Power resource scheduling method and electronic device

Abstract

The application provides a power resource scheduling method and an electronic device, which comprises the following steps: obtaining signaling and power data of a target area and aligning the time sequences to obtain signaling power alignment data; clustering the alignment data to obtain data clusters in multiple scenarios and determining a population flow coefficient and a signaling load mapping coefficient; taking a weighted sum of a network loss rate and a scheduling cost as an objective function, solving the objective function according to a power grid hard constraint condition, the population flow coefficient and the signaling load mapping coefficient to obtain multiple candidate scheduling schemes, performing multi-dimensional joint verification to screen qualified scheduling schemes, distributing weights according to scheduling performance indexes corresponding to the qualified scheduling schemes, and performing weighted summation on power scheduling parameters in the schemes according to the distributed weights to generate a target scheduling scheme; and scheduling power resources in each scenario according to the target scheduling scheme in each scenario. The application can quickly and accurately generate a scheduling scheme that can automatically adapt to different scenarios.

Description

Technical Field

[0001] This invention relates to the field of power dispatching technology, and more specifically, to a power resource dispatching method and electronic equipment. Background Technology

[0002] With the large-scale integration of distributed energy resources (such as rooftop photovoltaics and user-side energy storage) into the distribution network, Virtual Power Plants (VPPs), as intelligent platforms that aggregate and regulate multiple heterogeneous resources, have become a key technological path to improve the flexibility and economy of the power grid. Their core function is to generate dispatch instructions that meet safety constraints and are economically optimal based on real-time regional operational data, coordinating the collaborative responses of controllable units such as photovoltaic power output, energy storage charging and discharging, and adjustable loads.

[0003] However, in the existing technology, the virtual power plant real-time dispatch system still faces the following five interrelated objective technical defects that urgently need to be overcome: First, load forecasting heavily relies on static sources such as historical data and weather forecasts, resulting in large prediction errors in typical scenarios and inaccurate input benchmarks; Second, the parameters and rules of the MIP model lack regional characteristic adaptation mechanisms, and migration across business districts / residential areas / industrial areas requires manual recalibration, with adaptation cycles lasting up to several weeks and a significant decrease in accuracy; The above defects together restrict the system's ability to operate safely, economically, and efficiently under a high proportion of distributed energy access.

[0004] Therefore, how to enable the power grid dispatching system to complete power dispatching quickly and accurately, and to automatically adapt to different scenarios such as shopping malls, residential areas, and factories, is a technical problem that needs to be solved. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a power resource dispatching method and electronic device that enables the power grid dispatching system to complete power dispatching quickly and accurately, and can automatically adapt to different scenarios such as shopping malls, residential areas, and factories.

[0006] To achieve the above objectives, the technical solutions adopted in the embodiments of the present invention are as follows: In a first aspect, the present invention provides a power resource scheduling method, the method comprising: obtaining signaling data and power data of a target area in the current time period, and performing time-series alignment on the signaling data and power data to obtain signaling-power aligned data; clustering the signaling-power aligned data to obtain data clusters under multiple scenarios, and determining the population flow coefficient and signaling load mapping coefficient under each scenario based on each data cluster; for each scenario, using the weighted sum of minimizing network loss rate and scheduling cost as the objective function, solving the objective function according to preset hard constraints of the power grid, the population flow coefficient and the signaling load mapping coefficient to obtain multiple candidate scheduling schemes; performing multi-dimensional joint verification on the multiple candidate scheduling schemes corresponding to each scenario to screen out qualified scheduling schemes; assigning weights according to the scheduling performance indicators corresponding to each qualified scheduling scheme, and performing weighted summation on the power scheduling parameters in each qualified scheduling scheme according to the assigned weights, and generating a target scheduling scheme by combining the weighted summation results of each scheduling parameter; and scheduling power resources under each scenario according to the target scheduling scheme under each scenario in the target area.

[0007] In a second aspect, the present invention provides an electronic device including a processor and a memory, wherein the memory stores machine-executable instructions that can be executed by the processor, and the processor can execute the machine-executable instructions to implement the power resource scheduling scheme described in any of the foregoing embodiments.

[0008] This invention provides a power resource scheduling method and electronic device. First, it acquires signaling and power data for the target area in real time and aligns them temporally to ensure precise matching at the time granularity, providing a joint sensing foundation for subsequent modeling. Second, it performs clustering based on the aligned data to naturally identify differentiated data clusters corresponding to typical power consumption scenarios. Then, it separately labels the population flow coefficient and signaling load mapping coefficient specific to each scenario, transforming load forecasting from rough empirical estimation to fine-grained interpretable modeling of scenarios, significantly improving the accuracy of load response. Based on this, it independently constructs... An optimization model is constructed with the goal of minimizing the weighted sum of network loss rate and scheduling cost. The model incorporates real power grid hard constraints by combining the two coefficients mentioned above, ensuring that the generated candidate schemes are both physically feasible and cost-effective. Furthermore, qualified schemes are selected through multi-dimensional joint verification to ensure the accuracy and executability of scheduling instructions. Finally, the key performance indicators of each qualified scheme are weighted and the scheduling parameters are weighted and fused to generate the optimal target scheduling scheme. The entire process can enable the power grid scheduling system to quickly and accurately complete power dispatch without manual configuration and can automatically adapt to different scenarios such as shopping malls, residential areas, and factories.

[0009] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A diagram showing the functional modules of a traditional power dispatching system; Figure 2 A schematic flowchart illustrating a power resource scheduling method provided in an embodiment of the present invention; Figure 3 This is an example diagram illustrating the timing alignment of signaling data and power data according to an embodiment of the present invention; Figure 4 This is an example diagram of scene clustering provided in an embodiment of the present invention; Figure 5 An example diagram illustrating the composition of the scheduling scheme generated in an embodiment of the present invention; Figure 6 Examples of labeled samples provided in embodiments of the present invention; Figure 7 A functional block diagram of a power resource dispatching system provided in an embodiment of the present invention; Figure 8 This is a structural block diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0013] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0014] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0015] Please see Figure 1 , Figure 1 This diagram illustrates the functional modules of a traditional power dispatching system, including a traditional data acquisition module, a basic data preprocessing module, a MIP dispatching optimization module, and a dispatching scheme output module. The dispatching process of this power dispatching system is as follows: The traditional data acquisition module collects static / semi-static data such as photovoltaic power output forecast data (based on meteorological data), energy storage device status data (remaining power, charge and discharge thresholds), historical load data, and electricity price data, and provides this data to the basic data preprocessing module. The basic data preprocessing module deduplicates and smooths the collected traditional data, calculates surface statistics such as historical load averages and photovoltaic power output fluctuation ranges, and provides the preprocessed data to the subsequent MIP scheduling optimization module. The MIP scheduling optimization module constructs an MIP model with transmission capacity constraints, supply-demand balance constraints, and voltage deviation constraints as the objective function, directly solving the NP-hard problem of virtual power plant scheduling. The scheduling scheme output module directly outputs the MIP model solution results.

[0016] The inventors discovered during the research process that: the above Figure 1 The power grid dispatching system shown has the following defects in the power dispatching process: Disadvantage 1: Relying solely on static data such as historical load and weather conditions completely fails to reflect dynamic load drivers on the user side (such as regional population flow and real-time equipment activity status). When sudden scenarios such as large gatherings of people in commercial centers or concentrated start-ups and shutdowns of equipment in residential areas occur, static data cannot respond in a timely manner, leading to increased load forecasting errors. This, in turn, causes the input benchmark of the MIP model to become inaccurate, resulting in an inherent bias in the scheduling scheme.

[0017] Disadvantage 2: Existing mixed-integer programming (MIP) models use constraint parameters and load forecasting rules based on historical data from a single region, failing to consider the dynamic load characteristics differences between different regions (core urban / suburban areas, industrial / commercial areas) (e.g., dense population and frequent load fluctuations in core urban areas, while suburban areas have stable loads). When applied across regions, historical data needs to be manually re-collected and constraint parameters adjusted, resulting in an adaptation period of several weeks. Furthermore, the lack of dynamic data calibration after migration leads to decreased forecast accuracy, making it unsuitable for large-scale deployment.

[0018] Disadvantage 3: The constraint parameters (such as transmission capacity threshold and load boundary) of the existing MIP model are fixed values ​​and do not change dynamically with load. When the load fluctuates, the fixed constraints will cause the feasible solution space of the MIP model to shrink significantly or become redundant, requiring repeated iterations to find a compliant solution, further increasing the solution time and exacerbating the problem of solving NP-hard problems.

[0019] To address the aforementioned issues and enable the power grid dispatching system to quickly and accurately complete power dispatching, while automatically adapting to different scenarios such as shopping malls, residential communities, and factories, please refer to [link to relevant documentation]. Figure 2 , Figure 2 A schematic flowchart of a power resource scheduling method provided in an embodiment of the present invention includes steps S201 to S206, as described below: S201: Obtain the signaling data and power data of the target area in the current time period, and perform time-series alignment on the signaling data and power data to obtain signaling-power aligned data; S202: Cluster the signaling power alignment data to obtain data clusters under multiple scenarios, and determine the population flow coefficient and signaling load mapping coefficient under each scenario based on each data cluster; S203: For each scenario, the objective function is to minimize the weighted sum of network loss rate and scheduling cost. The objective function is solved based on the preset hard constraints of the power grid, population mobility coefficient and signaling load mapping coefficient to obtain multiple candidate scheduling schemes. S204: Perform multi-dimensional joint verification on multiple candidate scheduling schemes for each scenario, and select qualified scheduling schemes. S205: Based on the scheduling performance indicators corresponding to each qualified scheduling scheme, weights are allocated, and the power scheduling parameters in each qualified scheduling scheme are weighted and summed according to the allocated weights. The target scheduling scheme is generated by combining the weighted sums of each scheduling parameter. S206: Dispatch the power resources for each scenario according to the target scheduling plan for each scenario within the target area.

