An operation and maintenance personnel intelligent deployment method based on power load prediction

Abstract

This invention relates to the field of data processing technology and discloses an intelligent dispatch method for operation and maintenance personnel based on power load forecasting. The method includes collecting multi-source heterogeneous data and performing preprocessing such as cleaning, spatiotemporal alignment, and feature fusion; constructing an operation and maintenance demand intensity map based on the spatiotemporal distribution of load forecasting errors; optimizing collaborative dispatching within responsibility areas to address heterogeneous personnel skills and the risk of sudden load changes; converting personnel deployment plans into structured intelligent dispatching instructions, which are then distributed to operation and maintenance teams via mobile terminals; and achieving closed-loop control of "prediction-deployment-execution-feedback". The beneficial effects of this invention are: it constructs a composite reachability edge weight model that integrates electrical topology adjacency and road network travel time, simultaneously considering the feeder connection relationship of equipment, the degree of separation of switch nodes, and actual road travel time during the division of responsibility areas and the generation of cross-area support constraints, thereby avoiding the problems of weak electrical connections or excessive detour distances within responsibility areas caused by simple geographical division.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically a method for intelligent dispatching of operation and maintenance personnel based on power load forecasting. Background Technology

[0002] With the continuous expansion of the power distribution network and the in-depth advancement of the construction of new power systems, the operation mode of the power grid is becoming increasingly complex and volatile. The access of massive distributed resources such as distributed photovoltaics and electric vehicle charging facilities has significantly increased the uncertainty and volatility of the load side, which places unprecedented demands on the refined allocation of distribution operation and maintenance resources. Traditional operation and maintenance personnel allocation models mostly rely on deterministic load forecast curves, converting the forecast load into the number of personnel required through experience conversion or simple proportional coefficients, and then deploying manpower according to fixed geographical grids or administrative divisions. However, in actual operation, the intensity of operation and maintenance demand does not have a simple linear correspondence with the absolute value of the load. Although the overall load level is low during off-peak periods, the need for handling abnormalities and emergency preparedness of local equipment still objectively exists. While the peak periods and periods of rapid load ramp-up are limited in duration, due to the sharp increase in equipment thermal stress, the significant increase in operation frequency, and the obvious right-skewed and heavy-tailed characteristics of forecast errors in statistical distribution, these periods become the risk windows with the most prominent personnel allocation gaps. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides an intelligent dispatching method for operation and maintenance personnel based on power load forecasting, thereby resolving the problems existing in the prior art.

[0004] To achieve the above objectives, the present invention employs the following technical solution: A method for intelligent dispatching of operation and maintenance personnel based on power load forecasting includes the following steps: S1. Acquire multi-source heterogeneous data, including deterministic load forecast sequences, historical actual load data, equipment status monitoring data, meteorological forecast information, traffic network data, maintenance personnel skills, and historical work order data, and perform preprocessing such as cleaning, spatiotemporal alignment, and feature fusion. S2. Integrate load forecast results and their spatiotemporal error distribution, equipment health status assessment, and meteorological early warning information to construct a maintenance demand intensity map reflecting the distribution of maintenance pressure in future periods. S3. Based on the maintenance demand intensity map, combined with maintenance personnel skill level profiles, historical work order processing efficiency, and the upper limit of manpower available in the responsibility area, construct a multi-objective optimization model. Under the constraints of skill matching and response timeliness, achieve cross-responsibility area collaborative personnel deployment and output personnel deployment plans by time period and skill level. S4. Transform the personnel deployment plan into structured intelligent deployment instructions and distribute them to maintenance teams via mobile terminals. Simultaneously, monitor the actual load deviation and work order execution progress in real time. When the deviation exceeds a threshold, trigger a dynamic re-optimization mechanism.

[0005] Furthermore, S2 specifically refers to: S21. Load pattern adaptive correction and demand intensity mapping of load forecasting error: The current period is first classified into a specific load pattern, and then empirical quantiles are extracted from historical error samples of the same pattern to obtain the error correction value as the state changes over time, so that the operation and maintenance demand intensity is more in line with the actual operation risk. S22. By integrating electrical topology adjacency and road network travel distance, a composite reachability relationship between spatial grids is first constructed, and then the responsibility area division results and cross-area support constraints are generated accordingly.

[0006] Furthermore, S3 specifically refers to: S31. Based on the skill level efficiency coefficient and demand drastic change weight, construct a fitness value for evaluating candidate deployment schemes, and convert the fitness value into the light intensity of the firefly algorithm. S32. An improved firefly update with embedded support constraint projection and time consistency constraints performs feasible domain projection immediately after each location update, directly correcting the update result to a legal personnel deployment plan; at the same time, time consistency constraints are added to the individual difference calculation to make personnel arrangement changes between adjacent time periods smoother.

[0007] Furthermore, S21 specifically refers to: S211. Read the deterministic load forecast sequence of all spatial grids throughout the entire scheduling cycle; S212. Perform sliding window analysis on the deterministic load forecast sequence of each spatial grid to obtain the load pattern label for the current time period; S213. Collect historical forecast error samples according to load patterns and calculate the empirical quantile correction value for the corresponding pattern. S214. Add tail amplification correction to the basic quantile correction values ​​of peak mode and ramp mode to obtain the load forecast error correction value. S215, calculate the device importance weight for each spatial grid; S216. Calculate the environmental risk factor for each spatial grid and each time period; S217. The deterministic load forecast value, the load forecast error correction value, the equipment importance weight and the environmental risk factor are integrated to obtain the spatial grid-level operation and maintenance demand intensity. S218. Traverse all spatial grids and all time periods to generate a spatial grid-level operation and maintenance demand intensity time series, and summarize it to form an operation and maintenance demand intensity map.

[0008] Furthermore, S22 specifically refers to: S221. Treat all spatial grids as graph nodes, establish electrical connection relationships between spatial grids, and calculate electrical distances between pairs of spatial grids; S222. Using the center point, main equipment location, or station location of each spatial grid as the road network positioning point, calculate the average travel time between any two spatial grids in the road network. S223. Set a maximum emergency response time threshold. The maximum emergency response time threshold is used to indicate the upper limit of the passage time that allows inclusion in the same responsibility system or allows the establishment of cross-regional rapid support relationships. S224. Calculate the composite reachability edge weight for any two spatial grids; S225. Perform responsibility area division based on the composite reachability edge weights between all spatial grids to obtain the responsibility area label of each spatial grid; S226. Generate a cross-regional support constraint mask matrix based on the responsibility area division results; S227, Output the responsibility area label and cross-region support constraint mask matrix.

[0009] Furthermore, S31 specifically refers to: S311. Aggregate the spatial grid-level operation and maintenance demand intensity into the responsibility area-level operation and maintenance demand intensity. Directly sum the operation and maintenance demand intensity of all spatial grids belonging to the same responsibility area at the same time period to obtain the first... The responsibility area in the first The intensity of maintenance and operation needs at the responsibility area level during a given time period; S312. Establish skill level efficiency coefficients based on historical work order processing data; S313. Define the candidate personnel deployment plan and calculate the actual available operation and maintenance capabilities of the candidate personnel deployment plan in each responsibility area and time period. S314. Calculate the personnel shortage in each responsibility area and time period. For each responsibility area and time period, calculate the difference between the responsibility area-level operation and maintenance demand intensity and the actual available operation and maintenance capacity. If the difference is less than 0, it means that there is no shortage in that responsibility area and time period, and it is directly recorded as 0. If the difference is greater than 0, the difference is recorded as the personnel shortage. S315. Calculate the demand abrupt change weight based on the change range of the intensity of maintenance demand in adjacent time periods at the responsibility area level. S316. Calculate the local garrison cost and cross-regional support cost of the candidate personnel deployment plan. For any candidate personnel deployment plan, accumulate the product of the number of personnel stationed in the local area and the corresponding local garrison unit cost coefficient for each responsibility area, time period, and skill level to obtain the local garrison cost; then accumulate the product of the number of cross-regional support personnel, average support distance, and unit support cost coefficient for each responsibility area, time period, and skill level to obtain the cross-regional support cost. S317. Combine the local garrison cost, cross-regional support cost, and emergency weighted deficit penalty into a fitness value for the candidate personnel deployment plan. The smaller the fitness value, the better the candidate personnel deployment plan. S318. Convert the fitness value into the light intensity of the firefly algorithm.

[0010] Furthermore, S32 specifically refers to: S321. Encode the candidate personnel deployment plan as firefly positions. Arrange the number of personnel stationed in the area in a fixed order for all responsibility areas, all time periods, and all skill levels to form the first part of the code. Then arrange the number of cross-area support personnel on all allowed support paths in a fixed order to form the second part of the code. Finally, combine the two parts to form a complete firefly position vector. S322. Generate an initial firefly population that satisfies the boundary constraints. Set an adjustable upper limit for each responsibility area and each skill level. Generate an initial value of personnel stationed in the area based on the intensity of maintenance needs at the responsibility area level. Randomly generate a small number of initial cross-area support personnel between the responsibility area pairs allowed by the cross-area support constraint mask matrix. Prune the portion that exceeds the adjustable upper limit and set the negative value directly to 0 to obtain an initial firefly population that satisfies the boundary constraints. S323. Calculate the time consistency weighted distance between fireflies. First, apply time smoothing weights to the personnel sequence of each responsibility area and each skill level. Then, calculate the distance between two fireflies. Finally, assign high weights to the differences between adjacent time periods and low weights to the differences between non-adjacent time periods. S324. Move the low-brightness fireflies toward the high-brightness fireflies and add random perturbations. S325. Perform feasible domain projection on the updated candidate deployment plan; S326. Recalculate the fitness value and light intensity of the projected candidate personnel deployment scheme, and update the current optimal candidate personnel deployment scheme. S327. Repeat the position update, feasible region projection and fitness update until the iteration termination condition is met; S328. Output the final personnel deployment results. The final personnel deployment results shall include at least the number of personnel stationed in each responsibility area at each time period, the number of cross-area support personnel between each responsibility area at each time period, the configuration quantity of each skill level, and the cross-area support flow matrix formed therefrom between responsibility areas.

