and fusion. Among them, the mean filling method is used for numerical fields, the mode filling method is used for categorical fields, and the numerical data is mapped to the [0,1] interval using min-max normalization.

[0015] Preferably, in step S2, the data types include numerical data, text data, process data, and geographic information data; the differential cleaning includes: removing outliers from numerical data using the 3σ principle and combining it with secondary verification based on industry standards;

[0016] Text-based data is segmented and semantically extracted using a power supply-specific dictionary based on BERT, and redundant information with cosine similarity that does not meet the set value is removed.

[0017] For process-oriented data, the ProM tool is used to mine and remove invalid work orders with short processing times or many steps.

[0018] Geographic information data undergoes coordinate transformation via Mercator projection, and the least squares method is used to correct coordinate offsets, ensuring that the corrected coordinate error is less than a set threshold.

[0019] Preferably, in step S3, linear association mining is achieved by calculating the Pearson correlation coefficient, and nonlinear association mining is achieved by constructing a LightGBM regression model;

[0020] The knowledge graph construction includes designing core nodes for equipment, business, procedures, and constraints; defining the "association and influence" node relationships and assigning initial weights; defining inference rules in conjunction with the power supply company's procedures and guidelines; and updating the node association weights through a neural symbolic inference framework.

[0021] Structural dimension adjustments are achieved by using the ARIMA model and the AIC criterion to determine optimal parameters, predict changes in business volume caused by equipment upgrades and maintenance plans, and generate structural adjustment coefficients.

[0022] The DS evidence theory integrates quantitative dimension calibration coefficients and structural adjustment coefficients, and the integration is effective when the conflict coefficient meets the set value.

[0023] Preferably, in step S4, the standard time = preparation time + operation time + completion time, and the non-standard time = unavoidable waiting time + human-caused delay time + travel time.

[0024] Preferably, in step S5, the optimal route planning prioritizes highway routes, and the Haversine formula is used to calculate the straight-line distance. For non-straight-line routes, a route correction factor γ is multiplied, with γ=1.1 for urban areas, γ=1.2 for suburbs, and γ=1.5 for mountainous areas. The traffic speed calibration sets a benchmark value according to the geographical area, with 40km / h for urban areas, 30km / h for suburbs, and 20km / h for mountainous areas. It is also adjusted according to weather conditions, with a correction factor of 1.0 for sunny days, 0.8 for rainy days, and 0.6 for snowy days.

[0025] Preferably, in step S6, the total annual working hours = total annual quota working hours + total annual unavoidable waiting working hours + total annual travel time + reserved working hours for unexpected business; the comprehensive efficiency coefficient = skill level coefficient × age structure coefficient × regional adaptation coefficient.

[0026] A dynamic and precise staffing system for power supply enterprises based on multi-dimensional data correlation includes:

[0027] Data acquisition and processing module: Deployed in the power supply enterprise's data platform, it collects multi-source heterogeneous data from the power supply enterprise, performs differentiated cleaning and structured processing on the multi-source heterogeneous data according to data type, and converts the data into a unified structured format to obtain a standardized dataset;

[0028] Association Modeling Module: Based on features extracted from standardized datasets, and through linear and nonlinear association mining, knowledge graph construction and structural dimension adjustment, and fusion with DS evidence theory, a dynamic association model between equipment and business is generated;

[0029] The working time calculation module includes sub-modules for calculating standard time, non-standard time, and travel time. It constructs a multi-dimensional working time calculation model based on standardized datasets and equipment-business dynamic correlation models to calculate standard time, non-standard time, and travel time.

[0030] Staffing Calculation Module: Based on quota time, non-quota time and travel time, calculate the total annual working hours, and combine the average effective working time per person per year to calculate the theoretical staffing.

[0031] Dynamic optimization module: Enables data updates and incremental model training, monitors load deviations and triggers dynamic adjustments, and outputs a dynamic and precise staffing plan for the power supply company;

[0032] Output display module: Connects to the power supply company's human resource management system to output dynamic and accurate staffing plans.

[0033] The beneficial effects of this invention are as follows:

[0034] Significantly improved staffing accuracy: By constructing a full-chain connection between "equipment-business-personnel", the gap in the intermediary links of traditional methods is filled. Combined with GIS technology to specifically calculate travel time and incorporate multi-dimensional algorithm models, the staffing error is reduced to less than 8% compared with traditional methods, solving the one-sided problem of directly mapping equipment to personnel.

[0035] Significantly enhanced dynamic adaptability: A dynamic iteration mechanism with monthly data updates and quarterly incremental model training has been established, which can respond in real time to technological changes such as smart grid construction, equipment upgrades, and new energy grid connection. The 1.2 times reserve mechanism for sudden business has shortened the emergency response time from the traditional 7 days to within 72 hours.

[0036] Tailored to the actual business needs of power supply companies: Specifically adapted to the characteristics of multiple disciplines such as power transmission, substation, and distribution, and incorporating industry standards such as the "Electric Power Safety Work Regulations". The travel time calculation is tailored to the geographical distribution of equipment and the characteristics of the work area, improving the feasibility of implementation by 60% compared to the general staffing method;

[0037] Significant benefits of human resource optimization: By calibrating the comprehensive efficiency coefficients of skill level, age structure, and regional suitability, precise matching of people and positions can be achieved, which can improve the efficiency of human resource allocation in power supply companies by more than 15%, effectively reducing the cost of redundant personnel or the risk of emergency personnel shortages.

