An efficient method for dynamic allocation of cloud computing resources

Abstract

The present application discloses an efficient method for dynamic allocation of cloud computing resources, which belongs to the field of cloud computing technology. The method includes S1, real-time collection of multiple load source data under the influence of electric vehicle charging behavior, feature analysis and extraction of load source data, and prediction of short-term fluctuation trend of regional load in combination with deep learning model; S2, by simulating the local load pressure of electric vehicle charging peak, using multi-objective optimization method to evaluate the potential of demand-side response to load relief, and dynamically adjust the user behavior incentive mechanism; S3, setting multi-dimensional priority evaluation indicators, and multi-level classification of scheduling tasks through fuzzy logic model; S4, introducing reinforcement learning, dynamically adjusting the computing node allocation strategy according to the scheduling task priority and regional load forecast data, to achieve a balance between local efficiency and global performance; S5, deploying distributed computing nodes in the cloud computing platform, introducing regional collaborative optimization mechanism, and optimizing the allocation and load balancing of computing resources.

Description

Technical Field

[0001] This application relates to the field of cloud computing technology, and more specifically, to an efficient method for dynamic allocation of cloud computing resources. Background Technology

[0002] In smart grids, the randomness and regional concentration of electric vehicle (EV) charging behavior significantly increase the uncertainty of local loads, posing challenges to the coordination of overall grid dispatch and demand-side response. Demand-side response incentivizes users to adjust their electricity consumption to balance loads; however, its effectiveness may be limited or even conflict with overall grid dispatch objectives during peak EV charging periods. For example, excessive incentivization may lead to overly concentrated load transfer times, further exacerbating local load pressure. Furthermore, overall grid dispatch needs to seek global optimization among generation costs, transmission losses, and dynamic changes in electricity demand; balancing real-time performance and load balancing is crucial. Therefore, a collaborative optimization method based on dynamic allocation of cloud computing resources is urgently needed. This method should dynamically adjust computing resources and optimization strategies according to the predicted fluctuations in EV charging load, the actual effects of demand-side response, and the priorities of overall grid dispatch, in order to achieve load balancing, reduce transmission and distribution losses, and ensure system operational stability. Summary of the Invention

[0003] To overcome a series of shortcomings in existing technologies, the purpose of this application is to provide an efficient method for dynamic allocation of cloud computing resources, comprising the following steps:

[0004] Step S1: Collect data from multiple load sources affected by electric vehicle charging behavior in real time, including smart meters, charging piles, distributed energy, weather monitoring stations and power grid operation status data. Perform feature analysis and extraction on electric vehicle load, basic user electricity consumption behavior and distributed energy output in the load source data, and combine deep learning models to predict the short-term fluctuation trend of regional load.

[0005] Step S2: Based on the regional load forecast results, by simulating the local load pressure during peak electric vehicle charging, a multi-objective optimization method is used to evaluate the potential of demand-side response to alleviate the load, and by dynamically adjusting the user behavior incentive mechanism, the load distribution is ensured to be coordinated with the overall operation objectives of the power grid.

[0006] Step S3: Based on the grid dispatching requirements, set multi-dimensional priority evaluation indicators, including the regional impact of electric vehicle charging, demand-side response effect, grid generation cost and transmission loss, and classify the dispatching tasks into multiple levels through a fuzzy logic model.

[0007] Step S4: Construct a dynamic allocation model for cloud computing resources based on reinforcement learning. Adjust the allocation strategy of computing nodes dynamically according to the priority of scheduling tasks and regional load prediction data to achieve a balance between local efficiency and global performance.

[0008] Step S5: Deploy distributed computing nodes in the cloud computing platform, decompose and distribute the regional load allocation and scheduling tasks, introduce a regional collaborative optimization mechanism, coordinate the scheduling of computing resources and power load in neighboring regions, and optimize the allocation and load balancing of computing resources among different nodes.

[0009] Step S6: Execute the optimized resource allocation scheme in real time, collect the execution status of each computing node, load adjustment effect and power grid stability indicators, and evaluate the key performance indicators of the scheduling effect.

[0010] Step S7: Establish a real-time feedback and online learning mechanism, using the system's performance indicators as feedback to feed back to the cloud computing resource dynamic allocation model and the deep learning model for load forecasting, and continuously optimize the resource allocation strategy through online learning technology.

[0011] Furthermore, a multi-load source data input interface is constructed to integrate the collected multi-load source data. Different data analysis methods are used to extract the characteristics of different load source data. Specifically, for electric vehicle load analysis, the impact of charging time period, charging power change, and geographical location on charging demand is analyzed. For basic user electricity consumption behavior, user groups are classified through cluster analysis to extract typical electricity consumption behavior patterns. For distributed energy output, the impact of weather changes on energy output is obtained through regression analysis or time series analysis.

[0012] Furthermore, LSTM or Transformer in deep learning is used to predict the short-term fluctuation trend of regional load. The time-series characteristics and features of each load source data are extracted as input to the deep learning model. Data from different load sources are integrated into a unified deep learning framework to obtain the final short-term fluctuation trend curve of regional load and output the future short-term load demand fluctuation.

[0013] Furthermore, in step S2, a user behavior response model is constructed based on the grid load status, time-of-use electricity price, incentive mechanism, and user charging demand to adjust the charging time and power of electric vehicles. The model optimization objective is to minimize charging costs, which include economic costs and grid pressure. The user behavior response model is as follows:

[0014]

[0015] Among them, P priceLnet(t) is the electricity price at time t, and Lnet(t) is the load pressure on the power grid at time t, reflecting the power grid load situation. charge (i,t) represents the charging power of user i during time period t, where T charge (i) represents the charging period for user i, R user (t) is the reward value provided by the demand response incentive;

[0016] The charging time period is selected based on changes in grid load and electricity price. The charging time period selection function is expressed as follows:

[0017] T charge (i)=f(L net (t), P price (t), C demand (i))

[0018] Among them, C demand (t) represents the user's charging demand, indicating the amount of electricity the user needs to charge within a certain period of time. The function f is established through historical data and user behavior analysis.