[0020] Unlike existing technologies, this invention first acquires signaling and power data for the target area in real time and aligns them temporally to ensure precise matching at the time granularity, providing a joint sensing foundation for subsequent modeling. Secondly, based on the aligned data, clustering is performed to naturally identify differentiated data clusters corresponding to typical power consumption scenarios. Then, population flow coefficients and signaling load mapping coefficients specific to each scenario are labeled, transforming load forecasting from rough empirical estimation to fine-grained, interpretable modeling at the scenario level, significantly improving the accuracy of load response. Furthermore, for each scenario, a weighted sum of network loss rate and scheduling cost is independently constructed. The optimization model, with minimization as its objective, incorporates real-world power grid hard constraints by embedding the aforementioned two coefficients, ensuring that the generated candidate solutions are both physically feasible and cost-effective. Furthermore, qualified solutions are selected through multi-dimensional joint verification to guarantee the timely executability of dispatch instructions. Finally, based on the weights allocated to key performance indicators for each qualified solution, and the dispatch parameters are weighted and fused to generate a target dispatch solution that balances robustness and optimality. The entire process can adapt to load characteristics under different scenarios without manual configuration, thereby supporting the power grid dispatch system to complete highly accurate, robust, and adaptive power dispatch across all scenarios at the millisecond to minute level.

[0021] Next, the embodiments of the present invention will provide a detailed description of the provided power resource scheduling process in conjunction with the relevant accompanying drawings.

[0022] In this embodiment of the invention, the target area can be any distribution network area with virtual power plant dispatching capabilities, including but not limited to urban core business districts, new industrial parks, densely populated residential communities, or mixed power supply areas containing multiple load types. This embodiment of the invention does not limit this. The current time period is the time period during which power dispatching is required.

[0023] In this embodiment of the invention, the signaling data is regional-level anonymized signaling statistics provided by a compliant signaling service provider. The field format may include the following: total number of signaling interactions, base station coverage area label (this data will serve as the core identifier for scenario division and associate base station types, such as macro base stations in commercial areas, micro base stations in residential areas, macro base stations in industrial parks, and base stations in exhibition / remote areas), and a 1ms precision timestamp. The signaling data needs to undergo IMEI hashing and location information obfuscation (accuracy ≤ 100 meters) processing to meet compliance and anonymization requirements, and comply with the Personal Information Protection Law and GDPR requirements.

[0024] In this embodiment of the invention, the power data comes from the grid side and may include, but is not limited to: regional load (unit: MW), energy storage SOC (unit: %), and photovoltaic output (unit: kW).

[0025] Optionally, the signaling data in the embodiments of the present invention needs to meet the following requirements: the transmission delay is less than 100 ms, the data loss rate is less than or equal to 2%, the power data accuracy needs to retain 1 decimal place, and there is no abnormal jump value (the single fluctuation amplitude is less than or equal to 5%).

[0026] In the embodiments of the present invention, for the obtained signaling data, the criterion can be used to eliminate extreme outliers of the signaling interaction times (for example, if a base station suddenly has 1 million interactions within 1 ms, it is obviously a fault). For the missing data, the mean value of adjacent time points is used to complete it, providing an accuracy guarantee for subsequent time series alignment and calculating the population flow coefficient and signaling load mapping coefficient for different scenarios.

[0027] To avoid the problem of high load prediction error caused by the time series asynchronization between signaling data and power data, in step S201, the signaling data and power data can be aligned in time series to obtain signaling-power aligned data. The specific implementation is as shown in steps a1 to a3: Step a1: Align all power data timestamps with the signaling timestamps according to the time accuracy of the signaling data; Exemplarily, assume that the time accuracy of the signaling data is 1 ms, while the original time accuracy of the power data is 100 ms. Then, based on 1 ms, round all the timestamps of the power data to the nearest 1 ms scale and correspond them to the signaling timestamps one by one.

[0028] Step a2: For the time points without corresponding power data at this time accuracy, use the linear interpolation method to complete the power data corresponding to the time points; In the embodiments of the present invention, for the 1 ms time points without corresponding power data, use the linear interpolation method (like drawing a straight line connecting two known points) to complete the values. The interpolation formula is shown in formula (1) below:

[0029] where x is a certain time point without corresponding power data at the time accuracy, x0 and x1 are adjacent power data timestamps (x0 < x < x1), and y0 and y1 are the power data corresponding to x0 and x1 respectively.

[0030] Step a3: Obtain the signaling-power aligned data from the power data and signaling data corresponding to each time point at the time accuracy.

[0031] Finally, in the embodiments of the present invention, the signaling data and power data corresponding to each time point at the signaling time accuracy, that is, the signaling-power aligned data, can be obtained.

[0032] For the convenience of intuitively understanding the above time series alignment process, please refer to Figure 3 , Figure 3This is an example diagram illustrating the timing alignment of signaling data and power data in an embodiment of the present invention. Through the above steps a1 to a3, the embodiment of the present invention can realize the legitimate, accurate, and millisecond-level access and parsing of external social activity data by the power system, and use the signaling information for subsequent load forecasting, providing a reliable starting point for the entire system.

[0033] Optionally, to maintain system-level timing synchronization accuracy even with slight delays or jitter in external data sources, and to avoid cascading deviations in subsequent load forecasting and scheduling decisions due to single-point data anomalies, this embodiment of the invention can also monitor timing alignment errors in real time. If the error is less than a preset error threshold (e.g., 50ms), it is considered acceptable; otherwise, a data anomaly is flagged and fed back to the signaling service provider, triggering realignment. Therefore, after step a2, the alignment error can be determined based on the signaling data timestamp and the power data timestamp. When the error exceeds the preset threshold, a data anomaly feedback signal is sent to obtain the grid standard timestamp. Linear interpolation is performed based on the time precision of the signaling data and the grid standard timestamp to update the power data corresponding to the time point, ensuring data quality. Simultaneously, this invention ensures that timing accuracy meets the standards through an internal fallback mechanism.

[0034] In the above embodiments, the present invention employs a dual-reference alignment method, which re-interpolates based on the time precision of the signaling data and the standard timestamp of the power grid, controlling the alignment error within a preset threshold. After the service provider corrects the data, a data re-alignment process is automatically triggered to update the signaling power alignment data, and the verification frequency is temporarily increased (e.g., every 500ms) to dynamically avoid subsequent anomalies. This method ensures that the timing alignment accuracy continuously meets the signaling accuracy reference requirements, providing reliable data support for subsequent calculations of population flow coefficients and signaling load mapping coefficients in different scenarios.

[0035] It should be noted that the embodiments of the present invention are not long-term predictions in the traditional sense, but rather utilize the strong spatiotemporal correlation between signaling data and power load and the millisecond-level sensing advantage of signaling to achieve near real-time mapping of power load fluctuations.

[0036] Based on the signaling power alignment data obtained in the preceding steps, in order to improve the cross-domain adaptability of the scheduling scheme, in step S202 of this embodiment of the invention, by scenario-based clustering, combined with clustering initialization guided by base station coverage area labels and contour coefficient optimality verification, the entire domain data is dynamically divided into different scenario data clusters, and population flow coefficients are independently calculated for each cluster. ) and signaling load mapping coefficient ( The modeling and calibration of the system allows for automatic adaptation to the fundamental differences in population flow characteristics, load response sensitivity, and operational inertia across different regions without relying on manual experience for parameter tuning, thus providing accurate input for the subsequent generation of target scheduling schemes.

[0037] In step S202, this embodiment of the invention first clusters the signaling power alignment data to obtain data clusters under multiple scenarios. Specifically, this invention can cluster the signaling power alignment data according to signaling activity characteristics and load characteristics, thereby obtaining data clusters under multiple scenarios. The signaling activity characteristics can be normalized signaling interaction counts, with a normalization range of [0,1]; the load characteristics can be normalized regional load values, with a normalization range of [0,1]).

[0038] Optionally, embodiments of the present invention may employ clustering algorithms such as K-means. Initial cluster centers are set using a clustering algorithm, and clustering is completed quickly by setting iteration stopping conditions.

[0039] To facilitate an intuitive understanding of the above clustering results, such as Figure 4 As shown, Figure 4 This is an example diagram of scene clustering provided in an embodiment of the present invention. The K-means algorithm is used, with the number of cluster centers set to k=4. The iteration stopping condition is that the cluster center offset is less than 0.001, and the maximum number of iterations is 50. Finally, four homogeneous scenes are obtained: commercial peak, residential off-peak, industrial stable, and extreme scenes. Each scene cluster corresponds to a base station coverage area label, which can be directly used as the basis for validating the clustering results (e.g., signaling data labeled "commercial area macro base station" needs to be clustered into the commercial peak scene). Experimental verification shows that scene discrimination is optimal when k=4, with a silhouette coefficient of 0.82. The silhouette coefficients for k=2 / 3 / 5 are 0.58 / 0.71 / 0.78 respectively. Compared to a single global feature, scene-based processing improves the matching degree between signaling features and load change patterns.