[0038] Balancing interpretability and compliance: The knowledge graph integrates industry standards and corporate procedures, and the staffing results are accompanied by analysis reports on working hours composition, equipment-business correlation coefficients, and efficiency impact ratios, meeting the compliance audit requirements of power supply companies. Staffing verification passes industry benchmark testing and joint review by multiple departments, ensuring the rationality of the solution. Attached Figure Description

[0039] Figure 1 This is an overall flowchart of the dynamic and precise staffing method for power supply enterprises according to an embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram illustrating the construction of the device-service dynamic association model according to an embodiment of the present invention;

[0041] Figure 3 This is a schematic diagram of the structure of the multi-dimensional working time calculation model in an embodiment of the present invention. Detailed Implementation

[0042] Example 1: An embodiment of the present invention provides a method for dynamic and precise staffing of power supply enterprises based on multi-dimensional data correlation, such as... Figure 1 As shown, it includes the following steps:

[0043] S1. Collect multi-source heterogeneous data from power supply enterprises, including equipment basic data, business operation data, work order process data, work record data, working hours and personnel data, and procedure and standard data. Use ETL tools to extract, transform, and load the data, and then aggregate the processed multi-source heterogeneous data into the power supply enterprise's data platform.

[0044] The collected multi-source heterogeneous data covers all dimensions of "equipment-business-personnel-environment-procedures", ensuring that the data is closely aligned with the production and service characteristics of power supply companies.

[0045] Basic equipment data: ledger data and geographic information data (latitude and longitude, work area) of substations (capacity, voltage level, years of operation, intelligence level), transmission lines (length, model, number of towers, insulation level), service areas (number, user type, load characteristics), and power distribution equipment (number and distribution of switches and transformers) within the jurisdiction of the power supply company.

[0046] Operational data: Data on routine operations (transmission line inspection, substation maintenance, distribution line operation and maintenance, marketing and meter reading and billing, etc.) and emergency operations (fault repair, major power supply protection, emergency response, etc.) carried out by each specialty in the past three years, including information on business frequency, workload, and completion quality;

[0047] Work order process data: Work order data from the power supply service command system (distribution network emergency repair command platform) and the marketing business application system, including full-process information such as work order initiation time, response time, processing steps, completion time, and personnel involved;

[0048] Work record data: Work logs, work tickets, operation tickets, and pre-shift and post-shift meeting records of each professional team, including detailed data such as operators, work content, start / end time, and collaborating personnel;

[0049] Working hours and personnel data: Data on working hours (including quota / non-quota time) of various business operations collected through working hour interviews, work records, etc., as well as human resource data such as employee skill level (junior / intermediate / senior / technician), age structure, training records, job qualifications, etc.

[0050] Regulations and guidelines data: "Power Supply Labor Quota" (LD / T46-2016), production professional regulations (such as "Power Safety Work Regulations"), marketing service guidelines (such as "Power Supply Service Specifications"), and enterprise internal staffing management rules, etc.

[0051] ETL tools perform transformation operations including data type conversion, missing value imputation, data standardization, and data association and fusion. Numeric fields are imputed using the mean, categorical fields using the mode, and numeric data is normalized to the [0,1] interval using min-max normalization. Data from multiple tables is then joined and fused using primary keys. The specific ETL operation process and core formulas are as follows:

[0052] Extraction: Raw data is extracted from data sources such as PMS system, marketing system, and GIS system using methods such as direct database connection (JDBC / ODBC), file parsing (CSV / Excel), and API calls. The data is then categorized and stored in a temporary data layer according to a three-level directory structure of "data source - data table - field".

[0053] Transformation: The core processing step, including data cleaning, standardization, and correlation fusion. Specific formulas and operations are as follows:

[0054] (1) Data type conversion: Convert text-based device numbers, dates, etc., to standard formats, such as unifying the date format to "YYYY-MM-DDHH:MM:SS", the formula is: In this context, Date is the original date string, which is then reconstructed by extracting the year, month, day, hour, minute, and second using regular expressions.

[0055] (2) Filling missing values: For numeric fields (such as equipment capacity, working hours consumption), the mean value filling method is used, and the formula is:

[0056] ;

[0057] in, This is the total number of records in the field. This represents the number of missing records. For non-missing field values;

[0058] For categorization fields (such as device type), the mode fill method is used, and the formula is:

[0059] ;

[0060] in, For the set of all possible values ​​for the category field, For taking values Frequency of occurrence.

[0061] (3) Data standardization: Numerical data is normalized using min-max normalization and mapped to the [0,1] interval. The formula is as follows:

[0062] ;

[0063] in, The maximum value of the field. The minimum value of the field.

[0064] (4) Data Association and Fusion: Data from multiple tables is merged through primary key association (such as device ID, work order ID), using the following formula:

[0065] ;

[0066] in, Use the primary key (such as device ID) and employ inner joins to retain matching data from both sides.