[0019] The charging power is adjusted based on the grid load pressure and demand response signals. The charging power adjustment function is expressed as follows:

[0020]

[0021] in, γ is the maximum load capacity of the power grid during time period t, and γ is the sensitivity coefficient of users to charging power adjustment;

[0022] By dynamically adjusting the incentive signal to encourage users to charge outside of peak load periods, the change in the incentive mechanism is represented as follows:

[0023]

[0024] Among them, L low and L high R represents the low load and high load thresholds, respectively. max and R min These are the maximum and minimum rewards, respectively.

[0025] Furthermore, based on grid load forecasts and user charging behavior, the local load pressure in each region is assessed, expressed by the formula:

[0026]

[0027] Among them, L region (t) represents the total charging load within the time region t, reflecting the load pressure in that region, P charge (i,t) is calculated through dynamic adjustment in the user behavior response model;

[0028] By comparing the changes in local load pressure before and after demand response measures, the potential of demand-side response to alleviate local load pressure is quantified. The change in load pressure is expressed as:

[0029]

[0030] in, This is the sum of regional loads without demand response measures. The total regional load after the introduction of demand response measures, ΔL region (t) represents the change in load pressure;

[0031] A multi-objective optimization method is adopted to simultaneously consider the grid load balance and user charging demand, minimizing grid load fluctuations. The optimization objective function is:

[0032]

[0033] Where λ1 and λ2 are the weight coefficients of the optimization objective;

[0034] By introducing load balancing constraints and charging power limits, we ensure that the grid load does not exceed the grid's maximum carrying capacity during the optimization process.

[0035]

[0036] Among them, L min (t) and L max (t) represents the minimum and maximum allowable values ​​of the power grid load. and Minimum and maximum limits for charging power.

[0037] Furthermore, in step S3, the scheduling tasks are divided into three priorities: high priority, medium priority, and low priority. A fuzzy logic model is used to transform the regional impact of electric vehicle charging, demand response, grid generation costs, and transmission losses into processable fuzzy sets. Then, rule-based reasoning is used to dynamically allocate priorities to the scheduling tasks. Specifically, the regional impact degree I is used for each priority level. EV Load reduction ratio R DS Multi-source power generation cost C gen and line transmission loss L loss The influencing factors are characterized as follows:

[0038]

[0039] Among them, I EV (r,t) represents the impact of electric vehicle charging on region r within time period t, L net (r,t) represents the grid load in region r, RDS (r,t) represents the percentage reduction in load in region r during time period t after demand-side response, C gen (t) represents the weighted average of the generation costs of each type of power source used within time period t, where P is the overall generation cost. gen,k (t) represents the power generation of the k-th power generation unit during time period t, C cost,k L is the unit power generation cost of the k-th power generation unit. loss (t) represents the line transmission loss during time period t, I l (t)

[0040] R is the current in line l during time period t. l It is the resistance of line l;

[0041] The fuzzy inference algorithm is used to comprehensively evaluate the input fuzzy values ​​to determine the priority of each task, and the final priority value is obtained through a defuzzification process.

[0042] P task (t)=defuzzify(fuzzy({I EV R DS C gen L loss}))

[0043] Among them, P task (t) represents the final priority of the task, and defuzzify represents the defuzzification process.

[0044] Furthermore, in step S4, the state space of the cloud computing resource dynamic allocation model structure based on reinforcement learning contains input information that reflects the current power grid resource configuration and load status, including regional load forecast data, scheduling task priority, current resource configuration status and resource utilization efficiency. The action space includes the adjustment of computing resource allocation, resource adjustment strategy and resource scheduling method. The reward function comprehensively considers regional load, scheduling task priority, resource utilization efficiency and scheduling cost.

[0045] Furthermore, a real-time adaptive mechanism is introduced to dynamically adjust the allocation strategy of computing nodes based on load forecasting errors and the current resource allocation status. The dynamic adjustment mechanism is as follows:

[0046] a′ t =f(ΔL′) region (t), at)=α·a t +β·adjustment factor

[0047] Among them, a tLet ' be the adjusted resource allocation decision, f(·) be the function used to dynamically adjust the resource allocation strategy, α be the weight coefficient of the current resource allocation (α∈[0,1]), β be the adjustment coefficient (β∈[0,1]), and adjustment factor be composed of multiple factors, reflecting the impact of load changes, priorities, and resource utilization efficiency, i.e., adjustment factor(ΔL') region (t),P task (t),R util (t) represents the weighted sum of multiple factors, ΔL' region (t) represents the actual deviation of the grid load, R util (t) represents the resource utilization efficiency of the region.

[0048] adjustment factor=λ1·|ΔL′ region (t)|+λ2·|ΔL′ region (t)|γ+w1·P task (t)+w2·(1-R util (t))

[0049] Furthermore, computing resources and scheduling tasks are allocated to each region based on load ratio, and the task complexity of each region is evaluated based on its load fluctuations and task types. Based on the task complexity and load requirements of each region, tasks are decomposed and resources are allocated. Neighborhood node collaboration mechanism and inter-region load sharing mechanism are introduced to ensure that computing resources in different regions are shared and work collaboratively, load fluctuations in adjacent regions are transferred to avoid overload in a certain region, and task allocation strategies are adjusted in real time to ensure load balance.

[0050] Compared with the prior art, this application has the following beneficial effects.

[0051] This efficient cloud computing resource dynamic allocation method, through the coordinated operation of multiple steps, aims to optimize grid load scheduling and computing resource allocation during peak electric vehicle charging periods, thereby improving grid stability and response efficiency. The core of the method revolves around real-time data acquisition, load forecasting, demand-side response, priority scheduling, dynamic resource allocation, and collaborative optimization, ensuring optimal configuration and efficient operation of network resources under dynamic load changes and multi-task scheduling. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating the efficient dynamic allocation method for cloud computing resources disclosed in the embodiments of this application. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be described in more detail below with reference to the accompanying drawings. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are some embodiments of this invention, but not all embodiments.

[0054] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] The embodiments and directional terms described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0056] like Figure 1 As shown, an efficient method for dynamic allocation of cloud computing resources includes the following steps:

[0057] Step S1: Collect data from multiple load sources affected by electric vehicle charging behavior in real time, including smart meters, charging piles, distributed energy, weather monitoring stations and power grid operation status data. Perform feature analysis and extraction on electric vehicle load, basic user electricity consumption behavior and distributed energy output in the load source data, and combine deep learning models to predict the short-term fluctuation trend of regional load.