[0011] Furthermore, a cross-regional support constraint mask matrix is ​​generated based on the responsibility area division results, specifically as follows: The elements in the cross-region support constraint mask matrix are denoted as . , used to indicate the first The responsibility area and the first Is it permissible to establish cross-regional support relationships between different responsibility zones? Traverse any two responsibility areas and If there exists at least one pair that belongs to a specific area of ​​responsibility and responsibility area Spatial grids with composite reachable edge weights If it exceeds the preset support threshold, then Set to 1; otherwise, Set to 0.

[0012] Furthermore, a skill level efficiency coefficient is established based on historical work order processing data, specifically as follows: Extract the average completion time, first-time repair rate, and rework rate of similar work orders from the historical work order system, processed by personnel of different skill levels. First, convert the average completion time into processing capacity per unit time. Then, use the first-time repair rate for positive correction and the rework rate for negative correction to obtain the result. The overall processing efficiency of personnel at each skill level is calculated, and then the overall processing efficiency of all skill levels is normalized by comparing it with the baseline skill level to obtain the skill level efficiency coefficient.

[0013] Furthermore, the feasible domain projection for the updated candidate deployment plan is performed as follows: Set all negative numbers to 0, and round all non-integer numbers according to preset rules. Check the cross-region support constraint mask matrix. If the cross-region support constraint mask value between a certain responsibility region pair is 0, then set all cross-region support personnel on that responsibility region pair to 0, indicating that support is not allowed to occur on this path. For each responsibility area, each time period, and each skill level, the sum of the number of personnel stationed in the area and the total number of support personnel transferred out is calculated. If the calculated value is greater than the maximum number of personnel that can be transferred out for that skill level in the responsibility area, the number of personnel stationed in the area is retained first, and the number of support personnel transferred out is reduced proportionally. If the number of personnel stationed in the area has already exceeded the maximum number of personnel that can be transferred out, the number of personnel transferred out is reduced to the maximum number of personnel that can be transferred out. For each area of ​​responsibility and each time period, the actual number of available personnel after transfers in and out is recalculated. If the number of available personnel in the area is negative, or if there are duplicate transfers from multiple directions causing duplicate statistics or inconsistencies in the total number of personnel, the deployment plan is iteratively adjusted according to the correction order of prioritizing local personnel, prioritizing support from neighboring areas, and then reducing support from distant areas, until all constraints of the deployment plan are met.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. An adaptive error correction mechanism based on load patterns is introduced. Instead of using a fixed safety factor, the empirical quantiles of historical prediction errors are extracted based on four load patterns: low-end, flat-end, peak, and ramp. Tail amplification correction is added specifically for the right-skewed thick-tail characteristics of peak and ramp patterns, so that the intensity of operation and maintenance demand more accurately reflects the actual risk distribution under different operating conditions.

[0015] 2. A composite reachability edge weight model was constructed that integrates electrical topology adjacency and road network travel time. In the process of dividing responsibility areas and generating cross-area support constraints, the feeder connection relationship of the equipment, the degree of separation of switch nodes, and the actual road travel time are considered at the same time. This avoids the problem of weak electrical connection or excessive detour distance within the responsibility area caused by simple geographical division.

[0016] 3. An improved firefly algorithm with embedded cross-regional support constraint mask and feasible region projection operator was designed in the personnel allocation optimization. After each position update, the candidate solution is immediately modified into an executable deployment result that meets the requirements of the upper limit of personnel, support permission relationship and time continuity, which solves the engineering problem of conventional optimization algorithms outputting infeasible solutions.

[0017] 4. An operation and maintenance demand intensity map with equivalent standard working hours as the unified dimension was established. The load forecast value, error correction amount, equipment importance weight, and environmental risk factors under the combined effect of temperature, humidity, wind speed, and load rate were integrated. The ability to suppress the risk of staff shortage during periods of sharp load increase was strengthened through a rapid weight nonlinear penalty mechanism. Attached Figure Description

[0018] Figure 1 This is a flowchart of the present invention; Figure 2 This is a comparison chart showing the consistency between a certain grid's deterministic load forecast and historical actual loads. Figure 3 This is a box plot showing the historical forecast error distribution for each load pattern; Figure 4 It is a spatiotemporal distribution map of the intensity of operation and maintenance demand. Detailed Implementation

[0019] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined in this application.

[0020] like Figure 1 As shown, a method for intelligent dispatching of operation and maintenance personnel based on power load forecasting is disclosed, including the following steps:

[0021] S1. Acquire multi-source heterogeneous data, including deterministic load forecast sequences, historical actual load data, equipment status monitoring data, meteorological forecast information, traffic network data, operation and maintenance personnel skills and historical work order data, and perform cleaning, spatiotemporal alignment and feature fusion preprocessing. To achieve accurate characterization of operation and maintenance needs based on the spatiotemporal distribution of load forecasting errors and subsequent personnel allocation optimization, this step is responsible for comprehensively collecting the required multi-source heterogeneous data from the distribution network operation system, meteorological monitoring platform, geographic information system, and operation and maintenance management database, and performing necessary cleaning, alignment, and fusion preprocessing to provide standardized data input for the subsequent steps of constructing the operation and maintenance demand intensity map and optimizing personnel allocation.

[0022] First, a deterministic load forecast sequence for all spatial grids within the target distribution service area is acquired or generated. This data covers the entire daily scheduling cycle with a time granularity of 15 minutes and is stored in a structured manner according to the spatial grid number, time period number, and predicted load value, directly serving as the basic input for load pattern recognition and forecast error correction.

[0023] In one implementation, when obtaining a deterministic load forecast sequence, if a load forecasting module has been deployed in the distribution network operating environment (e.g., located in the dispatching master station or distribution automation master station), the output forecast results can be directly read through the standard data interface; if no existing interface is available, the system independently calculates and generates a deterministic load forecast sequence based on the historical load time series of each spatial grid collected, using the embedded time series forecasting algorithm (e.g., standard load forecasting models such as ARIMA and LSTM).

[0024] Secondly, historical actual load data with the same spatiotemporal granularity as the load forecast sequence are acquired and organized into a historical actual load sequence according to the same spatiotemporal granularity. This sequence is used to calculate the empirical quantile of historical forecast errors under the same load pattern. Simultaneously, feeder connection tables, switch node ledgers, equipment spatial coordinates, and key equipment lists are exported from the distribution network geographic information system and distribution automation system. These are used to construct the electrical topology adjacency relationships between spatial grids and calculate equipment importance weights.

[0025] In practice, historical load data comes from the load history archives that are inevitably generated and recorded during the operation of the distribution network. For example, it can be extracted from the power acquisition and metering database, the dispatch operation log storage unit, or the operation data history database built by this system.

[0026] Secondly, gridded meteorological forecast information corresponding to the scheduling cycle is obtained, including key elements such as temperature, humidity, and wind speed. This data is spatially interpolated according to the spatial grid to ensure that each grid has corresponding environmental parameters for each time period, which are used to calculate environmental risk factors. In addition, road topology data covering the service area and historical average traffic speeds for different traffic periods need to be extracted from the road network information database to calculate the road travel time between any two spatial grid center points.

[0027] In practice, gridded weather forecast information can come from public meteorological data services, data files released by the power meteorological network, or be generated by a locally deployed meteorological information receiving and processing module, and finally be converted into a sequence of environmental parameters such as temperature, humidity, and wind speed by spatial grid interpolation.

[0028] Finally, we obtain the skill level profiles of the maintenance team members, the processing efficiency statistics of historical maintenance work orders, and the upper limit of personnel that can be deployed in each responsibility area, so as to provide a basis for setting the skill level efficiency coefficient and the feasibility constraints of subsequent personnel deployment.

[0029] In one implementation, personnel skill level information can be determined based on the company's current job qualification certification records or the team personnel qualification ledger; the processing time and repair rate statistics are obtained by grouping completed historical work order records (such as emergency repair work orders, inspection records, and defect handling orders) by skill level and calculating the average time and first-time repair success rate.

[0030] In one embodiment, such as Figure 2 As shown, a comparison is made between the deterministic load forecast and historical actual load for a certain grid, displaying the curves showing the comparison between the deterministic load forecast and historical actual load values ​​for a single spatial grid (#12) within a complete scheduling day. The solid blue line represents the deterministic load forecast sequence output by the upper-level load forecasting system, the dashed red line represents the historical actual load record for the same period, the red shaded area represents positive forecast error (actual value is higher than forecast value), and the green shaded area represents negative error. Time-varying errors are common between forecast and actual values, and the errors are unevenly distributed across different operating periods. The horizontal axis represents scheduling time in hours, and the vertical axis represents power load in megawatts.

[0031] S2. By integrating load forecasting results and their spatiotemporal error distribution, equipment health status assessment and meteorological early warning information, an operation and maintenance demand intensity map reflecting the distribution of operation and maintenance pressure in future periods is constructed. Daily maintenance requirements for distribution networks do not simply correspond linearly to deterministic load forecasts. While load levels are lower during off-peak hours, localized fluctuations and anomaly preparedness still exist. Peak and ramp-up periods, though not necessarily the longest, involve greater thermal stress on equipment, higher operational frequency, and heavier error tails, easily leading to insufficient staffing. The conventional method of directly calculating staffing requirements based on deterministic load forecasts fails to differentiate between different operating states and does not incorporate equipment importance and environmental failure risks into the same input, potentially resulting in excessive staffing during low-risk periods and insufficient staffing during high-risk periods.