[0067] Loading: The transformed, standardized data is loaded in batches into the target data warehouse (such as Hive or Oracle). A partitioned storage strategy is used (partitioned by "year-major"). Transaction control is used during the loading process to ensure data consistency. The batch size is set to 100 records per batch to improve efficiency. The formula is:

[0068] ;

[0069] S2. Perform differentiated cleaning and structuring on the processed multi-source heterogeneous data according to data type, and convert the data into a unified structured format to obtain a standardized dataset.

[0070] Data types include numerical data, text data, process data, and geographic information data.

[0071] Label each data entry with unique attributes: Equipment data labeling: Equipment type, related discipline (transmission / substation / distribution), geographical coordinates, years of operation, intelligence level; Business data labeling: Business type (routine / emergency), related discipline, associated equipment ID, workload (e.g., inspection mileage, number of units repaired), operation date, weather conditions (outdoor operation); Working hour data labeling: Quota time (preparation + operation + completion), non-quota time (waiting + delay + distance), time consumption type (routine business / emergency repair business).

[0072] Differential data cleaning includes: removing outliers from numerical data using the 3σ principle and combining it with secondary validation based on industry standards; specifically: calculating the mean μ and standard deviation σ of the fields.

[0073] ;

[0074] ;

[0075] Set an outlier threshold: ,when When this happens, it is considered an outlier;

[0076] Secondary verification: A reasonable range is set based on the power supply company's industry standards (such as the "Construction Period Quota for Power Engineering"). If the abnormal value is within the reasonable range of the industry, it is retained; otherwise, it is discarded.

[0077] Text-based data was segmented and semantically extracted using a BERT-based power supply-specific dictionary, removing redundant information with a cosine similarity < 0.6; specifically:

[0078] Construction of a Power Supply Professional Dictionary: This involves collecting power supply-specific terms such as "two-ticket three-system," "live-line working," and "intelligent inspection" to build a dictionary. Assign weights to each term (Based on the frequency of the terminology in the procedure);

[0079] BERT word vector generation: Tokenization is performed on the text data, converting each character into a vocabulary index idxi, and word vectors are generated through the BERT embedding layer. ( (Assuming the vector dimension is 768), the formula is:

[0080] ;

[0081] in, Positional encoding is used to preserve text sequence information; semantic filtering: the cosine similarity between word vectors and term vectors from a specialized dictionary is calculated using the following formula:

[0082] ;

[0083] when If such information is deemed redundant (e.g., irrelevant security warnings), it will be deleted.

[0084] For process-oriented data, ProM tools were used to identify and remove invalid work orders with a processing time of <10s or more than 3 steps; specifically:

[0085] Event log parsing: Converting work order data into a ProM-compatible event log format:

[0086] ;

[0087] in, For work order number, For the work process, For the time stamp of the process;

[0088] Calculation of workflow time: For each work order's continuous operation steps, calculate the workflow time:

[0089] ;

[0090] Invalid work order identification: When the work order processing time is... (Test work order) or When a work order is submitted repeatedly, it will be deemed invalid and removed.

[0091] Geographic information data (equipment distribution, work area location) is transformed using Mercator projection, and coordinate offset is corrected using the least squares method to ensure that the corrected coordinate error is ≤50 meters; specifically:

[0092] Coordinate transformation: converting latitude and longitude coordinates Convert to Cartesian coordinates The Mercator projection formula is used:

[0093] ;

[0094] ;

[0095] in, The radius of the Earth is taken as 6,378,137 m.

[0096] Error correction: The least squares method is used to correct the coordinate offset, and the formula is as follows:

[0097] ;

[0098] ;

[0099] in, , The offset correction amount is calculated by matching with the standard map in the GIS system, ensuring that the corrected coordinate error is ≤50 meters.

[0100] Structured processing: The cleaned data is converted into a unified format: Numerical data: min-max normalization has been completed through the ETL process; Categorical data (business type, skill level): one-hot encoding is used, with the formula as follows: Where 1 corresponds to the index of the categorical variable; text data (procedure clauses): has been converted to 768-dimensional word vectors using BERT; geographic data: has been converted to GIS-compatible Cartesian coordinate format to generate a standardized dataset.

[0101] S3, such as Figure 2 As shown, using a standardized dataset, the equipment and business characteristics of power supply companies are extracted. Through linear and nonlinear association mining, knowledge graph construction and structural dimension adjustment, and fusion with DS evidence theory, a dynamic association model between equipment and business is obtained.

[0102] Equipment-business linear correlation mining is achieved by calculating the Pearson correlation coefficient, with a strong linear correlation defined as |r|≥0.6. Nonlinear correlation mining is achieved by constructing a LightGBM regression model. Knowledge graph construction includes core nodes for design equipment, business, procedures, and constraints, defining "association impact" node relationships and assigning initial weights. Inference rules are defined in conjunction with power supply company procedures and guidelines, and node association weights are updated using a neural symbolic inference framework. Structural dimension adjustment uses the ARIMA model and AIC criteria to determine optimal parameters (p,d,q), predicts changes in business volume brought about by equipment upgrades and maintenance plans, and generates structural adjustment coefficients. DS evidence theory integrates the quantitative dimension calibration coefficient and the structural adjustment coefficient; fusion is effective when the conflict coefficient K<0.8. The specific process described above is as follows:

[0103] 3.1 Quantitative Dimension Correlation Coefficient Calculation: Based on the core logic of power supply companies that "equipment determines business volume," a correlation model between equipment quantity and business volume is constructed using a combination of statistical analysis and machine learning.