[0058] Step S2: Based on the regional load forecast results, by simulating the local load pressure during peak electric vehicle charging, a multi-objective optimization method is used to evaluate the potential of demand-side response to alleviate the load, and by dynamically adjusting the user behavior incentive mechanism, the load distribution is ensured to be coordinated with the overall operation objectives of the power grid.

[0059] Step S3: Based on the grid dispatching requirements, set multi-dimensional priority evaluation indicators, including the regional impact of electric vehicle charging, demand-side response effect, grid generation cost and transmission loss, and classify the dispatching tasks into multiple levels through a fuzzy logic model.

[0060] Step S4: Construct a dynamic allocation model for cloud computing resources based on reinforcement learning. Adjust the allocation strategy of computing nodes dynamically according to the priority of scheduling tasks and regional load prediction data to achieve a balance between local efficiency and global performance.

[0061] Step S5: Deploy distributed computing nodes in the cloud computing platform, decompose and distribute the regional load allocation and scheduling tasks, introduce a regional collaborative optimization mechanism, coordinate the scheduling of computing resources and power load in neighboring regions, and optimize the allocation and load balancing of computing resources among different nodes.

[0062] As an optional solution to this application, the efficient dynamic allocation method for cloud computing resources also includes the following steps:

[0063] Step S6: Execute the optimized resource allocation scheme in real time, collect the execution status of each computing node, load adjustment effect and power grid stability indicators, and evaluate the key performance indicators of the scheduling effect.

[0064] Step S7: Establish a real-time feedback and online learning mechanism, using the system's performance indicators as feedback to feed back to the cloud computing resource dynamic allocation model and the deep learning model for load forecasting, and continuously optimize the resource allocation strategy through online learning technology.

[0065] In summary, this efficient dynamic allocation method for cloud computing resources effectively addresses grid load pressure during peak electric vehicle charging periods through intelligent data collection, deep learning prediction, optimization algorithms, and real-time feedback mechanisms. First, by collecting multi-source data and combining it with deep learning models to analyze electric vehicle load, user electricity consumption behavior, and distributed energy output, regional load fluctuation trends are predicted. Based on this, a multi-objective optimization method is used to assess the potential of demand-side response, and a dynamic adjustment of user behavior incentive mechanisms is employed to coordinate load distribution with grid operation objectives. Next, considering grid scheduling requirements, a fuzzy logic model is used to prioritize tasks, and reinforcement learning is used to optimize the dynamic allocation strategy for cloud computing resources, ensuring a balance between local efficiency and global performance. Through distributed computing nodes and regional collaborative optimization mechanisms, resources and loads achieve collaborative scheduling and load balancing, improving resource allocation efficiency. Finally, through real-time feedback and online learning mechanisms, the system continuously optimizes resource allocation strategies, ensuring grid stability and improving scheduling flexibility and accuracy. This method not only improves grid operation efficiency during peak electric vehicle charging periods but also provides sustainable technical support for smart grid optimization.

[0066] As an optional solution of this application, a multi-load source data input interface is constructed to integrate the collected multi-load source data, preprocess and clean the data, and use different data analysis methods to extract the characteristics of different load source data. The collection sources of multi-source data and the data characteristic analysis methods are as follows: (1) Smart meter data: Collect user electricity consumption data in each area, including power, peak and valley periods, power factor, etc. The data includes historical electricity consumption, load fluctuations, etc. Analyze the user's electricity consumption pattern, such as the electricity consumption habits of household users in different time periods (difference between daytime and nighttime electricity consumption), classify the user group through cluster analysis (such as K-means), and extract typical electricity consumption behavior patterns, such as peak and valley periods: changes in user electricity consumption during peak and valley periods; user behavior: extract the electricity consumption habit characteristics of different user groups based on historical electricity consumption data.

[0067] (2) Electric vehicle charging pile data: charging power, charging duration, charging location, charging frequency, etc. for each charging pile. Consider the time-varying and location-varying nature of charging behavior, and factors affecting charging demand (such as vehicle arrival time, charging duration, and charging power). Analyze the impact of charging time periods (such as peak and off-peak periods), changes in charging power, and geographical location on charging demand. Periodic characteristics can be extracted by analyzing the frequency, time period distribution, and peak and valley characteristics of historical charging data to obtain daily, weekly, and monthly periodic variation characteristics, such as time period characteristics: charging peak and off-peak periods, considering the differences between weekdays and weekends; charging frequency and charging power: statistically analyze the changes in charging power in different time periods and extract power fluctuation characteristics; geographical distribution: extract location-dependent characteristics based on the differences in charging demand in different regions.

[0068] (3) Distributed energy data: Real-time power generation data of renewable energy sources such as solar and wind power, including power output, weather conditions, and time characteristics. Analyzing the impact of weather conditions (such as sunshine, wind speed, and temperature) on the power generation of renewable energy sources such as solar and wind power can be achieved through regression analysis or time series analysis to obtain the influence of weather changes on energy output. For example, the relationship between weather and power generation: extracting the relationship between weather data (such as wind speed and temperature) and distributed energy output, especially seasonality and time dependence; power generation fluctuations: analyzing the power generation fluctuations at different time scales (hours, days, months).

[0069] (4) Weather data: External meteorological data (temperature, humidity, air pressure, wind speed, sunshine, etc.), which will directly affect load changes, especially in terms of air conditioning load and renewable energy power generation.

[0070] As an optional approach in this application, LSTM or Transformer in deep learning is used to predict the short-term fluctuation trend of regional load. The time-series features and characteristic extraction data of each load source are used as input to the deep learning model. For example, the time-series features in the input data include historical load data, charging behavior data, weather data, and user electricity consumption behavior data. The characteristic extraction data can select time-period features of electric vehicle charging demand, weather dependence of distributed energy resources, etc. Different load source data are fused into a unified deep learning framework to obtain the final short-term fluctuation trend curve of regional load. Taking the LSTM model as an example, its input layer inputs the time-series features and characteristic extraction data of the corresponding load source data. The LSTM layer uses multiple LSTM units to capture the long-term and short-term dependencies in the time series. The fully connected layer maps the time-series features output by the LSTM to obtain the final load forecast. The output layer outputs the predicted grid load demand for the next n hours. The mean squared error (MSE) is used as the loss function to minimize the error between the predicted load and the actual load, and the Adam optimizer is used for model training.