[0040] In this embodiment of the invention, for each scenario, the following can be calculated separately: and Prior to this, embodiments of the present invention may first set a benchmark number of permanent residents for each scenario. as well as Each corresponds to a reasonable range. The baseline permanent resident population is directly used in the calculation. ,and and Their respective reasonable ranges are then used to calculate the... and Each value is verified separately. These values ​​are determined by combining household registration data, base station coverage area, and base station coverage area labels. The labels are directly associated with the baseline resident population for each scenario (e.g., the label "residential micro base station" corresponds to a baseline value of 15,000 residents in an off-peak scenario). The baseline resident population for each scenario... and The specific reasonable ranges for each are shown in Table 1 below.

[0041] Table 1

[0042] It should be noted that the reference permanent resident population figures in Table 1 are different. , The specific values ​​of their respective reasonable ranges and anomaly detection conditions are merely examples and not intended to limit the embodiments of this invention. , Each of these parameters, along with their respective reasonable ranges and anomaly detection conditions, can be flexibly configured by relevant technical personnel according to actual needs; no restrictions are imposed here.

[0043] Based on the above scenario-based differentiated numerical settings, the following details the process by which the embodiments of the present invention determine the population flow coefficient and signaling load mapping coefficient for each scenario.

[0044] First, for each scenario, the population mobility coefficient represents the number of signaling interactions in that scenario divided by the resident population covered by the base station, i.e., the intensity coefficient of population mobility. This invention proposes the following implementation method to determine the population mobility coefficient. : Step b1: Use the ratio of the total number of signaling interactions in the data cluster to the number of permanent residents covered by the base station in the corresponding scenario of the data cluster as the population mobility coefficient; In this embodiment of the invention, for each scenario i, the population mobility coefficient The calculation process is shown in formula (2):

[0045] Step b2: If the population mobility coefficient exceeds the reasonable range of the population mobility coefficient in the scenario corresponding to the data cluster, then determine the signaling fluctuation coefficient and population mobility coefficient correction parameters in the scenario corresponding to the data cluster; otherwise, no action is taken. In this embodiment of the invention, for the calculation in step b1 It can also be judged If the population flow coefficient exceeds the reasonable range pre-set for scenario i, continue with the subsequent process if it does not exceed the limit; otherwise, determine the correction parameters to correct it.

[0046] In this embodiment of the invention, the signaling fluctuation coefficient The fluctuation ratio of the number of real-time signaling interactions to the historical average for the same period can be calculated as follows: Set a sliding window of a preset duration, such as 1 minute; iterate through the signaling data of the current period in the sliding window according to a preset step size (such as 10 seconds, adapted to the characteristics of 1ms high-precision signaling data), and count the total number of real-time signaling interactions within the sliding window during each iteration. Take the historical average number of signaling interactions with the sliding window within the same historical time period in the same scenario. (Exclude data from abnormal dates such as extreme weather and large-scale events); then calculate the signaling fluctuation coefficient according to the following formula (4):

[0047] in, A positive value indicates that the real-time signaling is higher than the historical average, while a negative value indicates that it is lower than the historical average.

[0048] In this embodiment of the invention, the population mobility coefficient correction parameter ( (∈[0.8,1.2]), can be obtained as follows: Step 1: Obtain historical signaling power alignment data for the scenario corresponding to this data cluster; Step 2: Select multiple valid data groups from historical signaling power alignment data; In this embodiment of the invention, historical signaling power alignment data refers to signaling power alignment data within a historical time period of a preset duration from the current time period, such as the past year. For each scenario, the historical signaling power alignment data can be first divided into multiple groups, and then... The criteria are to remove outliers and retain valid data sets. For example, if the data is split by month, there are 12 groups / scenarios, and n valid data sets are retained (n is less than or equal to 12).

[0049] Step 3: Calculate the historical population mobility coefficient for each valid data set, and use the historical population mobility coefficient and the actual population mobility coefficient corresponding to each valid data set to calculate the fitting deviation; In this embodiment of the invention, the historical population mobility coefficient can be calculated using formula (2), while the actual population mobility coefficient is obtained from real-time population statistics. The fitting deviation of the j-th effective data set is... The calculation formula (3) is shown below:

[0050] Step 4: Determine the correction parameters by averaging all fitting deviations.

[0051] Take the fit deviation for each valid data set j The average value δ_avg, the correction parameter for the population mobility coefficient. .

[0052] Step b3: Fine-tune the population mobility coefficient correction parameter based on the signaling fluctuation coefficient in the corresponding scenario of the data cluster, and then correct the population mobility coefficient using the fine-tuned population mobility coefficient correction parameter.

[0053] based on The formula (5) for fine-tuning the correction parameter of the population mobility coefficient is shown below:

[0054] in, This refers to the population mobility coefficient correction parameter after fine-tuning in this embodiment of the invention.

[0055] In this embodiment of the invention, before correcting the population mobility coefficient using the fine-tuned population mobility coefficient correction parameter, it is also possible to first determine... Does it exceed the preset value range, such as [0.8, 1.2]? If it does, then... The boundary values ​​within a preset range can be used (e.g., 0.8 for values ​​below 0.8, and 1.2 for values ​​above 1.2). Finally, the population flow coefficient for each scenario i can be obtained using the following formula (6):

[0056] Optionally, embodiments of the present invention may further determine the generated If the population mobility coefficient is within a reasonable range for scenario i, then retain the currently calculated value. Unchanged; if it exceeds, then... Replace it with the boundary value of the reasonable range of the population mobility coefficient. For example, assuming scenario i corresponds to a business peak, refer to the reasonable range of the population mobility coefficient [1.5, 1.8] in Table 1. When If <1.5, then take =1.5, when If the value is greater than 1.8, then take... =1.8. This is how to ensure... It can accurately reflect the population flow characteristics in scenario i, providing a reliable basis for the adaptation of subsequent scheduling schemes.

[0057] The following describes the scenario-specific calculation of the signaling load mapping coefficient in embodiments of the present invention. The process, To determine the load change magnitude corresponding to a unit signaling interaction, a quantitative relationship between population flow dynamics and power load is established. The calculation process is shown in steps c1 to c4 below: Step c1: Determine the minimum and maximum number of signaling interactions from the data cluster, and predict the maximum and minimum load based on the maximum and minimum values; In this embodiment of the invention, multiple random forest regression models, matching the number of scenarios, can be trained in parallel, with one model corresponding to each scenario.

[0058] For example, model parameters can be uniformly set as follows: number of decision trees n_estimators=100, maximum depth 10, minimum number of sample splits 5, minimum number of sample leaf nodes 2, and feature sampling method (m=1, only taking the number of signaling interactions); the training process can use 5-fold cross-validation, the training data is joint signaling load data, and the early stopping mechanism can be set to stop training when the validation set load prediction error decreases to less than or equal to 0.001 for three consecutive rounds to avoid overfitting. Of course, the above model parameter settings are just an example. Those skilled in the art can adjust the settings according to the actual data scale, scenario complexity, and deployment resource constraints, while ensuring... Under the premise of ensuring both the accuracy of coefficient extraction and the real-time performance of calculation, adaptive adjustments are made to hyperparameters such as the number of decision trees, depth, and pruning threshold.

[0059] For each scenario i, the minimum number of signaling interactions can be extracted. and maximum value Then and Input the data into the random forest regression model corresponding to this scenario to make predictions and obtain the minimum load. and maximum load ).

[0060] Step c2: Calculate the signaling load mapping coefficient from the minimum and maximum number of signaling interactions and the maximum and minimum load values; In this embodiment of the invention, the following formula (7) can be used to calculate: :

[0061] Optionally, embodiments of the present invention may also be based on each scenario i. Calculate the global Calculated by weighting the annual load percentage of each scenario As shown in formula (8):

[0062] Of which: load percentage Based on the electricity data statistics of the target area over the past year, for example, assuming I=4, then the load share is... It could be: (Business Peak) = 35% (Peak Residential Hours) = 40% (Industrial stability) = 20% (Extreme scenario) = 5%.

[0063] Step c3: If the signaling load mapping coefficient exceeds the reasonable range of the signaling load mapping coefficient in the scenario corresponding to the data cluster, then determine the load fluctuation coefficient and signaling load mapping coefficient correction parameters in the scenario corresponding to the data cluster; otherwise, no action is taken. In this embodiment of the invention, for the calculation in step c2 It can also be judged If the signaling load mapping coefficient exceeds the reasonable range pre-set for scenario i, continue with the subsequent process if it does not exceed the limit; otherwise, a correction parameter needs to be introduced to correct it.

[0064] In this embodiment of the invention, the signaling fluctuation coefficient is calculated. Similarly, the same sliding window method can be used to iterate through the power data for the current time period and calculate the real-time regional average load. At the same time, the historical average load was calculated. Then, the load fluctuation coefficient is calculated using the following formula (10). :

[0065] in, A positive value indicates that the real-time load is higher than the historical average, while a negative value indicates that it is lower than the historical average.

