Cloud computing power dynamic allocation and optimization management method

Abstract

The invention belongs to the technical field of power grid Internet of Things security, and more particularly relates to a cloud computing power dynamic distribution and optimization management method, which comprises the following steps of: acquiring current load data of a power grid cloud, preprocessing the data, ensuring the accuracy and the real-time performance of the data, and using a machine learning model to obtain a cloud computing power dynamic distribution model; and training and predicting the preprocessed load data to obtain a future load condition, carrying out dynamic computing power distribution through a fusion decision-making mechanism in combination with a current load and a prediction result to realize load harmony and resource optimization, and calculating resource requirements of each task and the whole system according to the fused load data so as to obtain a load prediction result. The cloud computing resources are dynamically adjusted, the optimal allocation and management of the resources are realized, the technical problems in the cloud computing power allocation and management of the power grid can be effectively solved, the stability of the system and the resource utilization efficiency are improved, and the intelligent and informatization management of the power grid is realized.

Description

Technical Field

[0001] The present application belongs to the field of power grid Internet of Things security technology, and more specifically, to a method for dynamic allocation and optimization management of cloud computing power. Background Art

[0002] With the continuous advancement of the intelligence and informatization of power systems, advanced technologies such as big data, cloud computing and artificial intelligence have gradually been introduced into the power grid field. The application of these technologies in the operation and management of power grids has not only improved the operating efficiency and stability of power grids, but also provided strong technical support for the intelligent development of power grids.

[0003] Popularization of smart grid:

[0004] Smart meters and sensors: The widespread deployment of smart meters and various sensors enables the power grid to collect a large amount of operating data in real time, including power load, voltage, current, etc.

[0005] Surge in data volume: With the development of smart grids, the amount of grid operation data has exploded, which has put higher requirements on data storage, processing and analysis.

[0006] Application of cloud computing in power grid:

[0007] Cloud computing platform: Power companies are gradually establishing cloud computing platforms for data storage, computing and analysis.

[0008] Elastic expansion: The elastic expansion capability of cloud computing enables power companies to flexibly adjust computing and storage resources according to actual needs and optimize resource utilization.

[0009] Artificial Intelligence and Machine Learning:

[0010] Load forecasting: Through machine learning models (such as LSTM), the grid load can be accurately predicted and measures can be taken in advance to avoid grid overload or resource waste.

[0011] Fault prediction and diagnosis: Machine learning and big data analysis can be used to predict and diagnose power grid faults in advance, thereby improving the reliability of the power grid.

[0012] Current status of power grid cloud computing power

[0013] Although the application of cloud computing and artificial intelligence technologies in power grids has made certain progress, there are still some problems in practical applications.

[0014] In view of the real-time nature of the power grid and the real-time changes in load, how to dynamically adjust cloud computing resources to achieve optimal resource allocation is a problem that needs to be solved. Summary of the invention

[0015] The present invention provides a method for dynamic allocation and optimization management of cloud computing power, which is intended to realize dynamic allocation of cloud computing resources of power grid.

[0016] The method for dynamic allocation and optimization management of cloud computing power is characterized by comprising the following steps:

[0017] Collect the current load data of the power grid cloud, pre-process the current load data, and use the pre-processed power grid load data as the input of the machine learning model to obtain the future load situation;

[0018] Based on the fusion decision-making mechanism, dynamic computing power allocation is carried out in combination with the current load and prediction results to achieve load balancing and resource optimization.

[0019] The present invention collects the current load data of the power grid, pre-processes the data to ensure the accuracy and real-time nature of the data, uses a machine learning model to train and predict the pre-processed load data, obtains the future load situation, combines the current load and the prediction result, and performs dynamic computing power allocation through a fusion decision-making mechanism to achieve load and resource optimization. In addition, based on the fused load data, the resource requirements of each task and the overall system are calculated, and cloud computing resources are dynamically adjusted to achieve optimal allocation and management of resources. This can effectively solve the technical problems in the allocation and management of cloud computing power in the power grid, improve the stability of the system and the efficiency of resource utilization, and realize the intelligent and information management of the power grid.