[0071] By using a weighted fusion method, the prediction results of electric vehicle charging load, basic user electricity consumption behavior, and distributed energy output are weighted and integrated to obtain the final regional load prediction result, namely the load curve after multi-source data fusion, which outputs the short-term load demand fluctuations in the future.

[0072] As an optional solution in this application, in step S2, a user behavior response model is constructed based on the grid load status, time-of-use electricity price, incentive mechanism, and user charging demand to adjust the charging time and power of electric vehicles. The optimization objective of the model is to minimize the charging cost, which includes economic cost and grid pressure. The user behavior response model is as follows:

[0073]

[0074] Among them, P price Lnet(t) is the electricity price at time t, and Lnet(t) is the load pressure on the power grid at time t, reflecting the power grid load situation. charge (i,t) represents the charging power of user i during time period t, where T charge (i) represents the charging period for user i, R user (t) is the reward value provided by the demand response incentive.

[0075] The charging period is selected based on changes in grid load and electricity price. If the grid load Lnet(t) is low and the electricity price P... price When (t) is low, users are more likely to charge during that time period. The charging time selection function is expressed as:

[0076] T charge (i)=f(L net (t), P price (t), C demand (i))

[0077] Among them, C demand (t) represents the user's charging demand, indicating the amount of electricity the user needs to charge within a certain period. The function f is established through historical data and user behavior analysis, taking into account the user's sensitivity to load, price, and demand.

[0078] After selecting a charging period, users need to adjust the charging power based on the grid load and demand response signals. The charging power adjustment function is expressed as follows:

[0079]

[0080] in, γ is the maximum load capacity of the power grid during time period t, and γ is the sensitivity coefficient of users to adjustments in charging power.

[0081] Incentive mechanisms and demand response signals play a crucial role in user behavior. By monitoring the grid load status in real time and dynamically adjusting incentive signals, users are encouraged to charge outside of peak load periods. Changes in the incentive mechanism are represented as follows:

[0082]

[0083] Among them, L low and L high R represents the low load and high load thresholds, respectively. max and R min These represent the maximum and minimum rewards, respectively. By dynamically adjusting the incentive signal, the user's charging behavior can adaptively respond to the grid's load status, thereby achieving grid load balance and optimization.

[0084] By adjusting the user behavior response model described above, users' charging time and power can be flexibly adjusted according to the grid load status and reward mechanism. This dynamic decision-making based on demand response and game theory can not only maximize the balance of grid load, but also dynamically incentivize users to adjust their charging behavior according to changes in grid load, thus achieving more efficient grid resource management.

[0085] As an optional approach in this application, based on the aforementioned grid load forecast and user charging behavior, the local load pressure of each region is assessed. The process of assessing and optimizing the local load pressure is closely integrated with adjustments to user behavior to evaluate the potential of demand-side response to alleviate load and optimize grid load distribution. The local load pressure of each region is expressed by the formula:

[0086]

[0087] Among them, L region (t) represents the total charging load within the time region t, reflecting the load pressure in that region, P charge (i,t) is calculated through dynamic adjustment in the user behavior response model.

[0088] By comparing the changes in local load pressure before and after demand response measures, the potential of demand-side response to alleviate local load pressure is quantified. The change in load pressure is expressed as:

[0089]

[0090] in, This is the sum of regional loads without demand response measures. The total regional load after the introduction of demand response measures, ΔL region (t) represents the change in load pressure.

[0091] For example, without a demand response strategy, the distribution of electric vehicle charging load would be concentrated during peak hours, resulting in excessive pressure on local power grids. However, with the introduction of demand response, users adjust their charging time and power, and the charging power of electric vehicles is evenly distributed throughout the entire load cycle. This will cause the load to be distributed in different time periods and regions, effectively reducing the charging load during peak hours and thus alleviating the load pressure on the power grid.

[0092] In order to achieve load balancing, user charging behavior needs to be adjusted through optimization algorithms to make the grid load more even across different time periods, given the interaction between user charging behavior and grid load demand. A multi-objective optimization method is adopted, simultaneously considering grid load balance and user charging demand, to minimize grid load fluctuations. The optimization objective function is:

[0093]

[0094] Where λ1 and λ2 are the weight coefficients of the optimization objective, and the first term The second item aims to minimize load pressure fluctuations and make the load more even. The aim is to balance each user's charging needs with the charging power adjustment, avoiding an excessive gap between user needs and actual charging power.

[0095] Multi-objective optimization constraints are set, including load balancing constraints and charging power limits, to ensure that the grid load does not exceed the grid's maximum carrying capacity during the optimization process.

[0096]

[0097] Among them, L min (t) and L max (t) represents the minimum and maximum allowable values ​​of the power grid load. and Minimum and maximum limits for charging power.

[0098] Local load pressure assessment and optimization is not a static process, but a dynamic one. In practical applications, grid load and user charging behavior change with time and external conditions. Therefore, it is necessary to monitor grid load and user charging behavior in real time through smart meters, charging pile data acquisition systems, etc., to obtain the current charging power distribution and grid load status. Based on the real-time monitoring results, demand response strategies and charging incentive mechanisms are adjusted to further optimize user charging time and power, thereby continuously alleviating local load pressure. The mitigation effect of demand-side response is evaluated by comparing the changes in local load before and after implementing the demand response strategy. If load peaks are effectively dispersed and load fluctuations are smoothed, it indicates that the potential of demand response to alleviate load pressure has been fully realized. If the demand response strategy fails to effectively alleviate load pressure, the system should readjust user charging behavior through optimization algorithms, such as further increasing reward incentives or modifying the charging time and power adjustment strategies.

[0099] Through the above design, the local load pressure assessment and optimization process can be effectively integrated with the user behavior response model, ensuring that electric vehicle charging behavior is effectively adjusted under the demand response mechanism, realizing the deep integration of electric vehicle charging behavior and grid load management, thereby achieving dynamic optimization and balance of grid load.