[0066] In this embodiment of the invention, the signaling load mapping coefficient correction parameter is denoted as... ( ∈[0.8,1.2]), the calculation method is the same as Similarly, as mentioned earlier, multiple valid data groups need to be selected from historical signaling power alignment data. The difference lies in: for Then, it is necessary to calculate the historical signaling load mapping coefficient for each valid data group, and use the historical signaling load mapping coefficient and the actual signaling load mapping coefficient corresponding to each valid data group to calculate the fitting deviation. The actual signaling load mapping coefficient is also obtained from real-time load statistics from relevant departments. The fitting deviation for the j-th valid data group... The calculation formula (9) is shown below:

[0067] Finally, the fit deviation for each valid data set j is taken. average _avg, correct parameter .

[0068] Next, the correction parameters are fine-tuned based on the load fluctuation coefficient in the corresponding scenario, and the signaling load mapping coefficient is corrected by the fine-tuned correction parameters. .

[0069] based on right Make fine adjustments, as shown in formula (11): (11) in, This refers to the signaling load mapping coefficient correction parameter after fine-tuning in this embodiment of the invention.

[0070] In this embodiment of the invention, before correcting the signaling load mapping coefficient with the fine-tuned signaling load mapping coefficient correction parameter, it is also possible to first determine... Does it exceed the preset value range, such as [0.8, 1.2]? If it does, then... The boundary values ​​within a preset range can be taken (e.g., 0.8 for values ​​below 0.8, and 1.2 for values ​​above 1.2). Finally, the value corresponding to each scene i can be obtained using the following formula (12). :

[0071] Optionally, embodiments of the present invention may further determine the generated If the signaling load mapping coefficient is within a reasonable range for scenario i, then retain the currently calculated value. Unchanged; if it exceeds, then... Replace it with the boundary value of the reasonable range of the signaling load mapping coefficient. For example, refer to Table 1. Assuming scenario i corresponds to a commercial peak, the reasonable range of the corresponding signaling load mapping coefficient is [0.002, 0.003] MW / time. If <0.002, then take =0.002, when If the value is greater than 0.003, then take... =0.003. This is how to ensure... It can accurately reflect the signaling load mapping characteristics in scenario i, providing a reliable basis for subsequent scheduling scheme adaptation.

[0072] Through the above implementation methods, the 1ms precision timing alignment method based on external signaling data controls the timing synchronization error within 50ms through linear interpolation and error verification, solving the load forecasting lag problem caused by the second-level alignment in the existing technology, while reducing the acquisition cost. A scenario-specific α / β feature extraction and calibration mechanism is proposed. Through K-means scenario clustering, interval validity verification, and historical data calibration, accurate feature parameters adapted to different regions are generated, reducing the load forecasting error to below 0.9%.

[0073] Based on the signaling power alignment data obtained in the previous steps and the population flow coefficient for each scenario And signaling load mapping coefficient In step S203, multiple candidate scheduling schemes can be generated for each scenario by combining preset power grid hard constraints.

[0074] In this embodiment of the invention, the objective function is set as minimizing the weighted sum of network loss rate and scheduling cost, as shown in formula (13):

[0075] Where f is the objective function value; , The weighting coefficients for network loss rate and scheduling cost are respectively (satisfying) + =1, which can be dynamically allocated according to the power grid security priority. For example, during peak periods, priority is given to ensuring security. =0.6、 =0.4; Cost control should be prioritized during off-peak periods, and can be taken as... =0.4、 =0.6).

[0076] In formula (13), The network loss rate (unit: %) is physically defined as "the proportion of power lost during power grid transmission to the total power supplied" (in accordance with the power industry network loss rate standard definition). It is calculated as shown in the following formula (14):

[0077] in, The total power supply (unit: MW) for the target area, that is, the total power output of all power supply resources such as photovoltaics and energy storage (total output on the power supply side), can be directly obtained from...; The total power loss of the power grid line (unit: MW) is calculated from the line resistance and transmission current. The specific calculation process is shown in formula (15):

[0078] In the formula: Let be the transmission current of line ij, in A. (where is the resistance of line ij, in Ω).

[0079] In formula (13), The unit time scheduling cost (unit: yuan / minute) is calculated using the following formula:

[0080] in, Unit output cost of distributed photovoltaic (unit: yuan / MW·minute); Real-time photovoltaic power output (unit: MW); The charging and discharging cost of energy storage units (unit: yuan / MW·minute); The cost reduction per unit of industrial load (unit: yuan / MW·minute, i.e., the economic cost required to reduce 1MW of industrial load per minute, including enterprise production reduction compensation, equipment start-up and shutdown losses, etc.) are all known quantities. Power reduction for industrial load (this is the power controllable by the power grid, i.e. whether power is output to industrial self-consumption units and how much power is output) is one of the variables to be solved in this embodiment of the invention (unit: MW, only takes a value when industrial load needs to be reduced, and is 0 when there is no reduction). , =2MW.

[0081] In formula (16), The energy storage charging and discharging power, i.e. the power generation to determine the energy storage discharge power and charging power, is another variable to be solved in this embodiment of the invention (unit: MW, discharge is positive and charging is negative, taking the absolute value can unify the calculation of charging and discharging cost). , =10MW. For ease of distinction, the energy storage discharge power can be used subsequently... It is indicated that the energy storage charging power can be used in the future. express.

[0082] In the process of solving the objective function, this embodiment of the invention sets several hard constraints that the power grid operation must meet, including but not limited to: supply and demand balance, energy storage SOC, photovoltaic output, industrial load reduction, signaling adaptation, transmission capacity, voltage constraints and voltage deviation. These constraints together constitute the physical feasibility boundary of the scheduling scheme.

[0083] It should be noted that the above-mentioned constraint types are only typical examples and not an exhaustive list; any other key power parameters that have a substantial impact on the safe and stable operation of the power grid can be included in the hard constraint system constructed in this embodiment of the invention, according to the needs of the actual scenario. The core of this embodiment of the invention is that by selecting only a few key, quantifiable hard constraints, it is possible to effectively ensure that the solution results not only meet the actual engineering requirements, but also support the reliable solution of the objective function (i.e., the weighted sum of network loss rate and scheduling cost minimization).

[0084] These constraints are explained below.

[0085] 1. Supply and demand balance constraint: The total power supplied must be equal to the real-time power and the total power loss of the power grid lines, as expressed by formulas (17) to (20):

[0086]

[0087]

[0088]

[0089] in, Indicates the total power supply. Indicates real-time power. This refers to the signaling load mapping coefficient in the current scenario. The total power loss of the power grid line is represented by r; the photovoltaic output allocation ratio is a variable to be solved in this embodiment of the invention. The amount of signaling interaction per unit time; The load is calibrated for historical periods and is a known quantity; The total regional load (MW) For energy storage discharge power, For energy storage charging power, The power reduction for industrial self-use load is represented by the amount of electricity used for industrial self-use by the generating unit (MW). Formulas (18) and (20) are two other ways of representing formula (17).

[0090] 2. Energy Storage SOC Constraint: The battery capacity must be within a preset SOC range, for example, the battery capacity cannot be lower than 20% or higher than 90%. This can be expressed by formula (21):

[0091] in, For the initial SOC (%), , The values ​​are the charge and discharge efficiencies (taken as 0.95). For charging time, Discharge duration Rated energy storage capacity (MW); [] represents the preset SOC range.

[0092] 3. Photovoltaic output constraint: The actual photovoltaic power used is within the preset power range, expressed by formula (22):

[0093] in, and This indicates the minimum and maximum battery capacity.

[0094] 4. Industrial load reduction constraint: The industrial load reduction shall not exceed 10% of the rated industrial load, as expressed by formula (23):

[0095] 5. Signaling adaptation constraint: The product of the population mobility coefficient and the total regional load in the current scenario is less than or equal to the upper limit of the transmission capacity, expressed by formula (24):

[0096] in, The total regional load (MW) For the upper limit of transmission capacity (e.g.) ).

[0097] This invention, through the addition of dynamic constraints adapted to signaling, allows constraint parameters to be adjusted in real time with the signaling α coefficient, dynamically optimizing the feasible solution space, reducing the number of iterations in the solution process, improving solution efficiency, ensuring constraint compliance, and laying an efficient foundation for subsequent distillation optimization.

[0098] 6. Transmission capacity constraint: The power transmitted by the power grid lines cannot exceed the upper limit of the transmission capacity (e.g., 100MW). 7. Voltage Constraint: The actual voltage of any power supply node must be within the safe fluctuation range, expressed by formula (25):

[0099] in: It is the actual voltage of power supply node i. It is the transmission power of each branch. The branch lines are blocked, [ , [ ] represents the safe fluctuation range, which are the maximum and minimum voltages at node i, respectively. Its design must refer to the requirements of "GB50217-2018 Standard for Design of Cables for Power Engineering". It is the reference operating voltage for power grid / equipment design.

[0100] 8. Voltage Deviation Constraint: The actual voltage deviation at any power supply node shall not exceed a preset deviation threshold (e.g., ...). ), which can be expressed by formula (26):

[0101] The embodiments of the present invention can use the existing branch and bound method to efficiently search for K feasible solutions in the feasible space composed of hard constraints of the power grid. Each feasible solution is a candidate scheduling scheme, which includes the specific values ​​of the three variables to be solved: energy storage charging and discharging power, industrial load reduction power and photovoltaic output allocation ratio, as well as the corresponding network loss rate, cost and other performance labels.