[0104] 3.1.1 Feature Extraction: Extracting device features from a standardized dataset. (Equipment quantity, capacity, and years of operation), business characteristics (Annual frequency, task volume) Construct "device-business" feature pairs .

[0105] 3.1.2 Calculation of Pearson correlation coefficient (linear correlation degree):

[0106] The formula is: ;

[0107] in, For equipment characteristics The mean, For business characteristics The mean, It was determined to be a strong linear association.

[0108] 3.1.3 LightGBM Nonlinear Association Mining: Constructing a LightGBM regression model, inputting device features Output business characteristics The core formula is as follows: Objective function (mean squared error):

[0109] ;

[0110] in, These are the model's predicted values. These are the model parameters. The splitting criterion (GOSS sampling) is as follows: Samples are sorted by absolute gradient value, and the top 20% of high-gradient samples and the top 80% of low-gradient samples are randomly selected as the remaining 10%. The information gain is then calculated.

[0111] ;

[0112] in, The entropy value. , This represents the number of samples in the left and right subtrees. The nonlinear correlation coefficients output by the model training are then fused with the Pearson correlation coefficient to generate the device-business basis correlation coefficient.

[0113] 3.2 Knowledge Graph Construction and Reasoning Calibration: Combining the power supply enterprise's regulations and guidelines, a multi-granularity logical reasoning knowledge graph is constructed to achieve regularized calibration of correlation coefficients.

[0114] 3.2.1 Core Node Design: This includes equipment-related nodes (transmission lines, smart substations), business-related nodes (inspection, maintenance, emergency repair), procedure-related nodes (LD / T 46-2016, safety regulations), and constraint-related nodes (weather conditions, equipment level).

[0115] 3.2.2 Definition of Relationships and Rules: The relationship between nodes is defined as “association influence” and assigned an initial weight (based on the basic correlation coefficient in 3.1). Inference rules are defined (such as “220kV transmission line + rainstorm weather → patrol workload increases by 20%”, “smart substation + remote monitoring → duty workload decreases by 30%”).

[0116] 3.2.3 Dynamic Weight Update: Using a neural symbolic reasoning framework and real-time business data, the inference error is calculated, and an adaptive gradient descent algorithm is used to update the node association weights, generating a set of association coefficients after regular calibration.

[0117] 3.3 Structural Dimension Adjustment and Optimization: Combining the characteristics of power supply enterprise equipment lifecycle management, structured factors such as equipment renewal and maintenance plans are introduced to adjust the correlation coefficients:

[0118] 3.3.1 Time Series Analysis (ARIMA Model) Forecasting: Construct an ARIMA(p,d,q) model and input historical business volume data. The core formula for predicting the impact of structured factors on business volume is as follows:

[0119] Difference processing: for non-stationary sequences Performing d-th order differencing yields a stationary sequence. ;

[0120] AR(p) model:

[0121] ;

[0122] in These are the autoregressive coefficients. It is white noise;

[0123] MA(q) model:

[0124] ;

[0125] in The moving average coefficient is... This represents the error term. The optimal parameters (p, d, q) are determined using the AIC criterion to predict changes in business volume resulting from annual equipment upgrades and maintenance plans, generating structural adjustment coefficients.

[0126] 3.3.2 Fusion of DS Evidence Theory: This involves fusing the quantitative dimension calibration coefficient and the structural adjustment coefficient. The core formula is as follows: Basic Probability Allocation (BPA): Given two sources of evidence, E1 (quantitative dimension coefficient) and E2 (structural adjustment coefficient), assign BPA to the set of correlation coefficients A: ;

[0127] Synthesis rules:

[0128] ;

[0129] in, The conflict coefficient is K, which is effective when K < 0.8. The final dynamic correlation coefficient set between "devices and services" is generated through fusion, enabling accurate estimation of "device quantity → service quantity".

[0130] S4, such as Figure 3 As shown, a multi-dimensional working time calculation model is constructed based on a standardized dataset and a dynamic correlation model between equipment and business, to calculate the standard time and non-standard time of working hours respectively.

[0131] The multi-dimensional working hour calculation model includes core features: business type (routine / emergency repair), associated equipment features (intelligence level, depreciation rate), personnel features (skill level, operational proficiency), and environmental features (outdoor / indoor, weather, geographical region); and special features for emergency business: fault level (general / major / extremely major), response priority, and nighttime operation identifier.

[0132] Quota time = preparation time + operation time + completion time, calculated using a combination of statistical analysis and machine learning.