[0020] Preferably, the pretreatment comprises the following steps:

[0021] Missing value processing: use the mean to fill in missing data;

[0022] Data smoothing: Using the moving average method, the smoothed data is but:

[0023]

[0024] Where: is the smoothed load data; n is the size of the sliding window;

[0025] Data normalization: Normalize the smoothed data to the same dimension.

[0026] Preferably, the machine learning model is an LSTM model, and the structure of the LSTM model is as follows:

[0027] Input layer: receives input data of shape: time steps and number of features;

[0028] Hidden layer: contains several LSTM units;

[0029] Output layer: a fully connected neuron used to predict future load values;

[0030] Loss function:

[0031]

[0032] Where: α, β and γ represent the weight coefficient of the mean square error part, the weight coefficient of the peak part and the weight coefficient of the load smoothness part respectively; represents the load predicted at the i-th moment; y i represents the actual load at the i-th moment; max(y) represents the actual load peak; represents the predicted load peak; t represents the length of the time series;

[0033] The Adam optimization algorithm is used in the training process to minimize the loss function.

[0034] Preferably, the specific steps of the Adam optimization algorithm for minimizing the loss function are as follows:

[0035] According to the current parameter θ t-1 Calculating the loss function Gradient with respect to the parameters:

[0036]

[0037] Update the dynamic estimate: Calculate the first-order momentum estimate m t and the second-order momentum estimate v t :

[0038] m t =β 1,t ·m t-1 +(1-β 1,t )·g t ;

[0039]

[0040] Where: β 1,t represents the adjusted momentum parameter; m t-1 represents the first-order momentum estimate at time step t-1; β2 represents the second-order moment parameter; v t-1 represents the second-order momentum estimate at time step t-1;

[0041] Applying AMSGrad to Calculate Corrected Second-Order Momentum Estimates

[0042]

[0043] Where: represents the corrected second-order momentum estimate at time step t-1;

[0044] Bias correction: Get the corrected first-order momentum based on the bias correction and second-order momentum

[0045]

[0046] Where: represents β at time step t 1,t t to the power of represents β2 to the power of t;

[0047] Update the parameters θ using the corrected momentum estimate t :

[0048]

[0049] Where: α t represents the learning rate after warm-up; ∈ t represents the bias correction term at time step t;

[0050] Repeat the above specific steps of minimizing the loss function and adjust the parameter θ to minimize the loss function.

[0051] Preferably, the fusion decision mechanism is as follows:

[0052] Make fusion decisions based on the future load forecast data combined with the current load data:

[0053] Load fused =μ·Load current +(1-μ)·Load pred ;

[0054]

[0055] Where: Load fused Represents the fused load data; Res fused represents the fused resource index; μ and Indicates the fusion parameter, between 0 and 1; Load current Indicates the current actual load data; Load pred Represents future load forecast data; Res current Indicates the current actual resource usage; Res pred Represents future resource demand forecast data.

[0056] Preferably, the dynamic computing power is dynamically allocated according to the fused load data and resource requirements:

[0057] Calculate the amount of resources Res required for each tasktask :

[0058]

[0059] Where: Load fused Represents the fused load data; Total tasks Indicates the amount of resources required for each task;

[0060] Resource allocation strategy:

[0061] When Res fused >Res current , it indicates that the system currently has insufficient resources and needs to increase computing resources:

[0062] Extra resources =Res fused -Res current ;

[0063] In the formula: Extra fesources Indicates the need for additional resources;

[0064] According to Extra resources Increase the number of VMs or containers; allocate additional resources to individual tasks:

[0065]

[0066] Where: Indicates the additional resources allocated to each task;

[0067] When Res fused <Res current , it indicates that the system currently has excess resources and needs to reduce computing resources:

[0068] Reduce resources =Res current -Res fused ;

[0069] Where: Reduce resources Indicates the resources that need to be reduced;

[0070] According to Reduce resources Reduce the number of virtual machines or containers and recycle excess resources:

[0071]

[0072] Preferably, when allocating tasks, the minimum number of connections method is used for allocation, that is, the tasks are allocated to the nodes with the least number of current connections to ensure load balancing.