[0102] Optionally, in embodiments of the present invention, step S203 can be accomplished using a mixed integer programming (MIP) model.

[0103] To facilitate a more intuitive understanding of the candidate scheduling schemes generated by the embodiments of the present invention, please refer to [link / reference needed]. Figure 5 , Figure 5 This is an example diagram illustrating the composition of the scheduling scheme generated in an embodiment of the present invention. It can be seen that, in peak commercial scenarios, based on the given signaling data and power data, scheduling parameters can be generated, including photovoltaic output settings (i.e., the product of the photovoltaic output allocation ratio and the real-time photovoltaic output), energy storage charging / discharging power, industrial load reduction power, and the photovoltaic output allocation ratio. Simultaneously, the scheduling performance indicators of this scheduling scheme can be generated, including the grid loss rate and the scheduling cost per unit time (yuan / minute).

[0104] Based on the multiple candidate scheduling schemes generated for each scenario in the preceding step S203, in order to ensure that the selected schemes have both security, compliance and scenario adaptability in the real operating environment, and to avoid constraint breaches or policy mismatches caused by idealized assumptions of the model, in step S204 of this embodiment of the invention, it is also necessary to perform multi-dimensional joint verification on these candidate scheduling schemes, that is, to perform signaling adaptability verification, power grid constraint verification and scenario adaptability verification on the candidate scheduling schemes respectively; if the candidate scheduling scheme passes each verification, it is determined to be a qualified scheduling scheme, otherwise it is an unqualified scheduling scheme and is eliminated.

[0105] In embodiments of the present invention, such as Figure 5 As shown, each candidate scheduling scheme is associated with signaling power alignment data. The scheme content includes a set of scheduling scheme parameters (including energy storage charging / discharging power, industrial load reduction power, and photovoltaic output allocation ratio) and scheme performance. Based on this, this embodiment of the invention designs multi-dimensional joint verification, including: signaling adaptability verification, grid constraint verification, and scenario adaptability verification.

[0106] (a) Signaling compatibility verification: Based on the scenario label associated with the candidate scheduling scheme and the corresponding population flow coefficient, determine whether the population flow coefficient is within the reasonable range of the preset population flow coefficient under the scenario label.

[0107] (b) Grid Constraint Verification: This involves determining the actual grid constraint parameters upon which the candidate dispatch schemes depend and verifying whether they meet the hard grid constraints, such as supply-demand balance constraints, energy storage SOC constraints, photovoltaic output constraints, industrial load reduction constraints, signaling adaptation constraints, transmission capacity constraints, voltage constraints, and voltage deviation constraints. For example, voltage deviation... Transmission capacity ≤ 100MW, 20% ≤ SOC ≤ 90%, etc., with no hard constraints to overcome.

[0108] (c) Scenario adaptability verification: Based on the signaling load mapping coefficient associated with the candidate scheduling scheme, evaluate whether its load prediction error in the corresponding scenario is lower than the preset threshold; when the load prediction error is less than the preset error threshold (e.g., 0.9%), it is determined to be satisfied; otherwise, it is not satisfied.

[0109] Finally, candidate scheduling schemes that fail any dimension of verification are eliminated, and schemes that pass all three dimensions of verification are retained as qualified scheduling schemes.

[0110] Optionally, when the number of qualified scheduling schemes is less than a preset threshold (e.g., 3), the candidate scheduling scheme is regenerated.

[0111] Based on the qualified scheduling schemes selected in the preceding steps, in step S205 of this embodiment of the invention, weights can be allocated according to the scheduling performance indicators corresponding to each qualified scheduling scheme, and the power scheduling parameters in each qualified scheduling scheme can be weighted and summed according to the allocated weights. The target scheduling scheme is generated by combining the weighted sums of each scheduling parameter.

[0112] In this embodiment of the invention, for each qualified scheduling scheme, weights can be assigned according to the network loss rate and the magnitude of the conflict risk of the qualified scheme (the scheme with the lowest network loss rate and the smallest conflict risk has a weight of 0.6, the second best scheme has a weight of 0.3, and the rest have a weight of 0.1). The core advantages of each scheme are integrated by combining the weights to generate a target scheduling scheme that fits the current scenario and completes the scenario-specific reconstruction of the scheduling chain.

[0113] As can be seen, the embodiments of this invention ensure the continued effectiveness of the scheme through two major mechanisms: a hard grid constraint adaptation mechanism, which anchors voltage, transmission capacity, and other constraints in real time to prevent the scheme from exceeding the safety threshold; and a signaling-power timing alignment mechanism, which requires data synchronization error ≤ 50ms. Experimental verification shows that when the synchronization error is less than 50ms, the conflict rate soars to 23%, while when it is greater than 50ms, the conflict rate can be controlled within 10%. The 50ms threshold provides a guarantee for the effectiveness of signaling-driven scheduling. Hard constraints: voltage deviation ≤ ±5%, transmission capacity ≤ 100MW, energy storage SOC ∈ [20%, 90%], signaling-power timing alignment error < 50ms. Hard constraints cannot be broken; if they are, the scheme will be directly eliminated. Soft constraints: grid loss rate ≤ 3%, scheduling cost ≤ benchmark cost (average scheduling cost in the past week). Soft constraints can be moderately relaxed (e.g., grid loss rate ≤ 3.5%), but the scheme score will be deducted.

[0114] Based on the target scheduling scheme generated in the preceding steps, in step S206, the power resources in each scenario can be scheduled according to the target scheduling scheme in each scenario within the target area.

[0115] In one embodiment of the present invention, the aforementioned target scheduling scheme can be generated by a MIP model. However, since the core optimization problem of power dispatch is essentially an NP-hard problem, when relying solely on the MIP model for direct solution, as the scale of distributed resources in the virtual power plant expands (e.g., more than 20 photovoltaic and energy storage units), the feasible solution space of the MIP model grows exponentially, increasing the time required for a single solution and making it unable to adapt to second-level load fluctuations, resulting in a loss of real-time performance. Furthermore, if the model is simplified in pursuit of speed (e.g., by reducing constraints or trimming the solution space), the optimality of the scheduling scheme will decrease significantly, the network loss rate will increase, and it will be easy to break through security constraints, making it impossible to balance efficiency and accuracy.

[0116] To address the issues of low efficiency and inability to deploy in real-time using the MIP model, and the inability of ordinary AI models to handle complex power grid constraints, this invention provides a lightweight AI model trained using high-quality labeled samples, based on the scheduling scheme generated by the MIP model. This model fully transfers the optimal scheduling functionality of the MIP model to the new model, resulting in a lightweight scheduling model with fast inference speed and high accuracy. The training process is as follows: Step d1: Construct the training sample set; Step d2: Using a mixed-integer programming model as the teacher model and an initial deep learning model as the student model, train the teacher and student models with a training sample set until the preset distillation loss function converges. Then, use the trained student model as the scheduling inference model to generate a scheduling scheme.

[0117] For ease of understanding, please refer to the embodiments of the present invention. Figure 6 , Figure 6 Examples of labeled samples provided for embodiments of the present invention. Figure 6 The training sample set includes multiple labeled samples, each containing at least input features and label information; the input features include signaling power alignment data and its associations. , The labeling information includes performance labels for scheduling scheme parameters generated by the mixed-integer programming model. Of course, the labeled samples can also include supplementary explanations to clarify the physical or business logic relationship between signaling characteristics and scheduling decisions, such as elucidating specific... The increase in value corresponds to the load growth driver (such as a surge in air conditioning load due to the gathering of people in commercial areas). The changes in values ​​reflect regional differences in electricity consumption behavior (such as the temporal coupling relationship between concentrated start-up and shutdown of residential equipment and signaling activity), as well as the priority response strategy of the MIP model under constrained critical conditions (such as prioritizing the reduction of industrial load rather than regulating energy storage when the voltage deviation approaches 5%). These auxiliary explanations not only enhance the interpretability and traceability of the samples, but also provide students with implicit domain knowledge for model learning.

[0118] In this embodiment of the invention, historical data of the target area over a historical period (e.g., one year) can be collected to construct labeled samples. This data can be obtained from the target virtual power plant dispatch center data platform and the base station data interface of cooperating communication operators. The collected data covers all time periods, including weekdays / holidays, peak / off-peak / valley times, and includes 1ms precision signaling interaction data, synchronous power parameters (PV output, energy storage SOC, regional load, etc.), historical dispatch records of the MIP model, and grid constraint verification results. The screening criteria remove samples with a data missing rate greater than 2%, a timing alignment error greater than or equal to 50ms, or MIP solution timeouts (greater than 1s). From 525,600 original data entries, 60,000 valid samples are ultimately selected, accounting for approximately 11.4%.