[0100] As an optional solution in this application, in step S3, the scheduling tasks are divided into three priorities: high priority, medium priority, and low priority. A fuzzy logic model is used to transform the regional impact of electric vehicle charging, demand response, grid generation costs, and transmission losses into a processable fuzzy set. Then, rule-based reasoning is used to dynamically allocate priorities for the scheduling tasks. Specifically, the regional impact degree I is used for each priority level. EV Load reduction ratio R DS Multi-source power generation cost C gen and line transmission loss L loss Characterizing influencing factors. Regarding the regional impact of electric vehicle charging, the impact of charging demand on the power grid in different regions is considered. For example, areas with high load concentration during peak charging periods exert greater pressure on the power grid and require priority treatment. The formula for calculating the impact of such areas is:

[0101]

[0102] Among them, I EV (r,t) represents the impact of electric vehicle charging on region r within time period t, L net (r,t) represents the grid load in region r.

[0103] Regarding the effectiveness of demand-side response, the mitigation effect of demand-side response on load peaks is evaluated using a user behavior response model. This effect can be represented by the load reduction ratio after demand-side response.

[0104]

[0105] Among them, R DS (r,t) represents the proportion of load reduction in region r during time period t after demand-side response.

[0106] For grid generation costs, considering the generation costs of different power sources (such as traditional power generation and renewable energy), the overall generation cost is obtained by weighting the generation costs of each power source:

[0107]

[0108] Among them, C gen (t) represents the weighted average of the generation costs of each type of power source used within time period t, where P is the overall generation cost. gen,k (t) represents the power generation of the k-th power generation unit during time period t, C cost,k It is the unit power generation cost of the kth power generation unit.

[0109] For transmission losses, based on the power grid topology, the transmission losses on different lines are calculated. The losses can be estimated by multiplying the line current by the impedance.

[0110]

[0111] Among them, L loss (t) represents the line transmission loss during time period t, I l (t) is the current in line l during time interval t, R l It is the resistance of line l.

[0112] The aforementioned influencing factors are transformed into tractable fuzzy sets. A set of rules is designed to describe the influence patterns of these factors on the priority of scheduling tasks. Fuzzy inference algorithms (such as Mamdani inference or Sugeno inference) are used to comprehensively evaluate the input fuzzy values ​​to derive the priority of each task. Finally, the priority value is obtained through a defuzzification process.

[0113] P task (t)=defuzzify(fuzzy({I EV R DS C gen L loss}))

[0114] Among them, P task(t) represents the final priority of the task, and defuzzify represents the defuzzification process.

[0115] For example, scheduling tasks can be classified based on the output of fuzzy inference and the specific nature of the task, such as whether it is a critical load area.

[0116] By designing priority decisions for scheduling tasks, factors such as the regional impact of electric vehicle charging, demand-side response effects, grid generation costs, and transmission losses are comprehensively considered. With the help of fuzzy logic models, subsequent task classification and rational allocation of computing resources can be effectively carried out, thereby achieving optimized scheduling of grid load.

[0117] As an optional solution in this application, in step S4, the state space of the cloud computing resource dynamic allocation model structure based on reinforcement learning contains input information reflecting the current power grid resource configuration and load status, including regional load forecast data, scheduling task priorities, current resource configuration status, and resource utilization efficiency. The state is represented as a vector:

[0118] s t =(L region (t), P task (t), R util (t), Q resources (t))

[0119] Among them, R util (t) represents the current resource utilization efficiency, Q resources (t) represents the current remaining amount of computing resources.

[0120] The action space includes adjustments to computational resource allocation, resource adjustment strategies, and resource scheduling methods. For example, computational resource allocation includes determining the number of computing nodes allocated to each region; resource adjustment strategies include prioritizing resource allocation based on task priorities, potentially involving weighted allocation or resource reduction for certain tasks; and resource scheduling methods include selecting a load balancing algorithm or employing a dynamic resource reallocation strategy. Therefore, the action space can be represented as:

[0121] a t =(X allocate (t), Y adjust (t))

[0122] Among them, X allocate (t) is the resource allocation for each region, Y adjust (t) represents the parameters of the resource adjustment strategy, such as the adjustment ratio and delayed allocation.

[0123] The reward function comprehensively considers regional load, scheduling task priority, resource utilization efficiency, and scheduling cost, specifically as follows:

[0124]

[0125] Among them, regional load, dispatch task priority, resource utilization efficiency, and dispatch cost are characterized by load balance R1, task priority satisfaction R2, resource waste R3, and transmission loss R4, respectively:

[0126]

[0127] in, The predicted load reflects the deviation between the actual load and the predicted load.

[0128] The Deep Q-Learning (DQN) model is used to learn the optimal resource allocation strategy through environmental feedback, such as grid load, computing demand and resource consumption.

[0129] As an optional solution in this application, to enable the model to cope with dynamic changes in grid load and demand, a real-time adaptive mechanism is introduced. Based on load forecasting errors and the current resource allocation status, the allocation strategy of computing nodes is dynamically adjusted. The dynamic adjustment mechanism is as follows:

[0130] a′ t =f(ΔL′) region (t), at)=α·a t +β·adjustment factor

[0131] Among them, a t ' represents the adjusted resource allocation decision, f(·) is a function used to dynamically adjust the resource allocation strategy, α is the weight coefficient of the current resource allocation, α∈[0,1], β is the adjustment coefficient, β∈[0,1], and the adjustment factor is composed of multiple factors, reflecting the impact of load changes, priorities, and resource utilization efficiency. The adjustment factor(ΔL') is a weighted adjustment factor. region (t),P task (t),R util Let (t) be represented, and ΔL' be... region (t) represents the actual deviation of the grid load, R util (t) represents the resource utilization efficiency of the region, that is,

[0132] adjustment factor=λ1·|ΔL′ region (t)|+λ2·|ΔL′ region (t)|γ+w1·P task (t)+w2·(1-R util (t))

[0133] It should be noted that the feedback control component for power grid load changes and resource allocation is the core of dynamic adjustment, ΔL' region (t) represents the difference between the actual load and the predicted load. The magnitude and direction of this deviation directly affect subsequent resource adjustments. When the load difference ΔL' region When (t) is large, more aggressive adjustments to resource allocation are needed, which may require increasing computing resources or changing resource allocation strategies. The adjustment factor will be larger, resulting in greater fluctuations in resource allocation. At this point,

[0134] adjustment factor(ΔL′ region (t))=λ1·|ΔL′ region (t)|

[0135] λ1 is a constant that determines the sensitivity of resource adjustment when load fluctuates.