[0073] The beneficial effects of the present invention include:

[0074] The present invention collects the current load data of the power grid, pre-processes the data to ensure the accuracy and real-time nature of the data, uses a machine learning model to train and predict the pre-processed load data, obtains the future load situation, combines the current load and the prediction result, and performs dynamic computing power allocation through a fusion decision-making mechanism to achieve load and resource optimization. In addition, based on the fused load data, the resource requirements of each task and the overall system are calculated, and cloud computing resources are dynamically adjusted to achieve optimal allocation and management of resources. This can effectively solve the technical problems in the allocation and management of cloud computing power in the power grid, improve the stability of the system and the efficiency of resource utilization, and realize the intelligent and information management of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0076] Figure 1 An overall step block diagram provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0077] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0078] See also Figure 1 As shown, the best embodiment of the present invention is further described;

[0079] The method for dynamic allocation and optimization management of cloud computing power is characterized by comprising the following steps:

[0080] Collect the current load data of the power grid (load data includes: CPU usage: CPU usage at different time points. Memory usage: memory usage at different time points. Network bandwidth usage: network bandwidth usage at different time points. Storage usage: storage resource usage at different time points.), and pre-process the current load data, and use the pre-processed power grid load data as the input of the machine learning model to obtain the future load situation;

[0081] The pre-processing comprises the following steps:

[0082] Missing value processing: use the mean to fill in missing data;

[0083] Data smoothing: Using the moving average method, the smoothed data is but:

[0084]

[0085] Where: is the smoothed load data; n is the size of the sliding window;

[0086] Data normalization: Normalize the smoothed data to the same dimension.

[0087] The machine learning model is an LSTM model, and the structure of the LSTM model is as follows:

[0088] Input layer: receives input data of shape: time steps and number of features;

[0089] Hidden layer: contains several LSTM units;

[0090] Output layer: a fully connected neuron used to predict future load values;

[0091] Loss function:

[0092]

[0093] Where: α, β and γ represent the weight coefficient of the mean square error part, the weight coefficient of the peak part and the weight coefficient of the load smoothness part respectively; represents the load predicted at the i-th moment; y i represents the actual load at the i-th moment; max(y) represents the actual load peak; represents the predicted load peak; t represents the length of the time series;

[0094] In this embodiment, by defining the loss function, including mean square error, peak error and smoothness error, the model performance can be optimized more comprehensively, not only focusing on the accuracy of load forecasting but also ensuring the smoothness and peak accuracy of the forecast results; it is of great significance for resource optimization and scheduling.

[0095] The Adam optimization algorithm is used in the training process to minimize the loss function.

[0096] Preferably, the specific steps of the Adam optimization algorithm for minimizing the loss function are as follows:

[0097] According to the current parameter θ t-1 Calculating the loss function Gradient with respect to the parameters:

[0098]

[0099] Update the dynamic estimate: Calculate the first-order momentum estimate m t and the second-order momentum estimate v t :

[0100] m t =β 1,t ·m t-1 +(1-β 1,t )·g t ;

[0101]

[0102] Where: β 1,t represents the adjusted momentum parameter; m t-1 represents the first-order momentum estimate at time step t-1; β2 represents the second-order moment parameter; v t-1 represents the second-order momentum estimate at time step t-1;

[0103] Applying AMSGrad to Calculate Corrected Second-Order Momentum Estimates

[0104]

[0105] Where: represents the corrected second-order momentum estimate at time step t-1;

[0106] Bias correction: Get the corrected first-order momentum based on the bias correction and second-order momentum

[0107]

[0108] Where: represents β at time step t 1,t t to the power of represents β2 to the power of t;

[0109] Update the parameters θ using the corrected momentum estimate t :

[0110]

[0111] Where: α t represents the learning rate after warm-up; ∈ t represents the bias correction term at time step t;

[0112] Repeat the above specific steps of minimizing the loss function and adjust the parameter θ to minimize the loss function.