[0133] Preparation time: The average time spent on each preparation stage (safety briefing, tool requisition) is calculated based on historical work order data. The adjustment coefficient α is assigned based on the skill level (α=0.8 for senior technicians / master technicians, α=1.0 for intermediate technicians, and α=1.2 for junior technicians), and the formula is as follows: ;

[0134] Task time: Using the feature system as input, a random forest model is used for training. Input feature vectors Output job time prediction value The model uses the mean squared error loss function: ;

[0135] Closure time: Average time spent based on records such as work order archiving and tool return. An adjustment coefficient β is assigned based on the complexity of the business (β=1.3 for emergency repairs, β=1.0 for routine operations), and the formula is as follows: ;

[0136] Integrated calibration: quota time If the operation meets the requirements of the "Electric Power Safety Work Regulations" by verifying the rules of the knowledge graph, then... If the scope exceeds the regulations, it shall be adjusted according to the upper / lower limits of the regulations.

[0137] Non-quota time = unavoidable waiting time + human-caused delay time + travel time. Unavoidable waiting time: the average time spent in power supply company-specific waiting scenarios (such as power outage permit approval, dispatch instruction issuance, and material delivery delays). Assign weights according to business type The formula is: in, Based on the frequency of occurrence of this service waiting scenario (such as emergency repair service) =1.2, routine business =1.0); Human-caused delays: Delay data caused by insufficient skills, rework, operational errors, etc., are marked, and outliers are identified using the Isolation Forest algorithm. The formula is:

[0138] ;

[0139] in, For the sample Path length, For indicator functions, when Values ​​that are identified as human-caused delays are removed.

[0140] To verify the formula calculation, the mean absolute error (MAE) of the test set must be ≤5%. The formula is as follows:

[0141] ;

[0142] If the conditions are not met, return to construct a multi-dimensional working time calculation model to supplement features (such as adding the feature of "frequency of work team collaboration").

[0143] S5. Based on GIS technology, preprocess the geographic information data in the equipment basic data, plan the route and calculate the distance, calibrate the travel speed according to the geographic region and weather conditions, calculate the single trip and the total annual travel time, and obtain the travel time.

[0144] Travel time is calculated based on GIS technology, including geographic data preprocessing, optimal route planning, traffic speed calibration, and annual travel time summary.

[0145] 5.1 Geographic Data Preprocessing and Layer Construction: The geographic coordinates (latitude and longitude) of production areas, substations, transmission line towers, and transformer substations are converted into Cartesian coordinates using the Mercator projection formula. The formula is the same as step 2; construct a "work area-equipment" spatial layer in the GIS system, classify and label the equipment according to its professional category (transmission / transformation / distribution) and geographical region (urban / suburban / mountainous area), and generate a spatial index:

[0146] ;

[0147] 5.2 Route Planning and Distance Calculation: For each task, the optimal route is planned based on GIS spatial analysis algorithms, prioritizing highway routes and avoiding restricted areas (such as military control zones and inaccessible mountainous areas). The route planning formula is as follows:

[0148] ;

[0149] in, For route length, The route delay time (based on road conditions) is calculated using the Haversine formula to determine the straight-line distance between each node on the route. The formula is as follows:

[0150] ;

[0151] in, R is the Earth's radius (6378137m).

[0152] For non-straight routes (such as highways), multiply by a route correction factor γ (γ=1.1 in urban areas, γ=1.2 in suburban areas, and γ=1.5 in mountainous areas) to obtain the actual route distance. .

[0153] 5.3. Speed ​​Setting and Single-Trip Time Calculation: Set a baseline speed according to the geographical area. (Urban Area) =40km / h, suburbs =30km / h, mountainous area =20km / h), and the speed is corrected according to weather conditions, using the following formula:

[0154] ;

[0155] in, Weather correction factor (sunny day) =1.0, Rainy Day =0.8, Snowy day =0.6);

[0156] The formula for calculating the time for a single trip (round trip) is: ,in, The round trip distance (km) The corrected travel speed (km / h) is converted to minutes.

[0157] 5.4 Annual Travel Time Summary: Statistics on the frequency of business operations conducted by each specialty throughout the year. (For example, if a work area conducts 20 maintenance visits per substation out of 10 substations annually), the formula is:

[0158] ;

[0159] in, This represents the total number of devices associated with this specialty. Annual business frequency for a single device;

[0160] The formula for calculating the total annual travel time for each major is as follows:

[0161] ;

[0162] in, This represents the number of business types within this specialty. For a single type of business, the single journey time. This refers to the annual frequency of this type of business.

[0163] S6. Based on the quota time, non-quota time and travel time, calculate the total annual working hours, introduce three types of adjustment coefficients (skill, age, and region) to calculate the comprehensive efficiency coefficient, and combine the national regulations on the average annual effective working time per person to calculate the theoretical number of employees and obtain the staffing plan for the power supply company.

[0164] Annual total working hours = Annual total standard working hours + Annual total unavoidable waiting time + Annual total travel time + Emergency business reserve working hours. The emergency business reserve working hours are 1.2 times the average annual working hours for emergency business over the past three years. The formula is:

[0165] ;

[0166] Annual total standard working hours ;

[0167] Waiting time is always unavoidable throughout the year. ;

[0168] Total travel time per year ;

[0169] Reserve working hours for unexpected business The average annual working hours for unexpected business events over the past three years are multiplied by a factor of 1.2 to cope with extreme situations.