[0136] When the load difference ΔL' region When (t) is small and its fluctuation is small, there is no need to adjust resources significantly. The adjustment factor becomes smaller, reducing the frequency and magnitude of adjustments and avoiding resource waste caused by over-adjustment. At this time,

[0137] adjustment factor(ΔL′ region (t))=λ2·|ΔL′ region (t)|γ,γ∈(0,1)

[0138] λ2 is a relatively small adjustment coefficient, and γ controls the reduction of the adjustment factor with load fluctuation.

[0139] During dynamic adjustment, in addition to real-time load differences, task priority and regional resource utilization efficiency are also important factors determining resource adjustments. Based on the task priority and resource utilization efficiency of different regions, the following weighted adjustment factors are designed:

[0140] adjustment factor (P) task (t))=w1·P task (t)

[0141] adjustment facrtor(R util (t))=w2·(1-R util (t))

[0142] More resources should be allocated to high-priority task areas, such as high-load areas for electric vehicle charging. If the resource utilization efficiency of a certain area is low, that is, there is a large waste of resources, resources should be adjusted in these areas first to improve resource utilization efficiency.

[0143] By designing a multi-factor weighted dynamic adjustment mechanism, the resource allocation strategy can be automatically adjusted based on load changes, task priorities, and resource utilization efficiency. This allows for real-time adaptation to load fluctuations, effective optimization of resource allocation strategies, and avoidance of the influence of single factors, thereby improving adjustment accuracy, maximizing resource utilization efficiency, and ensuring stable grid operation. Specifically, by introducing feedback control of load changes and combining it with a sensitivity adjustment factor, the magnitude of resource allocation can be flexibly controlled, avoiding excessive or insufficient resource adjustments, improving system response speed and stability. Furthermore, by comprehensively weighting task priorities and resource utilization efficiency, it ensures that critical task areas receive priority resource support while avoiding resource waste.

[0144] As an optional solution in this application, in step S5, computing resources and scheduling tasks are allocated to each region based on the load ratio, and the task complexity of each region is evaluated based on its load fluctuation and task type. Specifically, based on the regional load forecast, the load ratio of each region is first calculated, and corresponding computing resources are allocated accordingly. Assume the power grid contains multiple regions R1, R2, ..., R... n The load forecast for each region is as follows: Load ratio i The following formula can be used for calculation:

[0145]

[0146] Computing resources are allocated based on load ratio. Assume the system has M computing nodes, each with a computing capacity of C. node Then region R i The computing resources are allocated as follows:

[0147]

[0148] At this point, the load ratio is closely related to resource allocation, and computing resources for each region are allocated to meet its load requirements.

[0149] The scheduling tasks in each region have different complexities. To ensure efficient utilization of computing resources, it is necessary to analyze the task complexity and decompose the tasks according to the complexity. The task complexity analysis includes load fluctuation amplitude, scheduling task type, and regional power grid topology. The task complexity analysis function f... task (R i Calculate the complexity of the scheduling task for each region:

[0150]

[0151] Here, g() is the task complexity calculation function, which reflects the impact of load changes, task type, and system complexity on task execution time and computing resource requirements.

[0152] Based on the task complexity analysis and load requirements for each region, task decomposition and resource allocation are performed. Each task is broken down into several subtasks, which are then assigned to corresponding computing nodes for processing. The task allocation is adjusted according to the computing power and resource load of each node. The computational resource requirement C for each subtask is... subt ask (R i It can be calculated using the following formula:

[0153]

[0154] Building upon the aforementioned technical solutions, to further optimize the allocation of computing resources and load balancing among computing nodes, particularly in high-load areas, a regional collaborative optimization mechanism is introduced. The core objective of this mechanism is to enable computing nodes in adjacent regions to work collaboratively, sharing computing resources and information, thereby achieving load balancing and resource optimization. Specifically, a neighboring node collaboration mechanism and an inter-regional load sharing mechanism are introduced to ensure the sharing and collaborative work of computing resources across different regions. Load fluctuations in adjacent regions are transferred to prevent overload in any particular region. Task allocation strategies are adjusted in real-time to ensure load balancing, thereby improving resource utilization efficiency and the overall stability of the power grid.

[0155] Neighborhood node cooperation mechanism: Assume region R i Load changes affect adjacent areas R j To mitigate the impact, the following collaborative optimization strategies can be employed for resource sharing and load balancing:

[0156]

[0157] Where, λ ij This is a collaborative adjustment factor between adjacent regions, reflecting the degree of impact of load fluctuations in adjacent regions on the current region's load. This factor can be adjusted using empirical data or learning models.

[0158] Inter-regional load sharing mechanism: Through the neighboring region load sharing mechanism, load fluctuations in adjacent regions can be transferred through the corresponding sharing mechanism. For example, when a charging peak occurs in a certain region, some load can be transferred to a region with a lower load by sharing load resources, thereby avoiding overload in that region.

[0159]

[0160] Where, γi The load sharing factor reflects the contribution of each region to load regulation.

[0161] In regional collaborative optimization, it is necessary to ensure load balancing of computing nodes in each region. In this embodiment, the optimization objective is to minimize the load difference among computing nodes in all regions. Therefore, the load balancing model can be expressed as:

[0162]

[0163] Among them, L node,i (t) represents the load of the i-th computing node. This represents the average load across all computing nodes. This optimization objective allows for dynamic adjustment of the load on each computing node, ensuring a more even distribution of computing tasks.

[0164] The system adaptively adjusts load distribution based on real-time load and the status of computing nodes. If a node's load is too high, the system will dynamically adjust the task allocation for that node or transfer tasks to other nodes with lighter loads. For example, resources can be redistributed using optimal transmission algorithms (such as the shortest path algorithm or the Kuhn-Munkres algorithm) to ensure overall load balance.

[0165] By combining regional load forecasting, distributed task decomposition, and regional collaborative optimization mechanisms, resource optimization among regions is achieved. Based on real-time load and resource utilization, task allocation is intelligently adjusted to ensure efficient use of computing resources and balance of the entire power grid load. This ensures that the power grid system can efficiently allocate and adjust computing resources during peak electric vehicle charging periods or sudden load changes, not only optimizing load balancing but also improving resource utilization efficiency and ensuring stable operation of the power grid.