[0119] Optionally, in addition to historical data from the target area mentioned above, data from core urban areas of similar cities (supplementary data for generalization capabilities) and synthetic data from extreme scenarios can also be collected. The data collection entity for core urban areas of similar cities can be the dispatch center of a virtual power plant in the same core urban area (user density, load characteristics, and deviations from the target area ≤ 15%), obtaining authorized data through an industry data sharing platform; the data collection content is consistent with the sample dimensions of the target area, focusing on supplementing scenario data with different business district layouts and industrial load proportions; the data screening criteria are to exclude samples with an industrial load proportion exceeding 50% and significant differences in base station coverage. The synthetic data from extreme scenarios (20,000 records, supplementary data for scarce scenarios) is generated using the Monte Carlo simulation method, based on the physical laws of the power system and historical load fluctuation characteristics; the covered scenarios include 10 types of extreme operating conditions, 2,000 records for each type, specifically: a sudden drop in photovoltaic output due to heavy rain (a drop greater than or equal to 50%), a surge in visitor flow at a large exhibition, etc. Extreme high temperature air conditioning load surge (load increase greater than or equal to 30%), sudden equipment failure power supply gap, brief interruption of signaling data, etc.; Generation logic: set random variables such as the fluctuation range of people flow and the photovoltaic output attenuation coefficient, combined with the hard constraint verification of the power grid, to ensure that the generated data conforms to the physical laws and there is no constraint breakthrough situation.

[0120] Optionally, the amount of historical data, core urban area data of similar cities, and synthetic data of extreme scenarios for the target area can be flexibly set according to actual needs. For example, 60,000 historical data, 40,000 core urban area data of similar cities, and 20,000 synthetic data of extreme scenarios for the target area can be collected, respectively, without limitation here.

[0121] from Figure 6As can be seen, the labeled samples in this embodiment of the invention adopt a dual-label labeling mechanism. One label is the scheduling scheme parameters output by the MIP model (including photovoltaic output, energy storage charging and discharging power, industrial load reduction, and allocation ratio), with a labeling accuracy of 0.1MW and 0.01%. The other label is the network loss rate (accuracy 0.01%), scheduling cost (accuracy 0.1 yuan / minute), and constraint satisfaction status corresponding to the scheduling scheme. During implementation, professional power dispatchers and algorithm engineers can jointly label the samples, and 5% of each batch of samples can be randomly selected for manual verification.

[0122] Optionally, embodiments of the present invention can also filter labeled samples, such as removing samples that have timed out of MIP (e.g., greater than 1 second), have constraints that are not satisfied, or have missing data fields. Finally, some valid samples are retained in the original samples, and the training set, validation set, and test set are divided according to a preset ratio (e.g., 7:2:1) to ensure a balanced distribution of the dataset.

[0123] Optionally, embodiments of the present invention can also enhance the labeled samples. For example, data augmentation strategies can be adopted for extreme scenario samples. Enhanced samples can be generated by adding random load fluctuations within a preset fluctuation range (such as ±5%), adjusting the initial value of energy storage SOC, and fine-tuning the α coefficient, thereby improving the model's adaptability to extreme working conditions and avoiding insufficient generalization.

[0124] In this embodiment of the invention, all labeled samples are uniformly converted into JSON format, timestamps and scene labels are added, and the precision of numerical fields is normalized to ensure that the model input format is uniform and reduce training redundancy.

[0125] Next, this embodiment of the invention can be based on a scheduling knowledge transfer model architecture that combines mixed integer programming and lightweight deep learning, including a mixed integer programming model and a scheduling inference model. The scheduling inference model is a lightweight optimized deep neural network model, such as the Llama2-7B language model. In the knowledge distillation process, the mixed integer programming model will serve as the teacher model, and the scheduling inference model will serve as the student model. In order to enable the student model to effectively learn the decision logic of the teacher model, the distillation loss function can be defined in the model training stage as shown in the following formula (27):

[0126] in, The mean square error between the scheduling parameters output by the student model and the corresponding scheduling parameters of the teacher model is used to measure the numerical accuracy of power allocation and proportional control. The absolute deviation of the sum of network loss rate and scheduling cost per unit time of the scheduling scheme generated for the student model, relative to the corresponding value of the scheduling scheme for the teacher model, is used to measure the degree of achievement of the comprehensive optimization objective. The weight coefficients 0.6 and 0.4 are determined by minimizing the weighted sum of hard constraint violation rate and average performance deviation on the validation set, so as to prioritize the executability of scheduling parameters within the physical feasible region.

[0127] During model fine-tuning, the rank of the LoRA adaptation matrix, the optimizer learning rate, and the maximum number of training epochs can all be flexibly set according to the specific needs of the scenario. For example, the rank can be set to 8, and the optimizer learning rate can be set to... The maximum number of training epochs is 50. An early stopping mechanism can also be used during the training phase, meaning that training is terminated when the value of the validation set loss function decreases by no more than 0.001 over three consecutive training epochs.

[0128] During the model lightweight optimization process, the model can be structured and pruned (e.g., removing neuron connections with absolute weights less than 0.1 in the attention mechanism) and INT8 quantization can be applied, resulting in a compressed model parameter size compared to the original model. Simultaneously, inference latency can be stably controlled within a preset range (e.g., 80ms), supporting multiple scheduling scheme inferences per second on a single device, meeting the millisecond-level closed-loop control requirements of virtual power plants.

[0129] Finally, after complete training and lightweight deployment, the student model achieved the following performance on the independent test set: deviation rate from the optimal solution of the mixed integer programming model is less than or equal to 1.2%, network loss rate deviation is less than or equal to 0.1%, and scheduling cost deviation is less than or equal to 2%; the inference speed is increased to 350 times / second, which is more than 280 times faster than the mixed integer programming model (800ms / time in complex scenarios), fully meeting the sub-second real-time scheduling requirements.

[0130] In one embodiment of the present invention, considering that the existing technology lacks any dynamic data support for real-time monitoring and can only rely on subsequent power operation data to report conflicts (such as voltage exceeding limits and line overload), it belongs to the "post-event discovery" mode. Moreover, because no real-time dynamic data such as signaling is introduced, it is impossible to predict load abrupt conflicts (such as load surges caused by sudden large-scale events), and there are no targeted correction methods, resulting in a high conflict rate in the dispatching scheme, which seriously threatens the safe and stable operation of the power grid.

[0131] To address the aforementioned issues, after issuing the target scheduling scheme generated in the pre-execution steps, the following steps can also be performed: Step e1: Obtain signaling power alignment data and power grid operation parameters under the target scheduling scheme according to the preset acquisition frequency; Specifically, the system can synchronously acquire two types of key data according to a preset acquisition frequency: one type is signaling power alignment data, which is data that has been aligned with a 1-millisecond (ms) precision timestamp and includes the real-time population flow coefficient of the region. And signaling load mapping coefficient The data consists of two categories: First, the combined data from the grid. Second, the real-time data collected during the execution of the target scheduling scheme, including actual node voltage, line transmission power, state of charge (SOC) of energy storage, grid loss rate, and actual reduction in industrial load—values ​​reflecting the true physical operating state. During this process, the preset acquisition frequency is once per second. This frequency satisfies the observability requirements for second-level dynamic load fluctuations without excessively consuming system communication and computing resources. The obtained signaling power alignment data corresponds one-to-one with the grid operating parameters and is synchronized in time, forming the basic input for subsequent multi-level conflict determination.

[0132] Step e2: Perform multi-level conflict identification based on signaling power alignment data and grid operating parameters; Step e3: When an early warning of any level of conflict is triggered, the conflict is corrected based on the early warning correction strategy corresponding to any level of conflict, and a corrected target scheduling scheme is generated and issued for execution.

[0133] This invention constructs a priority-based dynamic conflict discrimination mechanism. This identification process is not a simple threshold comparison, but rather performs conflict determination at three levels based on the inherent logical relationship between the two types of data. Specifically: The first level is hard constraint conflict identification, which determines whether the grid operating parameters substantially violate the grid constraints mentioned above, including but not limited to: whether the deviation of the actual node voltage from the rated voltage exceeds 5%, whether the actual transmission power of the line exceeds 100 megawatts (MW), and whether the state of charge (SOC) of the energy storage is below 20% or above 90%, for example: The voltage deviation verification conditions are: ,in, This represents the actual voltage at the node. Rated voltage; The transmission capacity verification condition is: >100MW, of which This represents the actual transmission power of line ij. Energy storage SOC verification conditions: SOC < 20% or SOC > 90%.

[0134] When any grid constraint condition is triggered, a Level 1 conflict warning is activated. At this point, the corresponding correction strategy for that level of conflict can be immediately triggered, i.e., automatically adjusting the energy storage power adjustment amount based on the linear correlation formula between voltage deviation and energy storage power. The formula is as follows:

[0135] in The voltage-energy storage adjustment factor (tested and calibrated to 5MW / %). This is the difference between the actual voltage deviation and the threshold (±5%). If... If so, reduce the energy storage charging power or increase the discharge power; if If the voltage deviation is 5.2% (exceeding the threshold by 0.2%), then reduce the discharge power or increase the charging power; for example, a voltage deviation of 5.2% (exceeding the threshold by 0.2%). If the power is reduced, the energy storage charging power will be automatically reduced by 0.1MW.

[0136] In this embodiment of the invention, if the voltage deviation is still greater than 5% after a single adjustment, the industrial load reduction mechanism is further activated, and the reduction range is executed according to the following formula:

[0137] in The maximum reduction is strictly controlled at 10%.

[0138] The second level is data consistency conflict identification, which involves checking whether the time synchronization and change coordination between signaling power alignment data and grid operating parameters have deteriorated. This includes, but is not limited to, verifying: whether the timestamp deviation between signaling data and power data exceeds 50 milliseconds (ms), and whether the fluctuation range of the population mobility coefficient α or signaling load mapping coefficient β relative to their historical average for the same period exceeds 5%. For example: Timing alignment error verification conditions: ,in This is the time stamp discrepancy between signaling and power data.