[0170] Taking into account the human resource characteristics of power supply enterprises, three types of adjustment coefficients are introduced, and the comprehensive efficiency coefficient is calculated using the weighted product method. The formula is as follows:

[0171] ;

[0172] Among them, skill level coefficient Senior technician / master craftsman 1.0, intermediate technician 1.1, junior technician 1.3; Age structure coefficient Regional fit coefficient: 0.95 for ages 30-45, 1.0 for ages 25-30 and 45-50, and 1.05 for ages under 25 and over 50; Urban work area: 0.95, suburban work area: 1.0, mountainous work area: 1.1.

[0173] The national standard stipulates the average annual effective working hours per person. The theoretical staffing level is 2000 hours, calculated using a rounding function, with the following formula:

[0174] ;

[0175] in, This is a rounding function (to ensure that personnel meet the minimum workload requirements).

[0176] Staffing quota verification and tool operation process: Import total working hours, efficiency coefficients, and other data into SPSS. Use an independent samples t-test to verify the significance of the difference between the theoretical staffing quota and the industry benchmark (LD / T 46-2016). This meets industry requirements; a "Work Hour Composition - Staffing" relationship chart is generated using a Python visualization tool (Matplotlib) and submitted to the Human Resources Department and Production Department for joint review; if the review is approved, the final staffing plan is output; if it is not approved, the process is returned to adjust the efficiency coefficient or step 3 to adjust the equipment-business relationship coefficient.

[0177] S7. Establish a dynamic and precise staffing plan for power supply companies: Update multi-source heterogeneous data monthly according to actual conditions, incrementally train the equipment-business dynamic correlation model and multi-dimensional working hour calculation model quarterly, monitor the load deviation between the power supply company's staffing plan and the actual configuration, complete the verification through industry benchmark testing and joint review by multiple departments, and trigger dynamic optimization and adjustment when the deviation is >10%.

[0178] The incremental training sample accounts for no less than 20% of the total sample, and the load bias is the relative error between the actual average annual working hours per person and the theoretical average annual working hours per person.

[0179] The industry benchmark test used an independent sample test to verify the significant difference between the theoretical staffing and the "Power Supply Staffing" (LD / T46-2016).

[0180] The formula for calculating the load deviation Δ between the staffing plan and the actual staffing is:

[0181] ;

[0182] in, The actual average annual working hours per person. The theoretical average annual working hours per person; when When the deviation is caused by a new business type (such as new energy grid connection operation and maintenance), the features are supplemented and the knowledge graph is updated; if the deviation is caused by changes in personnel efficiency, the comprehensive efficiency coefficient η is adjusted.

[0183] Embodiments of the present invention also provide a dynamic and precise staffing system for power supply enterprises based on multi-dimensional data correlation, comprising:

[0184] Data acquisition and processing module: Deployed in the power supply enterprise's data platform, it collects multi-source heterogeneous data from the power supply enterprise, performs ETL processing on the multi-source heterogeneous data, and then obtains a standardized dataset through differential cleaning and structured processing.

[0185] Association Modeling Module: Based on features extracted from standardized datasets, a dynamic association model between equipment and business is generated through linear and nonlinear association mining, knowledge graph calibration, structural dimension adjustment, and DS evidence theory fusion.

[0186] The working time calculation module includes sub-modules for calculating standard time, non-standard time, and travel time. It constructs a multi-dimensional working time calculation model based on standardized datasets and equipment-business dynamic correlation models to calculate standard time, non-standard time, and travel time in all dimensions.

[0187] Staffing Calculation Module: Based on Python and SPSS tools, it calculates the total annual working hours, calculates the comprehensive efficiency coefficient, and obtains the theoretical staffing number by combining the average annual effective working time per person.

[0188] Dynamic optimization module: Enables monthly data updates and quarterly incremental model training, monitors load deviations and triggers adjustments, and works with the verification process to obtain a dynamic and accurate staffing plan for the power supply company;

[0189] Output and Display Module: Connects to the power supply company's human resource management system to output dynamic and accurate staffing plans, as well as analysis reports on working hours composition and the proportion of efficiency impact.

[0190] Example 2: This example combines a practical application case of a county-level power supply company to further illustrate the technical solution of the present invention. The company covers three core production specialties: power transmission, substation, and power distribution. Its service area includes urban areas, suburbs, and mountainous terrains. It currently has 10 substations, 500 kilometers of transmission lines, and 200 distribution substations.

[0191] Step 1: Multi-source data acquisition and processing.

[0192] Equipment data: ledgers of 10 substations (including 2 smart substations), geographical coordinates of 500 kilometers of transmission lines (including 100 kilometers of mountain lines), and distribution data of 200 distribution areas.

[0193] Business data: Records of routine business (4,500 power transmission inspections, 1,200 substation maintenances, and 3,800 power distribution maintenances) and emergency business (1,200 fault repairs and 150 power supply protection tasks) over the past three years.

[0194] Work order and work hour data: 12,000 work order data from the power distribution network emergency repair command platform, and 12,000 real-time work hour data and skill levels of 300 employees (80 senior workers, 150 intermediate workers, and 70 junior workers).