[0166] As an optional approach in this application, the optimized resource allocation scheme is executed in real time, and the execution status, load adjustment effect, and power grid stability indicators of each computing node are collected. Key indicators of the execution status include: CPU load, memory utilization, and network latency of each computing node; the load adjustment effect, i.e., the deviation between the actual load and the predicted load, with particular attention to the adjustment effect on electric vehicle charging load, basic user electricity consumption behavior, and distributed energy output; and power grid stability indicators include voltage deviation, frequency fluctuation, and power factor. These indicators directly affect the operational stability of the power grid and therefore require real-time monitoring.

[0167] By combining load adjustment effectiveness and grid stability, the effectiveness of resource allocation is evaluated, and the key performance indicators (KPIs) for evaluating dispatch effectiveness are designed as follows:

[0168]

[0169] Among them, Load Deviation: the deviation between predicted load and actual load; Cost of Energy: the cost based on optimization schemes and energy dispatch; Grid Stability: grid stability indicators, such as voltage deviation and frequency fluctuation; Computational Time: the time to complete the computation task, reflecting the efficiency of resource allocation.

[0170] This KPI formula comprehensively considers precise load regulation, grid stability, and computational efficiency, enabling a complete evaluation of the effectiveness of resource allocation schemes. Feedback signals are generated based on the evaluation results, informing users of the effectiveness of the current scheduling task, the rationality of resource allocation, and the stability status of the power grid.

[0171] As an optional solution in this application, by real-time monitoring of the actual operating results, the performance indicators in step S6 are used as feedback inputs to the reinforcement learning-based cloud computing resource dynamic allocation model and deep learning model for parameter updates and strategy optimization. Feedback information includes important indicators such as the effectiveness of resource allocation, the success or failure of load adjustment, and grid stability. These feedback signals help the model identify the merits of the scheduling strategy and guide subsequent optimization decisions. After receiving feedback, online learning technology is used to adjust the parameters of the resource allocation model. Taking reinforcement learning as an example, if the grid experiences overload, a penalty is imposed for load adjustment; conversely, a reward is given. Through repeated trials and feedback mechanisms, the resource allocation strategy is gradually adjusted to reduce load deviation and improve grid stability. The parameter update formula in reinforcement learning is as follows:

[0172]

[0173] Where: θ is the model parameter, η is the learning rate, and J(θ) is the loss function. The definition of the loss function J(θ) should be based on the load deviation, grid stability, and computation time in the KPIs, reflecting the effect of the model adjustment.

[0174] Based on the updated model, the system dynamically adjusts its strategy in each round of scheduling. For example, if a region experiences excessive load, the model adjusts resource allocation to prevent overburdening that region. As online learning progresses, the system gradually adapts to different load fluctuation patterns, optimizing resource allocation strategies to enable the power grid to flexibly adjust under varying load conditions, achieving the goals of load balancing and efficient scheduling across the entire network. After each round of scheduling, the model is updated based on new operational data, gradually achieving self-optimization and improving scheduling efficiency and power grid stability.

[0175] In summary, this efficient cloud computing resource dynamic allocation method, through the coordinated operation of multiple steps, aims to optimize grid load scheduling and computing resource allocation during peak electric vehicle charging periods, thereby improving grid stability and response efficiency. The core of the method revolves around real-time data acquisition, load forecasting, demand-side response, priority scheduling, dynamic resource allocation, and collaborative optimization, ensuring optimal configuration and efficient operation of network resources under dynamic load changes and multi-task scheduling.

[0176] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An efficient method for dynamically allocating cloud computing resources, characterized in that: The following steps are involved: Step S1, real-time collection of multiple load source data under the influence of electric vehicle charging behavior, including smart meters, charging piles, distributed energy, weather monitoring stations and power grid operation status data, feature analysis and extraction of electric vehicle load, basic user electricity consumption behavior, and distributed energy output in the load source data, and combined with a deep learning model to predict the short-term fluctuation trend of regional load; Step S2, based on the regional load forecast results, by simulating the local load pressure of the electric vehicle charging peak, a multi-objective optimization method is used to evaluate the potential of demand-side response for load relief, and by dynamically adjusting the user behavior incentive mechanism, the load distribution is ensured to be coordinated with the overall operation objectives of the power grid; Step S3, based on the grid dispatching demand, set multi-dimensional priority evaluation indicators, including the regional impact of electric vehicle charging, demand-side response effect, grid generation cost and transmission loss, and perform multi-level classification of dispatching tasks through a fuzzy logic model; Step S4, building a cloud computing resource dynamic allocation model based on reinforcement learning, dynamically adjusting the computing node allocation strategy according to the scheduling task priority and regional load forecast data, and achieving a balance between local efficiency and global performance; Step S5, deploy distributed computing nodes in the cloud computing platform, decompose and distribute the tasks according to regional load distribution and scheduling, introduce a regional collaborative optimization mechanism, coordinate the scheduling of computing resources and power loads in adjacent areas, and optimize the distribution of computing resources and load balancing among different nodes.

2. The efficient cloud computing resource dynamic allocation method according to claim 1, characterized in that: The following steps are also included: Step S6, executing the optimized resource allocation scheme in real time, collecting the execution status of each computing node, load adjustment effect and power grid stability index, and evaluating the key performance indicators of the scheduling effect; Step S7, establish a real-time feedback and online learning mechanism, use the system performance indicators as feedback, and transmit them back to the cloud computing resource dynamic allocation model and the deep learning model for load forecasting, and continuously optimize the resource allocation strategy through online learning technology.

3. The efficient cloud computing resource dynamic allocation method according to claim 1, characterized in that: Construct a multi-load source data input interface, integrate the collected multi-load source data, and use different data analysis methods to extract the characteristics of different load source data. Specifically: analyze the impact of charging time periods, charging power changes, and geographical locations on charging demand for electric vehicle loads; classify user groups through cluster analysis based on basic user electricity consumption behaviors, extract typical electricity consumption behavior patterns; and use regression analysis or time series analysis for distributed energy output to obtain the impact of weather changes on energy output.

4. The efficient cloud computing resource dynamic allocation method according to claim 1 or 3, characterized in that: LSTM or Transformer in deep learning is used to predict the short-term fluctuation trend of regional load. The time series characteristics and characteristic extraction data of each load source data are used as the input of the deep learning model. The data of different load sources are integrated into a unified deep learning framework to obtain the final regional load short-term fluctuation trend curve and output the future short-term load demand fluctuation.