[0139] The characteristic fluctuation test conditions include: , ,in , This represents the average characteristic value for the same historical period.

[0140] A Level 2 warning is triggered when any of the above conditions are met. At this time, a conflict correction strategy for this level is immediately activated. For example, when the timing alignment error is ≥50ms, the linear interpolation formula (1) is reused to re-interpolate and complete the missing or misaligned power data points. Simultaneously, for... The coefficients are statistically calibrated, as shown in the formula:

[0141] in, This represents the historical α mean for the same scenario and time period. (where α is the standard deviation of the fluctuation).

[0142] In addition, the system automatically switches to the backup scheduling scheme generated by the MIP teacher model. This scheme is sorted by network loss rate, constraint satisfaction and response delay weighted score and stored in the local cache.

[0143] The third level is scheme deviation conflict identification, which assesses whether there is a significant deviation between the execution result output by the current scheduling inference model and the theoretical optimal solution determined by the mixed integer programming model. Specifically, it uses whether the deviation rate of key scheduling parameters such as photovoltaic power output setpoint, energy storage charging and discharging power, and industrial load reduction power exceeds 1.2% as the criterion. For example, taking power as an example, the deviation rate verification condition is: ,in and These are any scheduling parameters generated by the scheduling inference model and the MIP model under the same input conditions, such as energy storage charging and discharging power, photovoltaic output setpoint, and industrial load reduction power.

[0144] When a Level 3 conflict is triggered, the execution model output weights undergo dynamic decay and parameter fusion correction: First, the original output weights of the scheduling inference model are adjusted based on the deviation rate between the execution result output by the scheduling inference model and the theoretical optimal solution determined by the mixed integer programming model. The correction is made, and the correction formula (31) is as follows:

[0145] in As the initial weights, The deviation rate between the execution result output by the scheduling inference model and the theoretical optimal solution determined by the mixed integer programming model; if the deviation is still greater than 1.2% after correction, the corresponding parameters of the MIP model are introduced as correction terms, and the final output value is calculated according to the formula. Calculation, where Represents any scheduling parameter; Meanwhile, embodiments of the present invention can also record the causes of deviations (feature deviation / insufficient model generalization) and correction parameters for use in the next round of model iteration training.

[0146] After all correction actions are completed, this embodiment of the invention can also generate a closed-loop feedback package containing a comparison of parameters before and after correction, execution time, constraint satisfaction status, and attribution flags. This feedback package is sent to the data quality monitoring interface and the signaling service provider's API endpoint to drive the adaptive adjustment of the front-end data processing logic and the quality traceability of external data sources.

[0147] In the aforementioned multi-level conflict identification process, all corrective actions are initiated upon receiving a conflict signal and completed within a specified time delay. The three levels of conflict are independent of each other and do not substitute for one another. Each type corresponds to a unique and clear triggering condition and a verifiable quantitative indicator, which together constitute a three-dimensional risk perception system.

[0148] The power resource scheduling method provided in this embodiment of the invention has the following verifiable technical effects in actual deployment and operation, as shown in Table 2: In terms of data processing, this embodiment of the invention, based on regional-level de-identified signaling statistics provided by an external compliant signaling service provider, can achieve 1ms precision timestamp benchmark alignment and scenario-specific feature extraction processes. Without adding dedicated acquisition hardware, it can stably control the signaling-power data timing synchronization error within 50ms. Correspondingly, the absolute error of load forecasting, after 72 hours of continuous measured statistics, has a root mean square error (RMSE) of no more than 0.9%, which is 82% lower than the control system (using the same historical load, weather, and electricity price inputs) without signaling data. This improvement in forecasting accuracy directly stems from… The coefficient's ability to dynamically represent population mobility has been enhanced, and High-fidelity modeling of the nonlinear mapping relationship between coefficients and signaling interaction quantity and load power.

[0149] In terms of system expansion, this embodiment of the invention, through a scenario-specific clustering architecture (K=4) and a scenario-specific feature calibration mechanism, enables the system to migrate to different geographical areas (core urban business districts, suburban residential communities, industrial parks, temporary exhibition venues) without requiring manual resetting of the model structure or constraint parameters. Initialization configuration can be completed within 5 minutes simply by loading the corresponding area's base station coverage label mapping table and historical calibration coefficient library. The lightweight student model, after INT8 quantization, has a size of 126MB, supporting deployment on distributed edge gateway devices. It has already implemented a unified scheduling strategy across all nodes in a pilot virtual power plant containing 217 distributed resource units, with end-to-end command latency (from signaling acquisition to terminal action execution) remaining stable at [value missing]. Within the range.

[0150] In terms of scheduling execution, the lightweight scheduling inference model constructed in this embodiment of the invention can achieve a single inference latency of less than or equal to 80ms and a throughput of 350 times / second on edge computing nodes. Compared with the optimal solution of the MIP model under the same input conditions, the average absolute percentage error (MAPE) of the output scheduling scheme parameters (energy storage charging and discharging power, photovoltaic output allocation ratio, and industrial load reduction) is 1.2%, the network loss rate deviation is 0.1 percentage points, and the unit time scheduling cost deviation is 2%. This performance enables the system to complete the entire closed loop from signaling data access, feature extraction, scheme generation to instruction issuance within a 1-second time window, meeting the response requirements of city-level virtual power plants to 1-second load fluctuations.

[0151] In terms of power grid operation, the sub-second-level conflict monitoring and closed-loop correction mechanism provided by this invention reduces the occurrence rate of hard constraint breach events (voltage exceeding limits, transmission capacity exceeding limits, and SOC exceeding limits) to 0.47% during actual operation of the system over 30 consecutive days, covering weekdays / holidays and peak / valley periods, which is lower than the industry standard threshold of 0.5%. The regional weighted average network loss rate has decreased from the baseline of 3.2% before scheduling to 2.1%, a reduction of 34.4%. The unit time scheduling cost is reduced by 18.3% compared to the historical average of the same period without this system. This economic improvement stems from the diversity of candidate schemes generated in the exploration phase of the three-stage scheduling chain, the dual screening of signaling adaptability and constraint compliance in the verification phase, and the dynamic weighted fusion mechanism based on network loss rate and conflict risk in the integration phase.

[0152] Table 2

[0153] Based on the above technical concept, embodiments of the present invention also provide a power resource dispatching system, such as... Figure 7 As shown, Figure 7 The functional block diagram of the power resource scheduling system provided in this embodiment of the invention includes: a data preprocessing and feature extraction module, a scheduling chain construction module, a knowledge distillation optimization module, and a dynamic conflict handling module. These modules can work together to execute the power resource scheduling method provided in this embodiment of the invention to achieve the corresponding technical effects. Specifically: The data preprocessing and feature extraction module is used to perform data preprocessing, time-series alignment, and scenario-specific calculations based on externally input compliant signaling data and power data. This can be used to perform the calculation process of the correction parameters for the population flow coefficient and signaling load mapping coefficient in steps S201 to S202, a1 to a3, b1 to b3, and c1 to c3 in the embodiments of the present invention, providing solid and reliable data support for subsequent scheduling.

[0154] The scheduling chain construction module receives data input from the data preprocessing and feature extraction modules, and constructs a three-stage scheduling decision-making process. The first stage... , Under the premise of hard grid constraints, multiple candidate scheduling schemes that meet the basic grid constraints are generated; in the second stage, the candidate schemes are subjected to triple verification, which not only requires them to meet hard safety constraints such as voltage deviation, transmission capacity, and energy storage SOC, but also requires them to pass the signaling scenario label matching test. The first stage is the load forecast accuracy verification, which selects qualified solutions that truly balance safety and adaptability. The second stage is the integration stage, which weights and merges the retained qualified solutions based on the network loss rate and conflict risk, and embeds the grid hard constraint anchoring and data timing alignment guarantee mechanism. Finally, it outputs a target scheduling scheme with a clear structure and well-defined parameters. That is, this module can be used to execute steps S203 to S206 in the embodiments of the present invention to achieve the corresponding technical effects.

[0155] The knowledge distillation optimization module aims to resolve the fundamental contradiction between the inherent computational complexity and real-time requirements of the MIP model. Essentially, it transfers the globally optimal scheduling knowledge inherent in the MIP model to a lightweight student model, balancing the optimality of the scheduling scheme with inference efficiency to adapt to the real-time scheduling needs of virtual power plants. It relies on the high-quality labeled samples constructed in this invention's embodiments to ensure the accuracy of knowledge transfer and the model's generalization ability. This module can be used to execute steps d1 to d2 in this invention's embodiments to achieve the corresponding technical effects.

[0156] The dynamic conflict post-processing module continuously and synchronously acquires signaling, power, and grid operation data at a frequency of 1 Hz, completing real-time monitoring of key indicators within millisecond delays, including timing alignment errors, voltage deviations, grid loss rates, energy storage SOC status, and... , The stability of core features is ensured. When an anomaly is detected, the module implements a tiered response based on the severity of the conflict, corrects the conflict, and modifies the current scheduling scheme before issuing the new module for execution. This module does not run in isolation but rather incorporates the root cause analysis results of each conflict (e.g., (Increased mutation frequency, large calibration error in certain scenarios, etc.) are fed back in real time to the data preprocessing and feature extraction module, driving it to dynamically optimize feature extraction weights or calibration strategies, thereby forming a complete positive reinforcement closed loop from the data source to scheduling decisions and then to operational feedback. This module can be used to execute steps e1 to e3 in the embodiments of the present invention to achieve the corresponding technical effects.