[0113] In this embodiment, AMSGrad is used to correct the second-order momentum estimation, which avoids the problem of learning rate decrease caused by excessive growth of the second-order momentum, and further improves the stability and robustness of the algorithm.

[0114] Based on the fusion decision-making mechanism, dynamic computing power allocation is carried out in combination with the current load and prediction results to achieve load balancing and resource optimization.

[0115] The fusion decision mechanism is as follows:

[0116] Make fusion decisions based on the future load forecast data combined with the current load data:

[0117] Load fused =μ·Load current +(1-μ)·Load pred ;

[0118]

[0119] Where: Load fused Represents the fused load data; Res fused represents the fused resource index; μ and Indicates the fusion parameter, between 0 and 1; Load current Indicates the current actual load data; Load pred Represents future load forecast data; Res current Indicates the current actual resource usage; Res pred Represents future resource demand forecast data.

[0120] Preferably, the dynamic adjustment is dynamically allocated according to the fused load data and resource requirements:

[0121] Calculate the amount of resources Res required for each task task :

[0122]

[0123] Where: Load fused Represents the fused load data; Total tasks Indicates the amount of resources required for each task;

[0124] Resource allocation strategy:

[0125] When Res fused >Res current , it indicates that the system currently has insufficient resources and needs to increase computing resources:

[0126] Extra resources =Res fused -Rescurrent ;

[0127] In the formula: Extra resources Indicates the need for additional resources;

[0128] According to Extra resources Increase the number of VMs or containers; allocate additional resources to individual tasks:

[0129]

[0130] Where: Indicates the additional resources allocated to each task;

[0131] When Res fused <Res current , it indicates that the system currently has excess resources and needs to reduce computing resources:

[0132] Reduce resources =Res current -Res fused ;

[0133] Where: Reduce resources Indicates the resources that need to be reduced;

[0134] According to Reduce resources Reduce the number of virtual machines or containers and recycle excess resources:

[0135]

[0136] In this embodiment, the fusion decision mechanism combines the current load data and the future load forecast data to provide comprehensive load information and improve the accuracy of decision making. Adjustment can achieve a smooth transition between the current load and the future predicted load, avoiding drastic fluctuations in resource allocation; the fusion parameters are dynamically adjusted according to the actual situation, so that the system can adapt to the resource allocation requirements under different load conditions. In addition, this embodiment dynamically adjusts the allocation of computing resources according to the fused load data and resource requirements to achieve the optimal use of resources. By calculating the amount of resources required for each task and allocating resources reasonably, it ensures the load balance of each task and avoids overloading of certain nodes. When resources are insufficient, virtual machines or containers are added, and when resources are excessive, virtual machines or containers are reduced to ensure the extensibility and scalability of the system.

[0137] Preferably, when allocating tasks, the minimum number of connections method is used for allocation, that is, the tasks are allocated to the nodes with the least number of current connections to ensure load balancing.

[0138] The present invention collects the current load data of the power grid, pre-processes the data to ensure the accuracy and real-time nature of the data, uses a machine learning model to train and predict the pre-processed load data, obtains the future load situation, combines the current load and the prediction result, and performs dynamic computing power allocation through a fusion decision-making mechanism to achieve load and resource optimization. In addition, based on the fused load data, the resource requirements of each task and the overall system are calculated, and cloud computing resources are dynamically adjusted to achieve optimal allocation and management of resources. This can effectively solve the technical problems in the allocation and management of cloud computing power in the power grid, improve the stability of the system and the efficiency of resource utilization, and realize the intelligent and information management of the power grid.

[0139] The above are only preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for dynamic allocation and optimization management of cloud computing power, characterized in that: The following steps are involved: Collect the current load data of the power grid cloud, pre-process the current load data, and use the pre-processed power grid load data as the input of the machine learning model to obtain the future load situation; Based on the fusion decision-making mechanism, dynamic computing power allocation is carried out in combination with the current load and prediction results to achieve load balancing and resource optimization.