[0032] This step first identifies the load patterns of the deterministic load forecast values, then extracts the empirical quantiles of historical forecast errors according to the load patterns, and adds tail amplification corrections for peak and ramp patterns. Next, the error-corrected load results are integrated with equipment importance weights and environmental risk factors to obtain an O&M demand intensity map unfolded by spatial grid and time period, which can be directly used for subsequent responsibility area division and personnel allocation. The specific steps are as follows: S21. Adaptive correction of load pattern and demand intensity mapping for load forecasting errors Traditional methods typically multiply the deterministic load forecast by a fixed safety factor or simply add the historical average error. This approach cannot distinguish between off-peak, flat-peak, peak, and climbing patterns, nor can it reflect the fact that large positive errors are more likely to occur in peak and climbing patterns.

[0033] This step first categorizes the current time period into a specific load pattern, then extracts empirical quantiles from historical error samples of the same pattern to obtain error correction values ​​that reflect changes in the state over time. This makes the intensity of operational and maintenance demands more consistent with actual operational risks. The specific steps are as follows: S211, Divide the target power distribution service area into There are 1 spatial grid, with spatial grid numbering as... ; Divide the intraday scheduling cycle into There are several time periods, each time period numbered as follows: ; The number of responsibility areas is recorded as The responsibility area number is ; The number of personnel skill levels is recorded as The skill level number is .

[0034] Furthermore, the deterministic load forecast sequence of all spatial grids over the entire scheduling cycle is read.

[0035] Specifically, for the first Each spatial grid is used to assemble its deterministic load forecasts for all time periods into a time sequence. ,in Indicates the first The spatial grid in the ... The deterministic load forecast values ​​for the time period come from the higher-level load forecasting system. This sequence serves as input for subsequent load pattern identification, error correction, and operational demand intensity calculation.

[0036] S212. Perform a sliding window analysis on the deterministic load forecast sequence for each spatial grid to obtain the load pattern label for the current time period. The load pattern label is denoted as... , used to indicate the first The spatial grid in the ... Which category does the time period fall under: off-peak, normal, peak, or uphill?

[0037] In practical implementation, a sliding window of length 3 or 5 is constructed centered on the current time period. For example, if a sliding window of length 3 is selected, the window will cover... , , Three time periods.

[0038] In one embodiment, for example, taking a sliding window length of 3 as an example, the window contains , , Deterministic load forecasts for three time periods. Load amplitudes are used directly. (MW). The rate of load change is measured by the absolute mean of adjacent differences within a window. A spatial grid has a deterministic load forecast of 32MW at 18:00, with a daily peak-to-valley difference of 20MW. The 80th percentile for the same period historically is 30MW, and the 20th percentile is 10MW. If the load is 31.5MW at 17:00 and 32.2MW at 19:00, the adjacent differences are 0.5MW and 0.2MW respectively, with an average rate of change of 0.35MW. 0.02 times the peak-to-valley difference of 20MW is 0.4MW. Since the current load of 32MW is higher than 30MW (80th percentile) and the rate of change is 0.35MW < 0.4MW, it is determined to be a peak load pattern.

[0039] Furthermore, the load amplitude and load change rate for the current time period are calculated within this sliding window. The load amplitude is directly taken from the deterministic load forecast for the current time period; the load change rate can be represented by the average absolute value of the differences between adjacent time periods, for example, by the adjacent differences within the window. and Find the average.

[0040] Furthermore, the current time period is determined based on the quantile thresholds obtained from historical samples of the same period. If the deterministic load forecast for the current time period is lower than the 20th percentile of the same period in history for that spatial grid, and the load change rate is less than 0.02 times the intraday peak-to-valley difference of that spatial grid, it is recorded as a low-valley mode; if the deterministic load forecast for the current time period is higher than the 80th percentile of the same period in history, and the load change rate is less than 0.02 times the intraday peak-to-valley difference of that spatial grid, it is recorded as a peak mode; if the load change rate is higher than 0.05 times the intraday peak-to-valley difference of that spatial grid, it is recorded as a climbing mode; all other cases are recorded as a flat-valley mode.

[0041] In one embodiment, as an example, suppose that the deterministic load forecast value of a spatial grid at 18:00 is in the high range of the same period in history, and the adjacent differences between 17:00 and 18:00 and between 18:00 and 19:00 are small, then 18:00 can be identified as a peak mode; suppose that a spatial grid rises continuously between 07:00 and 09:00, and the adjacent differences are significantly higher than 0.05 times the intraday peak-to-valley difference of the spatial grid, then this period can be identified as a climbing mode.

[0042] S213. Collect historical forecast error samples according to load patterns and calculate the empirical quantile correction value for the corresponding pattern. The historical forecast error is defined as the historical actual load value minus the historical deterministic load forecast value.

[0043] In specific implementation, for the first For each spatial grid, first read its current load pattern label for the current time period. Then, historical samples with the same or similar load pattern label, season, weekday attributes, and weather level are selected from the historical sample library; historical prediction errors are then extracted from these samples to form the error sample set corresponding to the current period.

[0044] Furthermore, the error sample set is sorted in ascending order of numerical value, and empirical quantiles are extracted according to a preset confidence level. The confidence level can be 0.90, 0.95, or 0.97, with 0.95 being the preferred value. If the number of error samples is... Then take the sorted number 1. Each sample value is used as the base quantile correction value.

[0045] In one embodiment, for example, the current time period is determined to be in peak mode, and the filtered results are obtained. Historical error samples (unit: MW). The sorted sample values ​​are as follows: .because Therefore, the 38th sample value is taken. If the 38th sample value is 12MW, then the base quantile correction value is... .

[0046] It should be noted that this step adopts the method of "screening samples by pattern and taking empirical quantiles by sorting", which does not rely on complex probability models. In engineering implementation, only historical sample screening, sorting and indexing values ​​are required, which is easy to implement in existing operation and maintenance systems.

[0047] S214. Add a tail amplification correction to the basic quantile correction values ​​for peak and ramp modes to obtain the load forecast error correction value. , Indicates the first The spatial grid in the ... The load forecast error correction value for the time period.

[0048] In practical implementation, if the load mode label for the current time period... If the load pattern is either a low-valley or flat-valley pattern, then the baseline quantile correction value obtained in step 3> is directly used as the load forecast error correction value; if the load pattern label for the current period is... If it is a peak mode or a climbing mode, a positive amplification factor is added to the basic quantile correction value. The positive amplification factor can be determined by the standard deviation of the current error sample set and the mode amplification factor. The mode amplification factor is used to characterize the degree of additional conservatism for the upper tail risk under the peak mode or climbing mode, and can be taken from 0.5 to 1.0.

[0049] In one implementation, the method for calculating the load forecasting error correction value under peak or ramp modes is expressed as follows: ; in, This represents the 1st quantile obtained based on the empirical quantile of historical error samples. The spatial grid in the ... The base adjustment value for the time period; This represents the amplification factor of the upper tail, which can be taken as 0.8; This represents the sample standard deviation of the current error sample set.

[0050] In one embodiment, as an example, suppose a spatial grid obtains 40 historical error samples in peak mode. After sorting them from smallest to largest, the 38th sample has a value of 12MW. Then the basic quantile correction value is... Take 12MW; if the standard deviation of this error sample set is 3MW, the upper tail amplification factor... If we take 0.8, then the load forecasting error correction value is... for MW. In engineering applications, 14MW or 14.5MW can be used for subsequent calculations.

[0051] It should be noted that this step does not simply amplify all time periods uniformly, but only adds tail amplification correction to peak mode and ramp mode, thereby distinguishing between the two types of risk periods of "high load and stability" and "rapid changes", so that operation and maintenance investment is more concentrated on the periods that are really prone to staff shortages.

[0052] S215, Importance weight of each spatial grid computing device , Indicates the first The device importance weight of the first spatial grid is used to characterize the first... The comprehensive level of critical equipment density, important user density, and historical power outage impact intensity within a spatial grid.

[0053] In practical implementation, three categories of basic indicators are statistically analyzed for each spatial grid: Category 1 is key equipment indicators, including the quantity or capacity percentage of main transformers, cable joints, ring main units, switch stations, and protection devices; Category 2 is important user indicators, including the percentage of hospitals, rail transit, data centers, dual-power users, and important public utility users; Category 3 is historical impact indicators, including the number of households affected by historical faults, the amount of electricity affected, or the average duration of power outages. Then, these three categories of basic indicators are normalized to a range of 0 to 1, and then weighted and summed according to preset weights to obtain a comprehensive score; finally, the comprehensive score is mapped to equipment importance weights. In an easy-to-implement approach, the weights for key equipment indicators, important user indicators, and historical impact indicators can be set to 0.4, 0.35, and 0.25, respectively.

[0054] In one embodiment, as an example, if the normalized value of a key equipment indicator for a spatial grid is 0.8, the normalized value of an important user indicator is 0.7, and the normalized value of a historical impact indicator is 0.6, then the comprehensive score is... This can be further mapped to obtain the device importance weights. (Mapping relationship is) ).

[0055] S216. Calculate environmental risk factors for each spatial grid and each time period. , Indicates the first The spatial grid in the ... Environmental risk factors for a given time period are used to reflect the susceptibility to failure under the combined effects of temperature, humidity, wind speed, and load rate.

[0056] In the specific implementation, first read the... The spatial grid in the ... The system uses meteorological data and load factor for a given time period, where the load factor is the ratio of the deterministic load forecast before error correction to the rated capacity of the spatial grid. Then, temperature, humidity, wind speed, and load factor are normalized to a range of 0 to 1. For example, temperature can be linearly normalized between historical minimum and maximum temperatures, humidity between 0% and 100%, wind speed between 0 and a preset upper limit, and load factor between 0 and 1.2, truncated to 0 to 1. Finally, the four normalized values ​​are weighted and summed to obtain an environmental risk score, which is then mapped to an environmental risk factor.