[0157] It should be noted that the power resource dispatching system provided in the embodiments of the present invention can be specific hardware on the device or software or firmware installed on the device. The system provided in the embodiments of the present invention has the same implementation principle and the same technical effects as the foregoing method embodiments. For the sake of brevity, any parts not mentioned in the system embodiments can be referred to the corresponding content in the foregoing method embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices and units described above can all be referred to the corresponding processes in the above method embodiments, and will not be repeated here.

[0158] Optionally, the above modules can be stored in the form of software or firmware. Figure 8 The memory shown is either stored in or embedded in the operating system (OS) of the electronic device 80, and can be used by... Figure 8 The processor executes the commands. Meanwhile, the data and program code required to execute these modules can be stored in memory.

[0159] Please see Figure 8 , Figure 8 The diagram illustrates the structure of an electronic device according to an embodiment of the present invention, including a memory 801, a processor 802, and a communication interface 803. The memory 801, processor 802, and communication interface 803 are electrically connected to each other directly or indirectly to achieve data transmission or interaction. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0160] Optionally, the bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 8 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0161] In this embodiment of the invention, the processor 802 may be a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in this embodiment of the invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in this embodiment of the invention can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor. The software modules may be located in the memory 801, and the processor 802 reads the program instructions from the memory 801 and, in conjunction with its hardware, completes the steps of the aforementioned methods.

[0162] In this embodiment of the invention, the memory 801 can be a non-volatile memory, such as a hard disk drive (HDD) or a solid-state drive (SSD), or it can be volatile memory, such as RAM. The memory can also be any other medium capable of carrying or storing desired executable program code having an instruction or data structure form and accessible by a computer, but is not limited thereto. The memory in this embodiment of the invention can also be a circuit or any other device capable of implementing a storage function for storing instructions and / or data.

[0163] The memory 801 can be used to store software programs and modules, such as the instructions / modules of the power resource dispatching system provided in this embodiment of the invention. These can be stored in the memory 801 in the form of software or firmware, or embedded in the operating system (OS) of the electronic device 80. The processor 802 executes various functional applications and data processing by executing the software programs and modules stored in the memory 801. The communication interface 803 can be used to communicate with other node devices for signaling or data.

[0164] Understandable. Figure 8 The structure shown is for illustrative purposes only; the electronic device 80 may also include components that are more advanced than those shown. Figure 8 The more or fewer components shown, or having the same Figure 8 The different configurations shown. Figure 8 The components shown can be implemented using hardware, software, or a combination thereof.

[0165] Based on the above embodiments, the present invention also provides a storage medium in which a computer program is stored. When the computer program is executed by a computer, the computer executes a power resource scheduling method provided in the above embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0166] Based on the above embodiments, the present invention also provides a program product, which includes a computer program. The processor can execute the computer program to implement the power resource scheduling method provided in the embodiments of the present invention. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0167] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and there may be other division methods in actual implementation. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interface; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0168] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the objectives of the embodiments of the present invention, depending on actual needs.

[0169] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0170] It should be noted that if the function is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes: USB flash drive, mobile hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, and other media capable of storing program code.

[0171] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A power resource dispatching method, characterized in that, The method includes: Obtain signaling data and power data of the target area in the current time period, and perform time-series alignment on the signaling data and power data to obtain signaling-power aligned data; The signaling power alignment data is clustered to obtain data clusters under multiple scenarios, and the population flow coefficient and signaling load mapping coefficient under the corresponding scenario are determined according to each data cluster; For each scenario, the objective function is to minimize the weighted sum of network loss rate and scheduling cost. The objective function is solved based on the preset hard constraints of the power grid, the population mobility coefficient and the signaling load mapping coefficient to obtain multiple candidate scheduling schemes. For each scenario, the multiple candidate scheduling schemes are jointly verified from multiple dimensions to select qualified scheduling schemes. Weights are allocated based on the scheduling performance indicators corresponding to each qualified scheduling scheme, and the power scheduling parameters in each qualified scheduling scheme are weighted and summed according to the allocated weights. The target scheduling scheme is generated by combining the weighted sums of the scheduling parameters. According to the target scheduling scheme for each scenario within the target area, the power resources in each scenario are scheduled.

2. The power resource dispatching method according to claim 1, characterized in that, Obtain signaling data and power data for the target area in the current time period, and perform time-series alignment on the signaling data and power data to obtain signaling-power aligned data, including: Based on the time precision of the signaling data, align all power data timestamps with the signaling timestamps; For time points where no power data is available at the specified time precision, linear interpolation is used to complete the power data corresponding to the time points. The signaling power alignment data is obtained by combining the power data corresponding to each time point under the specified time precision with the signaling data.

3. The power resource dispatching method according to claim 2, characterized in that, Before obtaining the signaling power alignment data from the power data corresponding to each time point at the time precision and the signaling data, the method further includes: For time points where there is no corresponding power data under the specified time precision, the alignment error is determined based on the signaling data timestamp and the power data timestamp. When the alignment error exceeds a preset threshold, a data anomaly feedback signal is sent to obtain the power grid standard timestamp; Linear interpolation is performed based on the time precision of the signaling data and the power grid standard timestamp to update the power data corresponding to the time point.

4. The power resource dispatching method according to claim 1, characterized in that, Determine the population mobility coefficient for each scenario based on each data cluster, including: The ratio of the total number of signaling interactions in the data cluster to the number of permanent residents covered by the base station in the corresponding scenario of the data cluster is used as the population mobility coefficient. If the population mobility coefficient exceeds the reasonable range of the population mobility coefficient in the scenario corresponding to the data cluster, then the signaling fluctuation coefficient and population mobility coefficient correction parameters in the scenario corresponding to the data cluster are determined; otherwise, no action is taken. The population mobility coefficient correction parameter is fine-tuned based on the signaling fluctuation coefficient, and the population mobility coefficient is corrected by the fine-tuned population mobility coefficient correction parameter.

5. The power resource dispatching method according to claim 4, characterized in that, The correction parameters determine the signaling fluctuation coefficient and correction parameters for the scenario corresponding to the data cluster, including: Obtain historical signaling power alignment data for the scenario corresponding to the data cluster; Multiple valid data groups were selected from the historical signaling power alignment data; Calculate the historical population mobility coefficient for each of the valid data sets, and calculate the fitting deviation using the historical population mobility coefficient corresponding to each of the valid data sets and the actual population mobility coefficient; The average of all fitting deviations is used as the correction parameter.

6. The power resource dispatching method according to claim 1, characterized in that, The signaling load mapping coefficients for each data cluster are determined according to the corresponding scenario, including: Determine the minimum and maximum number of signaling interactions from the data cluster, and predict the maximum and minimum load based on the maximum and minimum values; The signaling load mapping coefficient is calculated from the minimum and maximum values ​​of the number of signaling interactions and the maximum and minimum values ​​of the load. If the signaling load mapping coefficient exceeds the reasonable range of the signaling load mapping coefficient in the scenario corresponding to the data cluster, then the load fluctuation coefficient and signaling load mapping coefficient correction parameters in the scenario corresponding to the data cluster are determined; otherwise, no action is taken. The signaling load mapping coefficient correction parameter is fine-tuned based on the load fluctuation coefficient, and the signaling load mapping coefficient is corrected by the fine-tuned signaling load mapping coefficient correction parameter.

7. The power resource dispatching method according to claim 1, characterized in that, For each scenario, a multi-dimensional joint verification is performed on the multiple candidate scheduling schemes to select qualified scheduling schemes, including: The candidate scheduling schemes are subjected to signaling adaptability verification, power grid constraint verification, and scenario adaptability verification, respectively. If the candidate scheduling scheme passes each verification, it is determined to be a qualified scheduling scheme.

8. The power resource dispatching method according to claim 1, characterized in that, After scheduling the power resources in each scenario within the target area according to the target scheduling scheme for each scenario, the method further includes: Obtain signaling power alignment data and power grid operation parameters under the target scheduling scheme according to the preset acquisition frequency; Perform multi-level conflict identification based on the signaling power alignment data and the power grid operating parameters; When an alert for any level of conflict is triggered, the conflict is corrected based on the alert correction strategy corresponding to that level of conflict, and a corrected target scheduling scheme is generated and issued for execution.

9. The power resource dispatching method according to claim 1, characterized in that, The method further includes: Construct a training sample set; wherein the training sample set includes multiple labeled samples, each labeled sample containing at least input features and label information; the input features include signaling power alignment data and its associated population flow coefficient and signaling load mapping coefficient; the label information includes scheduling scheme parameters and scheme performance labels generated by the mixed integer programming model; Using the mixed integer programming model as the teacher model and the initial deep learning model as the student model, the teacher model and the student model are trained with the training sample set until the preset distillation loss function converges. The trained student model is then used as the scheduling inference model to generate a scheduling scheme.

10. An electronic device, characterized in that, It includes a processor and a memory, the memory storing machine-executable instructions that can be executed by the processor, the processor executing the machine-executable instructions to implement the power resource scheduling method according to any one of claims 1 to 9.

Images