[0195] Regulations and data: LD / T46-2016, "Electric Power Safety Work Regulations (Substation Section)" and other text data.

[0196] ETL processing: Data was extracted using the Talend tool and processed through formulas. Standardized numerical fields such as equipment capacity and working hours are integrated from multiple tables through primary key association and loaded into the Hive data warehouse for storage by partitioning into "2023-Power Transmission" and "2023-Power Transformation".

[0197] Numerical data: Calculation of average working hours for power transmission line inspection Hours, standard deviation Remove 230 abnormal data entries outside the scope.

[0198] Text data: The procedure text word vectors were generated using BERT, and 180 redundant pieces of information with a similarity of <0.6 to professional dictionaries were removed.

[0199] Process data: 120 test orders with a processing time of less than 10 seconds were removed using the ProM tool.

[0200] Geographic data: Coordinates were converted using Mercator projection to correct offset errors to within 30 meters, resulting in a standardized dataset of 11,340 records.

[0201] Step 2: Build a dynamic association model between devices and services.

[0202] Quantitative correlation: The Pearson correlation coefficient between transmission line length and inspection workload was calculated. The LightGBM model was used to find a nonlinear correlation coefficient of -0.2 between the number of smart substations and the maintenance workload.

[0203] Knowledge graph construction: Create nodes “220kV transmission line”, “intelligent substation”, “inspection business” and “safety procedure” in Neo4j, and define the rule “mountain transmission line → inspection business volume increases by 15%”.

[0204] Structural adjustment: Construct an ARIMA(2,1,1) model to predict that the annual renovation of 50 kilometers of old lines will lead to a 10% decrease in maintenance business volume in the following year, generating a structural adjustment coefficient of 0.9.

[0205] Fusion calibration: By fusing the quantitative dimension and structural adjustment coefficient through DS evidence theory, the final correlation coefficient of "transmission line-inspection" is 0.81.

[0206] Step 3: Accurate working time calculation based on machine learning.

[0207] Quota time: Statistical preparation time for power transmission line inspection Hours, closing time The LightGBM model predicts a job time of 4.2 hours, with an adjustment factor for advanced technicians. Quota time Hour.

[0208] Non-fixed time: unavoidable waiting time Hourly / time, 120 data entries were removed due to human error.

[0209] Step 4: Calculate travel time.

[0210] Convert the coordinates of the urban work area and the substation to Cartesian coordinates, calculate the straight-line distance of 10km from the urban work area to the substation, and apply the highway correction factor. The actual distance is 11km, the base speed in urban areas is 40km / h, and the correction factor is for sunny days. Corrected speed 40km / h, one-way travel time Minutes, 800 patrols per year, total annual travel time = 33 × 800 = 26400 minutes = 440 hours.

[0211] Step 5: Staffing calculation and optimization.

[0212] Total working hours: Hour, Hour, Hour, The total annual working hours are 46550 + 5700 + 440 + 1200 = 53890 hours.

[0213] Efficiency coefficient: Combining skills, age, and regional factors, average. Staffing quota: The number of employees was determined to be 27 after SPSS testing and compliance with the LD / T46-2016 standard, and the final staffing was determined after manual review.

[0214] Step 6: Dynamic optimization and update.

[0215] Assuming that three new smart substations are added in six months, the correlation coefficient between "substation-maintenance" will be updated to -0.25 after incremental training of the model, the staffing will be adjusted to 26 people, and the deviation will be controlled within 3%.

[0216] Those skilled in the art will understand that the above embodiments can adjust the data dimensions and model parameters according to the number of professional departments of the power supply company and the regional characteristics. For example, when adding marketing professionals, the number of customers and data on meter reading and billing can be supplemented, without departing from the core technology of the present invention.

[0217] This invention has been described through embodiments. Those skilled in the art will understand that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of this invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, this invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of this invention.