[0057] In one implementation, the environmental risk score is obtained by summing the values ​​of temperature (0.3), humidity (0.2), wind speed (0.2), and load factor (0.3). ; then adopt The calculation method is mapped to environmental risk factors. At this time, the environmental risk factors... The value range is approximately 1.0 to 1.5.

[0058] In one embodiment, as an example, if a spatial grid has a normalized temperature value of 0.9, a normalized humidity value of 0.8, a normalized wind speed value of 0.2, and a normalized load factor value of 0.85 for a certain period of time, then the environmental risk score is: Further environmental risk factors were obtained. .

[0059] It should be noted that the environmental risk factor does not directly represent the number of failures, but rather serves as a risk amplification factor for the intensity of operation and maintenance needs, reflecting the dispatching side's meaning that "higher assurance intensity is required during periods of high temperature, high humidity, and heavy load."

[0060] S217. By integrating deterministic load forecast values, load forecast error correction values, equipment importance weights, and environmental risk factors, the spatial grid-level operation and maintenance demand intensity is obtained. , Indicates the first The spatial grid in the ... The intensity of maintenance demand during a given time period, expressed in equivalent standard man-hours, characterizes the basic maintenance input required to complete inspections, temperature measurements, defect handling, and emergency preparedness at that specific time and location. The calculation method is as follows: ; in, This represents the time conversion factor (unit: equivalent standard working hours / MW), the value of which can be determined based on operation and maintenance experience (for example, each MW load increase corresponds to an increase of approximately 0.2 equivalent standard working hours).

[0061] It should be noted that the intensity of spatial grid-level operation and maintenance requirements uses "equivalent standard working hours" as a unified unit of measurement, rather than directly converting it into the number of personnel. This is because subsequent responsibility area aggregation and personnel allocation need to consider the efficiency differences of personnel with different skill levels and cross-regional support constraints. Using a unified workload scale to express the requirements first facilitates unified optimization later.

[0062] S218. Traverse all spatial grids and all time periods to generate a spatial grid-level operation and maintenance demand intensity time series, and summarize it to form an operation and maintenance demand intensity map.

[0063] The operation and maintenance demand intensity map can be stored in a two-dimensional matrix format of "spatial grid × time period" or in a three-dimensional structure of "scheduling day × spatial grid × time period", which can be directly accessed for subsequent division of responsibility areas and personnel allocation.

[0064] In one embodiment, as an example, suppose the region is divided as follows: Spatial grid, number of time periods within a day (Each segment lasts 15 minutes). The two-dimensional matrix form is as follows: Table, elements For the first Grid number The intensity of maintenance demand (equivalent standard working hours) for a given time period. The three-dimensional structure is as follows: If the scheduling day number is... Then store as The three-dimensional array facilitates cross-day comparative analysis. When subsequently dividing responsibility areas, a specific time period can be directly read. All grids below The requirements at the responsibility area level are obtained by aggregating them according to the spatial grid.

[0065] It should be noted that this step does not directly amplify the deterministic load forecast value by a fixed ratio. Instead, it first identifies the load pattern, then extracts empirical quantiles from the historical forecast errors of the same pattern, and adds tail amplification correction only for peak and ramp patterns. This allows both "error uncertainty" and "operational status risk" to be mapped into the intensity of operation and maintenance demand, making it less likely for personnel shortages to occur during critical periods.

[0066] In one embodiment, such as Figure 3 As shown, box plots analyzing the historical forecast error distribution for each load mode are presented, comparing the distribution characteristics of load forecast error samples extracted from historical data under four load modes: low-valley, flat-valley, peak, and ramp. The boxes represent the interquartile range, the whiskers represent the overall range, and the diamond markers indicate the 95th percentile empirical quantile. Experimental results show that the error distribution of the peak and ramp modes exhibits significant right skewness and upper tail characteristics, with their 95th percentiles being much higher than those of the low-valley and flat-valley modes. In the figure, the horizontal axis represents discrete load mode categories, and the vertical axis represents historical forecast error values, in megawatts (MW).

[0067] S22. Division of Responsibility Zones and Generation of Support Constraints Based on Integrated Electrical Topology Adjacency and Road Network Travel Distance Dividing responsibility areas solely based on geographical distance can easily lead to problems such as weak electrical connections within the same responsibility area, excessive detours for cross-regional support, and unpredictable emergency response times. Distribution operation and maintenance responsibility areas must not only be physically accessible by vehicles, but also be as consistent as possible with the electrical connections to the distribution network.

[0068] This step incorporates both electrical topology adjacency and road travel distance. First, it constructs composite reachability relationships between spatial grids, and then generates responsibility zone division results and cross-zone support constraints based on these relationships. The specific steps are as follows: S221. Treat all spatial grids as graph nodes, establish electrical connections between spatial grids, and calculate the electrical distances between pairs of spatial grids. .in, Indicates the first The spatial grid and the first The dimensionless electrical distance between spatial grids, which comprehensively considers the physical line lengths and the degree of separation between switching nodes in the distribution network, is used to reflect the tightness of electrical coupling. and This represents any two spatial grid numbers.

[0069] In practical implementation, the feeder, busbar segment, and tie switch information of equipment within each spatial grid are extracted from distribution network GIS data, feeder connection tables, switch connection tables, and equipment ledgers. If equipment in two spatial grids is located on the same continuous feeder segment or has a clearly defined tie path, then an electrical connection is considered to exist between these two spatial grids. Then, for two spatial grids with an electrical connection, the line length along the connection path is accumulated, and a fixed penalty value is added for each switch node crossed, thereby obtaining the electrical distance. .

[0070] In a convenient implementation method, the total line length along the path can be divided by the average feeder segment length of the service area to obtain the normalized line length; then a penalty of 0.2 is added for each switching node crossed; finally, the two are added together to obtain the electrical distance. The smaller the electrical distance, the more suitable it is for the two spatial grids to be collaboratively managed by the same area of ​​responsibility in terms of operation and handling.

[0071] In one embodiment, as an example, it is assumed that the average feeder segment length in a service area is 2.5 km. Grid With grid The total length of the line along the feeder path is 5.0 km, passing through 2 switch nodes. Therefore, the normalized line length = Switch penalty = electrical distance .

[0072] S222. Using the center point, main equipment location, or station location of each spatial grid as the road network positioning point, calculate the average travel time between any two spatial grids in the road network. , Indicates the first The spatial grid and the first The average road travel time (in minutes) between spatial grids is calculated based on the shortest path of the road network and can distinguish different traffic periods.

[0073] In practical implementation, the road network is constructed as a weighted graph model, with road nodes corresponding to intersections or key turning points, and road edge weights corresponding to historical average travel times. Then, a shortest path algorithm (such as Dijkstra's algorithm or A* algorithm, both standard algorithms for solving shortest paths in graph theory) is used to calculate the... The spatial grid to the first The shortest travel time for each spatial grid. Then, to better reflect actual operating scenarios, multiple sets of road edge weight databases can be established according to different traffic periods such as weekday off-peak, weekday peak, and holidays, and the corresponding edge weight database can be selected according to the current scheduling day attributes and time period.

[0074] In one embodiment, for example, if the road network edge weights are based on historical average congestion data during the morning rush hour on a weekday, from the grid... Center point to grid The shortest path to the center point is calculated to be 12.5 minutes. Minutes. The same grid pair may take 8.2 minutes during off-peak hours.

[0075] S223. Setting the maximum emergency response time threshold The maximum emergency response time threshold is used to indicate the upper limit of the passage time that allows inclusion in the same responsibility system or the establishment of cross-regional rapid support relationships.

[0076] In an easily implementable manner, the maximum emergency response time threshold A threshold of 30 minutes is acceptable. If the average travel time between two spatial grids exceeds this threshold, they will not be considered as fast-access reachable objects.

[0077] S224. Calculate the composite reachability edge weight for any two spatial grids. , Indicates the first The spatial grid and the first The composite reachability edge weight between spatial grids is dimensionless. It reflects both electrical adjacency and road accessibility. A larger value indicates that the two spatial grids are more suitable to be assigned to the same responsibility area, or more suitable to establish cross-area support relationships. The calculation method is as follows: ; in, This represents the electrical distance attenuation parameter, used to control the degree of influence of electrical distance on the weight of composite reachable edges, and can be set to 0.5.

[0078] It should be noted that if there is no electrical connection between two spatial grids, or if the average travel time exceeds the maximum emergency response time threshold, the composite reachability edge weight will be directly adjusted. Set to 0.

[0079] In one embodiment, for example, if two spatial grids are located in adjacent sections of the same feeder, have a small electrical distance, and have an average travel time of 8 minutes, then the composite reachable edge weight is larger; if two spatial grids are geographically close but belong to the ends of different feeders and have an average travel time greater than 30 minutes, then the composite reachable edge weight is directly set to 0.

[0080] S225. Perform responsibility area division based on the composite reachability edge weights between all spatial grids to obtain the responsibility area label of each spatial grid.

[0081] In the actual implementation, the number of responsibility areas should be determined first. The number of responsibility areas can be determined by the existing operations and maintenance organizational structure or by the existing number of on-site personnel. Then, select... A seed spatial grid serves as the initial responsibility area center. The seed spatial grid can be preferably a spatial grid with high operation and maintenance requirements, good road network accessibility, and existing nearby outposts or main sites.

[0082] In one embodiment, for example, let the number of responsibility areas be... Grids A, B, and C were selected as seed centers. The average composite reachable edge weight between unassigned grid D and the boundary grid of responsibility area 1 (center A) is 0.45, with an average travel time of 18 minutes; the average composite reachable edge weight between unassigned grid D and the boundary grid of responsibility area 2 (center B) is 0.22, with an average travel time of 32 minutes (exceeding...). (Minutes, not feasible); the average composite reachable edge weight of the boundary grid of responsibility zone 3 (center C) is 0.31. Therefore, grid D is preferentially merged into responsibility zone 1.