5. The efficient cloud computing resource dynamic allocation method according to claim 1 or 4, characterized in that: In step S2, a user behavior response model is constructed according to the load status of the power grid, the electricity price during the period, the incentive mechanism and the user charging demand to adjust the charging period and power of the electric vehicle. The model optimization goal is to minimize the charging cost, which includes the economic cost and the pressure of the power grid. The user behavior response model is: Among them, P price (t) is the electricity price in period t, Lnet(t) is the load pressure of the power grid in period t, reflecting the load of the power grid, P charge (i,t) is the charging power of user i in time period t, T charge (i) is the charging period of user i, R user (t) is the reward value provided by the demand response incentive; The charging period is selected according to the changes in grid load and electricity price. The charging period selection function is expressed as: T charge (i)=f(L net (t),P price (t),C demand (i)) Among them, C demand (t) is the user's charging demand, which means the amount of electricity that the user needs to charge in a certain period of time. Function f is established through historical data and user behavior analysis; The charging power is adjusted according to the load pressure of the power grid and the demand response signal. The charging power adjustment function is expressed as: in, is the maximum load capacity of the power grid in time period t, γ is the user's sensitivity coefficient to charging power adjustment; By dynamically adjusting the incentive signal to encourage users to charge outside the peak load period, the change of the incentive mechanism is expressed as: Among them, L low and L high Represent low load and high load thresholds, R max and R min are the maximum and minimum rewards respectively.

6. The efficient cloud computing resource dynamic allocation method according to claim 5, characterized in that: According to the grid load forecast and user charging behavior, the local load pressure in each area is evaluated and expressed as: Among them, L region (t) represents the total charging load in the area at time t, reflecting the load pressure in the area, P charge (i,t) is calculated through dynamic adjustments in the user behavior response model; By comparing the changes in local load pressure before and after the demand response measures, the potential of demand-side response to alleviate local load pressure is quantified. The change in load pressure is expressed as: in, is the sum of the regional loads in the absence of demand response measures, is the total regional load after the introduction of demand response measures, ΔL region (t) represents the change in load pressure; The multi-objective optimization method is used to simultaneously consider the grid load balance and user charging demand to minimize the grid load fluctuation. The optimization objective function is: Among them, λ1 and λ2 are the weight coefficients of the optimization objective; Load balancing constraints and charging power limits are introduced to ensure that the load on the grid during the optimization process does not exceed the maximum carrying capacity of the grid: Among them, L min (t) and L max (t) is the minimum and maximum permissible value of the grid load, and The minimum and maximum limits for charging power.

7. The efficient cloud computing resource dynamic allocation method according to claim 6, characterized in that: In step S3, the scheduling tasks are divided into three priorities: high priority, medium priority and low priority. The regional impact of electric vehicle charging, demand response, grid generation cost and transmission loss are converted into processable fuzzy sets using the fuzzy logic model, and the scheduling tasks are dynamically assigned priorities through rule reasoning. EV , load reduction ratio R DS 、Multi-source power generation cost C gen and line transmission loss L loss Characterize the influencing factors, specifically: Among them, I EV (r,t) represents the impact of electric vehicle charging in area r during time period t, L net (r,t) is the grid load in region r, R DS (r,t) represents the load reduction ratio of region r in period t after the demand-side response, C gen (t) represents the overall power generation cost obtained by weighting the power generation cost of each power source in time period t, P gen,k (t) is the power generation of the kth power generation unit in time period t, C cost,k is the unit power generation cost of the kth power generation unit, L loss (t) represents the line transmission loss in time period t, Il(t) is the current of line l in time period t, R l is the resistance of line l; The fuzzy inference algorithm is used to comprehensively evaluate the input fuzzy values ​​to obtain the priority of each task, and the final priority value is obtained through the defuzzification process: P task (t)=defuzzify(fuzzy({I EV ,R DS ,C gen ,L loss })) Among them, P task (t) is the final priority of the task, and defuzzify is the defuzzification process.

8. The efficient cloud computing resource dynamic allocation method according to claim 1, characterized in that: In step S4, the state space in the cloud computing resource dynamic allocation model structure based on reinforcement learning contains input information for reflecting the resource configuration and load conditions of the current power grid, including regional load forecast data, scheduling task priority, current resource configuration conditions and resource utilization efficiency. The action space includes the adjustment of computing resource allocation, resource adjustment strategy and resource scheduling method. The reward function comprehensively considers regional load, scheduling task priority, resource utilization efficiency and scheduling cost.

9. The efficient cloud computing resource dynamic allocation method according to claim 8, characterized in that: A real-time adaptive mechanism is introduced to dynamically adjust the allocation strategy of computing nodes according to the load forecast error and the current resource allocation status. The dynamic adjustment mechanism is as follows: a′ t =f(△L′ region (t),a t )=α·a t +β·adjustment factor Where at′ is the adjusted resource allocation decision, f(·) is the function used to dynamically adjust the resource allocation strategy, α is the weight coefficient of the current resource allocation, α∈[0,1], β is the adjustment coefficient, β∈[0,1], and the adjustment factor is composed of multiple factors, reflecting the impact of load changes, priority and resource utilization efficiency, that is, adjustment factor(ΔL' region (t),P task (t),R util (t)) is the weighted sum of multiple factors, ΔL' region (t) represents the actual deviation of the grid load, R util (t) represents the resource utilization efficiency of the region. adjustment factor=λ1·|ΔL′ region (t)|+λ2·|ΔL′ region(t) | γ +w1P task (t)+w2·(1-R util (t)) 10. The efficient cloud computing resource dynamic allocation method according to claim 1, characterized in that: Allocate computing resources and schedule tasks to each region based on load ratio, and evaluate the task complexity of each region based on its load fluctuation and task type; Decompose tasks and allocate resources based on the task complexity and load requirements of each area; A neighborhood node coordination mechanism and an inter-regional load sharing mechanism are introduced to ensure that computing resources in different regions are shared and work together, load fluctuations in adjacent regions are transferred to avoid overload in a certain area, and task allocation strategies are adjusted in real time to ensure load balancing.

Images