Claims

1. A method for dynamic and precise staffing of power supply enterprises based on multi-dimensional data correlation, characterized in that, Includes the following steps: S1. Collect multi-source heterogeneous data from power supply enterprises, including equipment basic data, business operation data, work order process data, work record data, working hours and personnel data, and procedures and guidelines data; S2. Perform differentiated cleaning and structuring processing on multi-source heterogeneous data according to data type, and convert the data into a unified structured format to obtain a standardized dataset; Data types include numerical data, text data, workflow data, and geographic information data; differential cleaning includes: Numerical data is processed using the 3σ principle to remove outliers and then further validated using industry standards. Text-based data is segmented and semantically extracted using a power supply-specific dictionary based on BERT, and redundant information with cosine similarity that does not meet the set value is removed. For process-oriented data, the ProM tool is used to mine and remove invalid work orders with short processing times or many steps. Geographic information data undergoes coordinate transformation via Mercator projection, and coordinate offset is corrected using the least squares method. The corrected coordinate error is less than a set threshold. S3. Based on standardized datasets, feature data is extracted, and through linear and nonlinear association mining, knowledge graph construction and structural dimension adjustment, and DS evidence theory fusion, a dynamic association model between equipment and business is obtained. The linear correlation mining between equipment and business is achieved by calculating the Pearson correlation coefficient, while the nonlinear correlation mining is achieved by constructing a LightGBM regression model. The nonlinear correlation coefficient is output through model training and fused with the Pearson correlation coefficient to generate the basic correlation coefficient between equipment and business. The knowledge graph construction includes designing core nodes for equipment, business, procedures, and constraints; defining the relationships between nodes and assigning initial weights; defining inference rules in conjunction with the power supply company's procedures and guidelines; calculating inference errors using a neural symbolic inference framework and real-time business data; updating node association weights using an adaptive gradient descent algorithm; and generating regularized and calibrated association coefficients. Structural dimension adjustment uses the ARIMA model and the AIC criterion to determine the optimal parameters, predicts changes in business volume caused by equipment upgrades and maintenance plans, and generates structural adjustment coefficients. The DS evidence theory integrates quantitative dimension calibration coefficients and structural adjustment coefficients. The integration is effective when the conflict coefficient meets the set value. The final device-business dynamic correlation coefficient is generated through the integration. S4. Based on standardized datasets and equipment-business dynamic correlation models, construct a multi-dimensional working time calculation model to calculate the standard time and non-standard time of working hours respectively. S5. Based on multi-source heterogeneous data, combined with GIS spatial analysis algorithms, the optimal highway route is planned. The Haversine formula is used to calculate the straight-line distance between each node of the route. For non-straight-line routes, the route correction coefficient is multiplied to obtain the actual route distance. The benchmark speed is set according to the geographical region, and the corrected speed is obtained by multiplying the speed by the weather correction coefficient in combination with the weather conditions. Finally, the round-trip travel time for a single operation is calculated, and the total annual travel time is obtained by combining the annual operation frequency. S6. Based on the quota time, non-quota time and travel time, calculate the total annual working hours, and combine the annual average effective working time per person to calculate the theoretical number of employees and obtain the staffing plan for the power supply company.

2. The method for dynamic and precise staffing of power supply enterprises based on multi-dimensional data association as described in claim 1, characterized in that, It also includes step S7, establishing a dynamic and precise staffing plan for power supply companies: updating multi-source heterogeneous data according to actual conditions, incrementally training the equipment-business dynamic correlation model and the multi-dimensional working hour calculation model, monitoring the load deviation between the power supply company's staffing plan and the actual configuration, and triggering dynamic optimization and adjustment when the deviation exceeds the set threshold.

3. The method for dynamic and precise staffing of power supply enterprises based on multi-dimensional data association as described in claim 1, characterized in that, In step S1, the extraction, transformation and loading of multi-source heterogeneous data are completed by ETL tools. The transformation specifically includes data type conversion, missing value imputation, data standardization and data association and fusion. Among them, the mean imputation method is used for numerical fields, the mode imputation method is used for categorical fields, and the numeric data is mapped to the [0,1] interval using min-max normalization.

4. The method for dynamic and precise staffing of power supply enterprises based on multi-dimensional data association as described in claim 1, characterized in that, In step S4, the standard time = preparation time + operation time + completion time, and the non-standard time = unavoidable waiting time + human-caused delay time + travel time.

5. The method for dynamic and precise staffing of power supply enterprises based on multi-dimensional data association as described in claim 1, characterized in that, In step S5, the route correction coefficients are 1.1 for urban areas, 1.2 for suburbs, and 1.5 for mountainous areas. The base speeds are 40 km / h for urban areas, 30 km / h for suburbs, and 20 km / h for mountainous areas. The weather correction coefficients are 1.0 for sunny days, 0.8 for rainy days, and 0.6 for snowy days.

6. The method for dynamic and precise staffing of power supply enterprises based on multi-dimensional data association as described in claim 1, characterized in that, In step S6, the total annual working hours = total annual standard working hours + total annual unavoidable waiting time + total annual travel time + reserved working hours for unexpected business. Overall efficiency coefficient = skill level coefficient × age structure coefficient × regional adaptation coefficient.

7. A dynamic and precise staffing system for power supply enterprises based on multi-dimensional data association, used to execute the method described in any one of claims 1-6, characterized in that, include: Data acquisition and processing module: Deployed in the power supply enterprise's data platform, it collects multi-source heterogeneous data from the power supply enterprise, performs differentiated cleaning and structured processing on the multi-source heterogeneous data according to data type, and converts the data into a unified structured format to obtain a standardized dataset; Association Modeling Module: Based on features extracted from standardized datasets, and through linear and nonlinear association mining, knowledge graph construction and structural dimension adjustment, and fusion with DS evidence theory, a dynamic association model between equipment and business is generated; The working time calculation module includes sub-modules for calculating standard time, non-standard time, and travel time. It constructs a multi-dimensional working time calculation model based on standardized datasets and equipment-business dynamic correlation models to calculate standard time, non-standard time, and travel time. Staffing Calculation Module: Based on quota time, non-quota time and travel time, calculate the total annual working hours, and combine the average effective working time per person per year to calculate the theoretical staffing. Dynamic optimization module: Enables data updates and incremental model training, monitors load deviations and triggers dynamic adjustments, and outputs a dynamic and precise staffing plan for the power supply company; Output display module: Connects to the power supply company's human resource management system to output dynamic and accurate staffing plans.

Images