[0083] Furthermore, each seed spatial grid is initialized as a responsibility region, and then the unassigned spatial grids are traversed. For any unassigned spatial grid, the average weight of the composite reachable edge between it and the boundary spatial grids of each responsibility region is calculated; if the average travel time between a responsibility region and the unassigned spatial grid does not exceed the maximum emergency response time threshold, and the average weight of the composite reachable edge is the largest, then the unassigned spatial grid is incorporated into the responsibility region.

[0084] Furthermore, after all spatial grids have completed their initial assignment, connectivity checks are performed on each responsibility area. If an edge spatial grid that is electrically disconnected or has obvious detours on the road network appears in a responsibility area, that edge spatial grid is reassigned to an adjacent responsibility area with a higher composite reachability edge weight, until each responsibility area meets the requirements of "internal reachability and boundary continuity".

[0085] In one implementation, the average travel time between any major spatial grids within the responsibility area is preferably no more than 30 minutes, and the average composite reachability edge weight between the major spatial grids within the responsibility area and the center of the responsibility area is preferably no less than 0.2.

[0086] It should be noted that this step adopts the approach of "first selecting the seed space grid, then gradually expanding according to the composite reachable edge weights, and finally making boundary corrections". It does not rely on a complex graph cut optimizer. When the system is implemented, only edge weight calculation, neighborhood traversal and boundary redistribution are required, which is convenient for engineering implementation.

[0087] S226. Generate a cross-regional support constraint mask matrix based on the responsibility area division results. The elements in the cross-region support constraint mask matrix are defined as denoted as , used to indicate the first The responsibility area and the first Is it permissible to establish cross-regional support relationships between different responsibility areas?

[0088] In the actual implementation, traverse any two responsibility areas. and If there exists at least one pair that belongs to a responsibility area. and responsibility area Spatial grids with composite reachable edge weights If it exceeds the preset support threshold, then Set to 1; otherwise, Set to 0. The preset support threshold can be 0.15. To facilitate scheduling calculations, it can be set to 0. This ensures that the cross-regional support constraint mask matrix remains symmetric. Furthermore, the diagonal elements... A value of 1 is fixed to indicate that personnel within the area of ​​responsibility are naturally permitted to participate in the handling of matters within that area.

[0089] In one embodiment, as an example, the responsibility area With responsibility area There exists at least one pair of spatial grids, whose (If the value is greater than the preset support threshold of 0.15), then This indicates permission to provide support. If another pair of responsibility areas... and All spatial grid pairs between If all values ​​are ≤0.15, then Matrix example ( )for:

[0090] S227. Output responsibility area labels and cross-region support constraint mask matrix. Responsibility area labels are used to aggregate spatial grid-level operational demand intensity into responsibility area-level operational demand intensity; the cross-region support constraint mask matrix is ​​used to restrict combinations of responsibility areas that do not meet support conditions in subsequent personnel allocation optimization. That is, in subsequent personnel allocation optimization, if... Then it is prohibited to leave the area of ​​responsibility. To the area of ​​responsibility Dispatch any cross-regional support personnel.

[0091] It should be noted that this invention does not simply divide areas based on geographical distance or administrative boundaries, but rather considers both electrical topological adjacency and road travel distance simultaneously. It first constructs a composite reachability relationship, and then divides responsibility areas and generates cross-regional support constraints based on this relationship. The resulting responsibility areas better reflect the actual organizational methods of power distribution operation and maintenance, and the cross-regional support paths are closer to truly executable paths.

[0092] In one embodiment, such as Figure 4 As shown, the spatiotemporal distribution of maintenance demand intensity is analyzed. A two-dimensional heatmap is used to centrally display the spatiotemporal evolution of maintenance demand intensity (R_i,t) for all 36 spatial grids within the previous 12 hours (48 time periods). Darker areas (deeper red) represent greater maintenance demand intensity. This map visually reflects the final output of step S2: a refined demand distribution using "equivalent standard working hours" as the unified unit, obtained by integrating deterministic load forecasting, patterned error correction, equipment importance, and environmental risk. The horizontal axis represents time in hours (0 to 12 hours), and the vertical axis represents the spatial grid number (1 to 36). The color levels correspond to numerical units of equivalent standard working hours.

[0093] S3. Based on the maintenance demand intensity map, combined with the maintenance personnel skill level profiles, historical work order processing efficiency, and the upper limit of manpower available in each responsibility area, a multi-objective optimization model is constructed. Under the constraints of skill matching and response timeliness, this model enables collaborative personnel deployment across responsibility areas, outputting personnel deployment plans based on time periods and skill levels. After obtaining the responsibility area division results, it is also necessary to determine the number of personnel stationed in each area and the number of personnel providing cross-area support based on the responsibility area-level maintenance demand intensity of each area in each time period. This problem is simultaneously affected by skill level differences, the upper limit of available personnel, cross-area support permit relationships, and the amplified risks during periods of rapid change. If a standard minimum cost model is still used, it often underestimates the risk of staff shortages during peak and ramp-up periods and fails to reflect the efficiency advantages of highly skilled personnel.

[0094] This step first establishes skill level efficiency coefficients based on historical work order data, then constructs a fitness function that combines "local garrison cost + cross-regional support cost + sudden change weighted deficit penalty"; next, it uses an improved firefly algorithm with feasible region projection to immediately correct the personnel deployment plan after each location update, ultimately outputting personnel deployment results at the responsibility area level, by time period, and by skill level. The specific steps are as follows: S31. Fitness Construction Based on Skill Level Efficiency Coefficient and Demand Abrupt Change Weight S311. Aggregate spatial grid-level operation and maintenance demand intensity into responsibility area-level operation and maintenance demand intensity. , Indicates the first The responsibility area in the first The intensity of maintenance and operation needs at the responsibility area level during a given time period.

[0095] In practical implementation, the maintenance demand intensity of all spatial grids belonging to the same responsibility area at the same time period is directly summed to obtain the first... The responsibility area in the first The intensity of maintenance demand at the responsibility area level during a given time period .

[0096] In one embodiment, as an example, the responsibility area It includes three spatial grids: grid 15, grid 16, and grid 17. (During the time period...) At 10:00 AM, the maintenance demand intensity (equivalent standard working hours) for each grid was 4.2, 3.8, and 5.0 respectively. Therefore, the demand intensity at the responsibility area level was: .

[0097] Furthermore, if it is necessary to consider the degree of traffic dispersion within the responsibility area, a dispersion coefficient of 1.0 to 1.2 can be multiplied by the summation result; in a more convenient implementation method, the summation result can be directly used as the intensity of operation and maintenance demand at the responsibility area level.

[0098] S312. Establish skill level efficiency coefficients based on historical work order processing data. Skill level efficiency coefficient Used to indicate the first The equivalent work efficiency of an individual at a skill level relative to a baseline skill level.

[0099] In practical implementation, the average completion time, first-time repair rate, and rework rate of the same type of work orders processed by personnel of different skill levels are extracted from the historical work order system. First, the average completion time is converted into processing capacity per unit time. Then, the first-time repair rate is used for positive correction, and the rework rate is used for reverse correction, to obtain the... The overall processing efficiency of personnel at each skill level is calculated. Then, the overall processing efficiency of all skill levels is normalized by comparing it to a baseline skill level, resulting in a skill level efficiency coefficient. .

[0100] In one embodiment, as an example, a junior staff member (benchmark) has the following characteristics: an average processing time of 1.2 hours, a first-time repair rate of 0.85%, a rework rate of 0.10, and an overall processing efficiency of 1.0. An intermediate staff member (benchmark) has the following characteristics: an average processing time of 0.8 hours, a first-time repair rate of 0.92%, a rework rate of 0.05, and an overall efficiency approximately 1.6 times that of the junior staff member. Senior personnel (benchmark): Their overall efficiency is approximately 2.4 times that of junior personnel. .

[0101] It should be noted that the skill level efficiency coefficient is not a wage coefficient, but a workload conversion coefficient, used to uniformly convert personnel of different skill levels to the scale of "standard personnel equivalent".

[0102] S313. Define the candidate personnel deployment plan and calculate the actual available operation and maintenance capabilities of the candidate personnel deployment plan in each responsibility area and time period.

[0103] In the specific implementation, the first The deployment plan for each candidate is denoted as In the candidate deployment plan middle, Indicates the first The candidate deployment plan is in the first The responsibility area, the first The first time slot arrangement Number of personnel stationed in this area for each skill level; Indicates the first The candidate deployment plan is in the first The time period starts from the The responsibility area was transferred to the first The first responsibility area Number of personnel providing cross-regional support at each skill level.

[0104] In one embodiment, as an example, the candidate deployment scheme It contains two parts of encoding: a>The area under the jurisdiction of this unit: For each area of ​​responsibility Each time period Each skill level Give the number of people .

[0105] b>Cross-region support section: For each allowed support path Each time period Each skill level Give the number of people .

[0106] Example snippet ( )for: ; The rest are 0. This means that during time period 1, there are 2 junior staff and 1 intermediate staff stationed in responsibility area 1, and 1 junior staff member is transferred from responsibility area 1 to support responsibility area 2.

[0107] Furthermore, for any area of ​​responsibility, any time period, and any skill level, first calculate the actual number of available personnel for that area of ​​responsibility and that skill level during that time period, which is the number of personnel stationed in the area plus the total number of support personnel transferred in, and then subtract the total number of support personnel transferred out; then multiply by the corresponding skill level efficiency coefficient; finally sum over all skill levels to obtain the actual available operation and maintenance capacity of that area of ​​responsibility during that time period.