2. The method for dynamic allocation and optimization management of cloud computing power according to claim 1, characterized in that: The pre-processing comprises the following steps: Missing value processing: use the mean to fill in missing data; Data smoothing: Using the moving average method, the smoothed data is but: Where: is the smoothed load data; n is the size of the sliding window; Data normalization: Normalize the smoothed data to the same dimension.

3. The method for dynamic allocation and optimization management of cloud computing power according to claim 1, characterized in that: The machine learning model is an LSTM model, and the structure of the LSTM model is as follows: Input layer: receives input data of shape: time steps and number of features; Hidden layer: contains several LSTM units; Output layer: a fully connected neuron used to predict future load values; Loss function: Where: α, β and γ represent the weight coefficient of the mean square error part, the weight coefficient of the peak part and the weight coefficient of the load smoothness part respectively; represents the load predicted at the i-th moment; y i represents the actual load at the i-th moment; max(y) represents the actual load peak; represents the predicted load peak; t represents the length of the time series; The Adam optimization algorithm is used in the training process to minimize the loss function.

4. The method for dynamic allocation and optimization management of cloud computing power according to claim 3, characterized in that: The specific steps of the Adam optimization algorithm for minimizing the loss function are as follows: According to the current parameter θ t-1 Calculating the loss function Gradient with respect to the parameters: Update the dynamic estimate: Calculate the first-order momentum estimate m t and the second-order momentum estimate v t : m t =b 1,t ·m t-1 +(1-β 1,t )·g t ; Where: β 1,t represents the adjusted momentum parameter; m t-1 represents the first-order momentum estimate at time step t-1; β2 represents the second-order moment parameter; v t-1 represents the second-order momentum estimate at time step t-1; Applying AMSGrad to Calculate Corrected Second-Order Momentum Estimates Where: represents the corrected second-order momentum estimate at time step t-1; Bias correction: Get the corrected first-order momentum based on the bias correction and second-order momentum Where: represents β at time step t 1,t t to the power of represents β2 to the power of t; Update the parameters θ using the corrected momentum estimate t : Where: α t represents the learning rate after warm-up; ∈ t represents the bias correction term at time step t; Repeat the above specific steps of minimizing the loss function and adjust the parameter θ to minimize the loss function.

5. The method for dynamic allocation and optimization management of cloud computing power according to claim 1, characterized in that: The fusion decision mechanism is as follows: Make fusion decisions based on the future load forecast data combined with the current load data: Load fused =μ·Load current +(1-μ)·Load pred ; Where: Load fused Represents the fused load data; Res fused represents the fused resource index; μ and Indicates the fusion parameter, between 0 and 1; Load current Indicates the current actual load data; Load pred Represents future load forecast data; Res current Indicates the current actual resource usage; Res pred Represents future resource demand forecast data.

6. The method for dynamic allocation and optimization management of cloud computing power according to claim 5, characterized in that: The dynamic computing power is dynamically allocated according to the fused load data and resource requirements: Calculate the amount of resources Res required for each task task : Where: Load fused Represents the fused load data; Total tasks Indicates the amount of resources required for each task; Resource allocation strategy: When Res fused >Res current , it indicates that the system currently has insufficient resources and needs to increase computing resources: Extra resources =Res fused -Beef current ; In the formula: Extra resources Indicates the need for additional resources; According to Extra resources Increase the number of VMs or containers; allocate additional resources to individual tasks: Where: Indicates the additional resources allocated to each task; When Res fused <Res current , it indicates that the system currently has excess resources and needs to reduce computing resources: Reduce resources =Res current -Res fused ; Where: Reduce resources Indicates the resources that need to be reduced; According to Reduce resources Reduce the number of virtual machines or containers and recycle excess resources:

7. The method for dynamic allocation and optimization management of cloud computing power according to claim 1, characterized in that: When allocating tasks, the minimum number of connections method is used, that is, the tasks are allocated to the node with the least number of current connections to ensure load balancing.

Images