[0108] In one implementation, the candidate deployment scheme In the The responsibility area, the first The actual available operational capacity for a given period is denoted as: The calculation method is expressed as follows: ; in, Indicates the first The candidate deployment plan is in the first The time period starts from the The responsibility area was transferred to the first The first responsibility area Number of personnel providing cross-regional support at each skill level; Indicates the first The candidate deployment plan is in the first The time period starts from the The responsibility area was transferred to the first The first responsibility area Number of personnel providing cross-regional support at each skill level.

[0109] S314. Calculate the personnel shortage in each responsibility area for each time period. Specifically, for each responsibility area and each time period, calculate the difference between "responsibility area-level operation and maintenance demand intensity minus actual available operation and maintenance capacity"; if the difference is less than 0, it means that there is no shortage in that responsibility area for that time period, and it is directly recorded as 0; if the difference is greater than 0, the difference is recorded as a personnel shortage. In specific implementation, the first... The candidate deployment plan is in the first The responsibility area, the first Staff vacancies during a given period are recorded as .

[0110] In one embodiment, as an example, the responsibility area During the period Operation and maintenance demand intensity (Equivalent standard working hours), actual available operation and maintenance capacity (Standard personnel equivalent). Then the vacancy... ,in, This indicates the operation of retrieving the maximum value.

[0111] It should be noted that the penalty for staff shortages only applies to the shortfall. While redundant staff may increase costs, they do not pose a risk of staff shortages, so they are not penalized again in this category.

[0112] S315. Calculate the demand drastic weight based on the magnitude of changes in the intensity of maintenance demand at the responsibility area level over adjacent time periods. , Indicates the first The responsibility area in the first The demand drastic change weighting is used to amplify the shortage penalty for the corresponding time period in peak mode and ramp mode.

[0113] In practical implementation, the change in the intensity of operation and maintenance requirements at the responsibility area level is calculated for each responsibility area in adjacent time periods. For example, the calculation... Then, this change is mapped to a weight value greater than or equal to 1, expressed as: ; in, This represents the amplification factor for sudden changes, which can be taken as 2.0; This represents the scale parameter, which can be set to 5.0; This represents the hyperbolic tangent function.

[0114] In one embodiment, for example, if the area-level maintenance demand intensity of a certain area in the previous period was 10 and the area-level maintenance demand intensity in the current period is 24, then the change is 14. Substituting this into the above formula, the resulting demand change weight will be significantly greater than 1, thereby significantly increasing the deficit penalty for that period.

[0115] It should be noted that the weighting of sudden demand changes is not uniformly amplified across all time periods, but only increases the weighting of periods with significant changes between adjacent time periods, thereby reflecting the risk difference between "stable shortages" and "sudden demand shortages" in the optimization objective.

[0116] S316. Calculate the local deployment cost and cross-regional support cost of the candidate personnel deployment plan. Specifically, set a local deployment unit cost coefficient for each skill level. , used to indicate the first The stationing cost of personnel at each skill level per unit of time; setting the average support travel distance or average support travel time for each pair of responsibility areas, and setting a unit support cost coefficient. .

[0117] Furthermore, for any candidate personnel deployment plan, the cost of stationing personnel in the current area is obtained by multiplying the number of personnel stationed in the current area by the corresponding unit cost coefficient for each area of ​​responsibility, time period, and skill level; then, the cost of cross-regional support is obtained by multiplying the number of personnel supporting other areas, the average support distance, and the unit support cost coefficient for each area of ​​responsibility, time period, and skill level.

[0118] In an easy-to-implement approach, the unit cost coefficients for junior, intermediate, and senior personnel stationed in the area can be set to 1.0, 1.4, and 2.0, respectively; the average support distance can be directly adopted as the average road distance between the center points of the responsibility area.

[0119] In one embodiment, as an example, the cost of stationing personnel in this area is: Area 2 during the time period Two people are stationed at the primary level. ), 1 intermediate-level person ( If the time period is 1 hour, then the cost of stationing personnel in that area during that time period = Cost unit. The cost of cross-regional support is: dispatching one junior personnel from responsibility area 2 to support responsibility area 3, with an average support distance of 12km, and a unit support cost coefficient. (Cost unit / person·km), then the support cost = Cost unit.

[0120] S317. Combine the local garrison cost, cross-regional support cost, and emergency weighted deficit penalty into a fitness value for the candidate personnel deployment plan. The smaller the fitness value, the better the candidate personnel deployment plan.

[0121] In one implementation, the candidate deployment scheme The fitness value is denoted as The calculation method is expressed as follows: ; in, Indicates the first The local stationing cost of each candidate's personnel deployment plan (calculated as "number of local personnel stationed in each responsibility area at each time period and skill level × corresponding unit cost coefficient") (This is obtained by summing the duration of the time period) Indicates the first The cross-regional support cost for each candidate's personnel deployment plan (calculated as "number of personnel supported across regions × average distance of support path × unit support cost coefficient") (Accumulated) This represents the deficit penalty coefficient, used to adjust the proportion of the deficit penalty in the total fitness, and can be taken from 50 to 200. Indicates the first The candidate deployment plan is in the first The responsibility area, the first Staffing vacancies during the period (through (Calculated) This represents the nonlinearity index of the deficit, used to impose a stronger penalty on larger deficits, and can be taken as 1.2 to 2.0.

[0122] It should be noted that this step uses a non-linear penalty instead of a linear penalty for the deficit because a small deficit may be mitigated by temporary scheduling, while a large deficit will significantly increase the risk of equipment malfunction, repair timeouts, and handling delays. Therefore, it should be more strongly suppressed during the optimization process.

[0123] S318. Convert the fitness value into light intensity for the firefly algorithm. Specifically, set the light intensity to the reciprocal of the fitness value, or to the difference between "the current population's maximum fitness value and the fitness value of the candidate deployment scheme". Based on this, the fireflies corresponding to candidate deployment schemes with smaller fitness values ​​are brighter, and are more likely to attract other fireflies to approach them in subsequent iterations.

[0124] It's important to note that the Firefly Algorithm is a swarm intelligence optimization algorithm that searches by simulating the behavior of fireflies being attracted to brighter individuals. In the optimization process, the objective function value (fitness value) is mapped to light intensity; that is, individuals with better fitness (where lower is better) correspond to individuals with stronger light intensity. Light Intensity The setup method is as follows: (The smaller the adaptability, the greater the light intensity); or ( (This is the current maximum fitness value of the population).

[0125] S32, Improved Firefly Update with Embedded Support Constraint Projection and Temporal Consistency Constraints

[0126] The conventional firefly algorithm directly updates the position of continuous variables, which often produces unexecutable results such as negative number of people, non-integer number of people, unauthorized support, and excessive dispatch of local personnel. It then relies on a penalty function to repeatedly correct these results, resulting in low convergence efficiency.

[0127] This step performs feasible region projection immediately after each location update, directly correcting the update results to a valid personnel deployment plan; simultaneously, it incorporates time consistency constraints into the individual difference calculation to make personnel arrangement changes between adjacent time periods smoother. The specific steps are as follows: S321. Encode the candidate personnel deployment plan as firefly locations. Specifically, this involves assigning the number of personnel stationed in each area of ​​responsibility, across all time periods, and for all skill levels. The first part of the code is formed by arranging the data in a fixed order; then the number of cross-regional support personnel on all permitted support paths is... The two parts are arranged in a fixed order to form the second part of the encoding; finally, the two parts are concatenated to form a complete firefly position vector.

[0128] In the actual implementation, to facilitate subsequent feasible domain projection, a three-dimensional array of "responsibility area × time period × skill level" and a four-dimensional array of "inbound responsibility area × outbound responsibility area × time period × skill level" can be retained inside the program, and then temporarily expanded into vector form when calculating the difference in firefly positions.

[0129] In one embodiment, for example, let Z=2, T=3, K=2, then the garrison area includes There are two variables. If there are two supporting paths ( and Then the cross-region support part includes There are two variables. These two parts are concatenated in order into a vector of length 24, which is the firefly location code.

[0130] S322. Generate an initial firefly population that satisfies the basic constraints.

[0131] Specifically, first set an adjustable upper limit for each responsibility area and each skill level, denoted as... , used to indicate the first The responsibility area in the first The maximum number of personnel that can be deployed at each skill level is determined. Then, an initial number of personnel stationed in the area is generated based on the intensity of maintenance needs at the responsibility area level. It is preferable to allocate the basic number of personnel according to the proportion of maintenance needs at the responsibility area level, and then allocate them to each skill level according to the preset skill ratio; the preset skill ratio can be determined based on the existing team structure, for example, 0.5, 0.3 and 0.2 for the three skill levels respectively.

[0132] Furthermore, in cross-regional support constraint mask matrix Between allowed responsibility zones, a small initial number of cross-zone support personnel are randomly generated to increase population diversity. Then, any portion exceeding the deployable limit is pruned, and negative values ​​are directly set to 0, thus obtaining an initial firefly population that meets basic feasibility requirements.

[0133] S323. Calculate the time-consistency weighted distance between fireflies. Specifically, instead of directly using ordinary Euclidean distance, first apply time-smoothing weights to the personnel sequence for each responsibility area and skill level, and then calculate the distance between two fireflies. Then, assign higher weights to differences between adjacent time periods and lower weights to differences between time periods further apart. For example, the weight between the current time period and one time period before and after can be 1.0, and the weight between the current time period and two time periods before and after can be 0.6.

[0134] Based on this, when two candidate personnel deployment plans have similar values ​​in a single time period, but one plan has a sharp jump while the other plan is smoother, the time consistency weighted distance will prioritize the smooth plan as being closer to the optimal direction, thus helping to form a personnel deployment result that is more in line with the scheduling pattern.

[0135] S324. Move weaker fireflies towards stronger fireflies, and add random perturbations. Specifically, the... During the nth iteration, the 1st Only fireflies face the first The update method for only optimizing fireflies is expressed as follows: ; in, Indicates the first Only fireflies in the first The candidate deployment plan for the next iteration. Indicates the first Only fireflies in the first The candidate deployment plan for the next iteration; Indicates compared to the first The firefly is better than the first Only fireflies in the first The candidate deployment plan for the next iteration; Indicates the first Only fireflies and the first Time-consistency weighted distance between individual fireflies; This represents the initial attraction coefficient, which can be set to 1.0. This represents the attraction attenuation coefficient, which can be taken as 0.5; Indicates the first The random perturbation step size of each iteration can be decreased according to the number of iterations; To represent a random perturbation vector, a Lévy-distributed random vector or a Gaussian random vector can be used; This represents the feasible region projection operator.

[0136] It should be noted that the random perturbation during the update process is used to prevent the population from converging too early; the feasible region projection operator is used to immediately correct the update result back to an executable solution, rather than directly carrying infeasible solutions into the next iteration.

[0137] S325. Perform feasible region projection on the updated candidate deployment plan. Specifically, perform projection corrections in the following order: First, set all negative personnel counts to 0. Since the personnel count cannot be negative, any negative result is considered an intermediate invalid value generated during the update process and needs to be cleared during projection.

[0138] Furthermore, all non-integer numbers are rounded according to preset rules, such as rounding to the nearest integer or rounding down.

[0139] b>Check the cross-regional support constraint mask matrix If a certain responsibility region has a cross-regional support constraint mask value... If the value is 0, then all cross-regional support personnel in that responsibility area will be set to 0, indicating that support is not allowed to occur on that path.

[0140] c> For each responsibility area, each time period, and each skill level, calculate the sum of "number of personnel stationed in this area + total number of personnel transferred out for support". If this calculated value exceeds the maximum number of personnel that can be deployed for that responsibility area and skill level... If the number of personnel stationed in the current area is already exceeded, the number of personnel transferred out for support will be reduced proportionally. If the number of personnel stationed in the current area has already exceeded the available allocation limit, then the number will be reduced to the available allocation limit.

[0141] In one embodiment, for example, if responsibility zone 3 only allows 5 people to be deployed at a certain skill level during a certain period, and the updated result shows 4 people stationed in this zone, 2 people supporting responsibility zone 1, and 1 person supporting responsibility zone 2, then the total number of people occupied is 7, exceeding the deployment limit of 5 people. In this case, first retain the 4 people stationed in this zone, and then reduce the total number of people deployed for support from 3 to 1. If the original values ​​of the two support paths are 2 and 1 respectively, they can be reduced proportionally to 1 and 0, or the support path with the shorter travel time can be retained first.

[0142] d> Recalculate the actual number of available personnel after transfers in and out for each responsibility area and each time period. If there are situations such as negative available personnel in the area, duplicate transfers from multiple directions causing duplicate statistics, or inconsistencies in the total number of personnel, the correction will continue in the order of "priority for personnel stationed in the area, priority for support from neighboring areas, and reduction after support from distant areas" until all constraints are met.

[0143] In one embodiment, as an example, the responsibility area During the period Skill Level (Basic) Available allocation limit The updated plan is as follows: 3 people stationed in this area, 2 people transferred to support area 1, and 2 people transferred to support area 3, for a total of [number missing] people. The number of people exceeds the limit of 5.

[0144] In this embodiment, the correction steps are as follows: retain 3 people stationed in this area, and the remaining available quota is... People. The ratio of people in the original two support routes was: The number of personnel will be reduced proportionally to a total of 2, meaning each support route will retain 1 person. If a certain support route... If the number of participants on a given path is zero, then the remaining paths will be allocated proportionally.

[0145] It should be noted that feasible region projection is not a simple truncation, but rather the personnel deployment rules are directly written into the correction process after each location update, thereby ensuring that each iteration revolves around the real executable solution as much as possible.

[0146] S326. Recalculate the fitness value and light intensity of the projected candidate personnel deployment scheme, and update the current optimal candidate personnel deployment scheme.

[0147] In practice, if a firefly's fitness value is better than the current historical best fitness value after location update and feasible region projection, then the candidate deployment scheme corresponding to that firefly is saved as the new historical best candidate deployment scheme.

[0148] In one embodiment, for example, the first After the next iteration, individual fireflies The fitness value after projection of the feasible region is The current global historical best fitness value is .because Therefore, Save it as a new historical best candidate solution.

[0149] S327. Repeat the position update, feasible region projection, and fitness update until the iteration termination condition is met.

[0150] In practical implementation, the iteration termination condition can be any one or a combination of the following: reaching the maximum number of iterations; the improvement of the historical best fitness value is less than a preset threshold for several consecutive rounds; or the positional change of most fireflies in the population is already very small. In a convenient implementation method, the maximum number of iterations can be 100 to 300, and the iteration can be terminated early when the improvement of the best fitness value is less than 1% for 10 consecutive rounds.

[0151] S328. Output the final personnel deployment results. The final personnel deployment results shall include at least the number of personnel stationed in each responsibility area at each time period, the number of cross-area support personnel between each responsibility area at each time period, the configuration quantity of each skill level, and the cross-area support flow matrix formed therefrom between responsibility areas.

[0152] Furthermore, the final personnel deployment results can be organized into a structured instruction table of "responsibility area-time period-skill level", or output as a data file that the scheduling system can directly read.

[0153] In one embodiment, for example, the final output is a structured instruction table in CSV or JSON format, with fields including: responsibility area number. Time period number Skill Level Number of personnel stationed in this area Transferred to the source responsibility area and number of people (If there is no support, it will be empty). Example (partial row):

[0154] It's important to note that this step doesn't simply use the Firefly algorithm. Instead, it employs two key techniques: firstly, by using a cross-regional support constraint mask matrix and feasible region projection, operational constraints such as "cannot support," "cannot exceed the upper limit," and "cannot excessively deploy personnel" are directly embedded into each iteration; secondly, by using time-consistency weighted distance, the actual scheduling patterns where personnel allocation shouldn't drastically change between adjacent time periods are incorporated into the search process. This results in personnel deployment that not only meets cost and security objectives but also more closely approximates a practically feasible solution.

[0155] S4, Intelligent Deployment of Maintenance Personnel

[0156] After completing the construction of the operation and maintenance demand intensity map in step S2 and completing the collaborative allocation optimization of the responsibility area and outputting the final personnel deployment results in step S3, this step is responsible for transforming the optimized personnel deployment plan by time period and skill level into intelligent allocation instructions that can be executed by the operation and maintenance team, and establishing a closed-loop mechanism for plan issuance and dynamic fine-tuning.

[0157] Specifically, the structured personnel deployment table of "responsibility area-time period-skill level" output by step S3 is first parsed into a detailed daily schedule for each responsibility area. The schedule clearly specifies the number of junior, intermediate and senior skilled personnel required to be stationed in each 15-minute time period of the day in each responsibility area, as well as the flow and volume of cross-area support personnel between each responsibility area during specific time periods.

[0158] Then, the system interfaces the scheduling plan with the maintenance work order system and mobile terminal application, pushing the on-site tasks in this area to the mobile terminals of the corresponding responsibility area team leaders in the form of an electronic schedule. For time periods involving cross-regional support, a support dispatch task to be confirmed is generated simultaneously on the terminals of both the departure and destination responsibility areas. The task information includes the support start time, destination grid, estimated travel time, and the type of tools and equipment to be carried. During the execution of the plan, the system continuously monitors real-time load fluctuations, sudden fault alarms, and changes in the actual road network traffic status. If the actual load of the current time period exceeds the error range after considering the upper tail amplification correction in step S2, or if a regional equipment trip causes a sudden change in local maintenance needs, the system will trigger a dynamic allocation mechanism based on a preset threshold. Based on the original personnel deployment plan, and under the premise of meeting the cross-regional support constraint mask matrix and the upper limit of personnel allocation, the system will prioritize the scheduling of nearby idle personnel with matching skills for rapid replacement, and generate an immediate allocation command to be issued to the relevant personnel terminals.

[0159] At the end of the scheduling day, the system archives the deviation data between the actual personnel allocation records and the S3 optimization plan. This data is used to provide feedback and correction for key parameters such as the upper tail amplification coefficient and skill level efficiency coefficient in peak and ramp modes in future periods, thereby realizing a complete intelligent allocation closed loop from data collection, demand forecasting, optimization allocation to execution feedback.

Claims

1. A method for intelligent dispatching of operation and maintenance personnel based on power load forecasting, characterized in that, Includes the following steps: S1. Acquire multi-source heterogeneous data, including deterministic load forecast sequences, historical actual load data, equipment status monitoring data, meteorological forecast information, traffic network data, operation and maintenance personnel skills and historical work order data, and perform cleaning, spatiotemporal alignment and feature fusion preprocessing. S2. By integrating load forecasting results and their spatiotemporal error distribution, equipment health status assessment and meteorological early warning information, an operation and maintenance demand intensity map reflecting the distribution of operation and maintenance pressure in future periods is constructed. S3. Based on the intensity map of operation and maintenance demand, combined with the skill level files of operation and maintenance personnel, the efficiency of historical work order processing and the upper limit of manpower that can be allocated in the responsibility area, a multi-objective optimization model is constructed. Under the constraints of skill matching and response timeliness, the collaborative allocation of personnel across responsibility areas is realized, and personnel deployment plans are output according to time period and skill level. S4. Transform personnel deployment plans into structured intelligent dispatch instructions and distribute them to maintenance teams via mobile terminals; at the same time, monitor actual load deviation and work order execution progress in real time, and trigger a dynamic re-optimization mechanism when the deviation exceeds the threshold. Specifically, S2 is: S21. Load pattern adaptive correction and demand intensity mapping of load forecasting error: The current period is first classified into a specific load pattern, and then empirical quantiles are extracted from historical error samples of the same pattern to obtain the error correction value as the state changes over time, so that the operation and maintenance demand intensity is more in line with the actual operation risk. S22. Integrating electrical topology adjacency and road network travel distance, first construct the composite reachability relationship between spatial grids, and then generate the responsibility area division results and cross-area support constraints based on this, specifically: S221. Treat all spatial grids as graph nodes, establish electrical connection relationships between spatial grids, and calculate electrical distances between pairs of spatial grids; S222. Using the center point, main equipment location, or station location of each spatial grid as the road network positioning point, calculate the average travel time between any two spatial grids in the road network. S223. Set a maximum emergency response time threshold. The maximum emergency response time threshold is used to indicate the upper limit of the passage time that allows inclusion in the same responsibility system or allows the establishment of cross-regional rapid support relationships. S224. Calculate the composite reachability edge weight for any two spatial grids. The composite reachability edge weight is used to reflect both electrical adjacency and road reachability. S225. Perform responsibility area division based on the composite reachability edge weights between all spatial grids to obtain the responsibility area label of each spatial grid; S226. Generate a cross-regional support constraint mask matrix based on the responsibility area division results. The elements in the cross-regional support constraint mask matrix are used to indicate whether cross-regional support relationships are allowed between responsibility areas. S227. Output responsibility area labels and cross-region support constraint mask matrix. The responsibility area labels are used to aggregate the spatial grid-level operation and maintenance demand intensity into the responsibility area-level operation and maintenance demand intensity. The cross-region support constraint mask matrix is ​​used to restrict the combination of responsibility areas that do not meet the support conditions in subsequent personnel allocation optimization.

2. The intelligent dispatching method for operation and maintenance personnel based on power load forecasting according to claim 1, characterized in that, S3 specifically refers to: S31. Based on the skill level efficiency coefficient and demand drastic change weight, construct a fitness value for evaluating candidate deployment schemes, and convert the fitness value into the light intensity of the firefly algorithm. S32. An improved firefly update with embedded support constraint projection and time consistency constraints performs feasible domain projection immediately after each location update, directly correcting the update result to a legal personnel deployment plan; at the same time, time consistency constraints are added to the individual difference calculation to make personnel arrangement changes between adjacent time periods smoother.

3. The intelligent dispatching method for operation and maintenance personnel based on power load forecasting according to claim 1, characterized in that, S21 specifically refers to: S211. Read the deterministic load forecast sequence of all spatial grids throughout the entire scheduling cycle; S212. Perform sliding window analysis on the deterministic load forecast sequence of each spatial grid to obtain the load pattern label for the current time period; S213. Collect historical forecast error samples according to load patterns and calculate the empirical quantile correction value for the corresponding pattern. S214. Add tail amplification correction to the basic quantile correction values ​​of peak mode and ramp mode to obtain the load forecast error correction value. S215, calculate the device importance weight for each spatial grid; S216. Calculate the environmental risk factor for each spatial grid and each time period; S217. The deterministic load forecast value, the load forecast error correction value, the equipment importance weight and the environmental risk factor are integrated to obtain the spatial grid-level operation and maintenance demand intensity. S218. Traverse all spatial grids and all time periods to generate a spatial grid-level operation and maintenance demand intensity time series, and summarize it to form an operation and maintenance demand intensity map.

4. The intelligent dispatching method for operation and maintenance personnel based on power load forecasting according to claim 2, characterized in that, S31 specifically refers to: S311. Aggregate the spatial grid-level operation and maintenance demand intensity into the responsibility area-level operation and maintenance demand intensity. Directly sum the operation and maintenance demand intensity of all spatial grids belonging to the same responsibility area at the same time period to obtain the first... The responsibility area in the first The intensity of maintenance and operation needs at the responsibility area level during a given time period; S312. Establish skill level efficiency coefficients based on historical work order processing data; S313. Define the candidate personnel deployment plan and calculate the actual available operation and maintenance capabilities of the candidate personnel deployment plan in each responsibility area and time period. S314. Calculate the personnel shortage in each responsibility area and time period, and calculate the difference between the responsibility area-level operation and maintenance demand intensity and the actual available operation and maintenance capacity for each responsibility area and time period. If the difference is less than 0, it means that there is no shortage of personnel in the area of ​​responsibility during that period, and it is directly recorded as 0; if the difference is greater than 0, the difference is recorded as a shortage of personnel. S315. Calculate the demand abrupt change weight based on the change range of the intensity of maintenance demand in adjacent time periods at the responsibility area level. S316. Calculate the local garrison cost and cross-regional support cost of the candidate personnel deployment plan. For any candidate personnel deployment plan, accumulate the product of the number of personnel stationed in the local area and the corresponding local garrison unit cost coefficient for each responsibility area, time period, and skill level to obtain the local garrison cost; then accumulate the product of the number of cross-regional support personnel, average support distance, and unit support cost coefficient for each responsibility area, time period, and skill level to obtain the cross-regional support cost. S317. Combine the local garrison cost, cross-regional support cost, and emergency weighted deficit penalty into a fitness value for the candidate personnel deployment plan. The smaller the fitness value, the better the candidate personnel deployment plan. S318. Convert the fitness value into the light intensity of the firefly algorithm.

5. The intelligent dispatching method for operation and maintenance personnel based on power load forecasting according to claim 2, characterized in that, S32 specifically refers to: S321. Encode the candidate personnel deployment plan as firefly positions. Arrange the number of personnel stationed in the area in a fixed order for all responsibility areas, all time periods, and all skill levels to form the first part of the code. Then arrange the number of cross-area support personnel on all allowed support paths in a fixed order to form the second part of the code. Finally, combine the two parts to form a complete firefly position vector. S322. Generate an initial firefly population that satisfies the boundary constraints. Set an adjustable upper limit for each responsibility area and each skill level. Generate an initial value of personnel stationed in the area based on the intensity of maintenance needs at the responsibility area level. Randomly generate a small number of initial cross-area support personnel between the responsibility area pairs allowed by the cross-area support constraint mask matrix. Prune the portion that exceeds the adjustable upper limit and set the negative value directly to 0 to obtain an initial firefly population that satisfies the boundary constraints. S323. Calculate the time consistency weighted distance between fireflies. First, apply time smoothing weights to the personnel sequence of each responsibility area and each skill level. Then, calculate the distance between two fireflies. Finally, assign high weights to the differences between adjacent time periods and low weights to the differences between non-adjacent time periods. S324. Move the low-brightness fireflies toward the high-brightness fireflies and add random perturbations. S325. Perform feasible domain projection on the updated candidate deployment plan; S326. Recalculate the fitness value and light intensity of the projected candidate personnel deployment scheme, and update the current optimal candidate personnel deployment scheme. S327. Repeat the position update, feasible region projection and fitness update until the iteration termination condition is met; S328. Output the final personnel deployment results. The final personnel deployment results shall include at least the number of personnel stationed in each responsibility area at each time period, the number of cross-area support personnel between each responsibility area at each time period, the configuration quantity of each skill level, and the cross-area support flow matrix formed therefrom between responsibility areas.

6. The intelligent dispatching method for operation and maintenance personnel based on power load forecasting according to claim 1, characterized in that, Based on the responsibility area division results, a cross-regional support constraint mask matrix is ​​generated, specifically as follows: The elements in the cross-region support constraint mask matrix are denoted as . , used to indicate the first The responsibility area and the first Is it permissible to establish cross-regional support relationships between different responsibility zones? Traverse any two responsibility areas and If there exists at least one pair that belongs to a specific area of ​​responsibility and responsibility area Spatial grids with composite reachable edge weights If it exceeds the preset support threshold, then Set to 1; otherwise, Set to 0.

7. The intelligent dispatching method for operation and maintenance personnel based on power load forecasting according to claim 4, characterized in that, A skill level efficiency coefficient is established based on historical work order processing data, specifically as follows: Extract the average completion time, first-time repair rate, and rework rate of similar work orders from the historical work order system, processed by personnel of different skill levels. First, convert the average completion time into processing capacity per unit time. Then, use the first-time repair rate for positive correction and the rework rate for negative correction to obtain the result. The overall processing efficiency of personnel at each skill level is calculated, and then the overall processing efficiency of all skill levels is normalized by comparing it with the baseline skill level to obtain the skill level efficiency coefficient.

8. The intelligent dispatching method for operation and maintenance personnel based on power load forecasting according to claim 5, characterized in that, The feasible domain projection for the updated candidate deployment plan is as follows: Set all negative numbers to 0, and round all non-integer numbers according to preset rules. Check the cross-region support constraint mask matrix. If the cross-region support constraint mask value between a certain responsibility region pair is 0, then set all cross-region support personnel on that responsibility region pair to 0, indicating that support is not allowed to occur on this path. For each responsibility area, each time period, and each skill level, the sum of the number of personnel stationed in the area and the total number of support personnel transferred out is calculated. If the calculated value is greater than the maximum number of personnel that can be transferred out for that skill level in the responsibility area, the number of personnel stationed in the area is retained first, and the number of support personnel transferred out is reduced proportionally. If the number of personnel stationed in the area has already exceeded the maximum number of personnel that can be transferred out, the number of personnel transferred out is reduced to the maximum number of personnel that can be transferred out. For each area of ​​responsibility and each time period, the actual number of available personnel after transfers in and out is recalculated. If the number of available personnel in the area is negative, or if there are duplicate transfers from multiple directions causing duplicate statistics or inconsistencies in the total number of personnel, the deployment plan is iteratively adjusted according to the correction order of prioritizing local personnel, prioritizing support from neighboring areas, and then reducing support from distant areas, until all constraints of the deployment plan are met.

Images