Resource allocation method based on power safety target system

Abstract

The present application relates to a kind of resource allocation methods based on electric power safety target system, including according to the establishment of virtual resource training model of electric power target implementation resource data, resource training model is trained in combination with current allocation item, under the premise of reserving the greater emergency resource threshold of emergency deployable resource, output resource allocation scheme, and real-time fusion processing is carried out to output resource allocation scheme as time goes on.The present application provides a kind of resource allocation methods based on electric power safety target system, the traditional electric power target resource model is improved by machine learning technology, proposes the dynamic configuration model of deployable resource, fuses big data technology, solves model by training big data and cross validation, and in calculation, with emergency resource threshold instead of uncertain resource allocation data, avoids the uncertainty of model, so that the model is more reasonable, reduces resource overflow, improves resource utilization efficiency.

Description

Technical Field

[0001] This invention relates to a resource allocation method, and more particularly to a resource allocation method based on a power safety test range system. Background Technology

[0002] As a critical national infrastructure, the power system faces immense cybersecurity pressure. In particular, uneven resource allocation within the power system can lead to sudden outages of certain important modules or hardware. Therefore, we propose a resource allocation method based on a power security test range system. Summary of the Invention

[0003] The purpose of this invention is to address the deficiencies mentioned in the background art by proposing a resource allocation method based on a power safety test range system.

[0004] The technical solution adopted in this invention is as follows:

[0005] A resource allocation method based on a power safety test range system is provided, comprising the following steps:

[0006] A virtual resource training model is established based on the resource data of the power test site. The model is trained in conjunction with the current allocation items. Under the premise that the available emergency resources are greater than the emergency resource threshold, a resource allocation plan is output. The output resource allocation plan is then integrated and processed in real time over time.

[0007] As a preferred embodiment of the present invention, the steps of the virtual resource training model are as follows:

[0008] S1: Model initialization;

[0009] S2: Input model parameters and perform the i-th iteration of training on the training dataset (i = 1, 2, ..., n);

[0010] S3: Cross-validate the model on the validation dataset and determine if it meets the requirements;

[0011] S4: If the requirements are not met, adjust the model and return to S1;

[0012] S5: If the requirements are met, output the model that performs best in each cross-validation group, and perform aggregate training on the two models;

[0013] S6: Output the final model.

[0014] As a preferred embodiment of the present invention, the constraints of the virtual resource training model are as follows:

[0015] The total amount of resources available for allocation within any time period cannot exceed the maximum resource utilization of each module in the system during that time period, and the maximum resource utilization cannot exceed twice the amount of resources available for emergency allocation. Specifically:

[0016]

[0017] Among them, S t0 S represents the total amount of resources available for allocation within the time period t0. i S represents the maximum resource utilization of each module in the system during the time period t0. y The amount of resources that can be urgently allocated, where n is the number of modules in the system.

[0018] As a preferred embodiment of the present invention, the real-time fusion processing of the output resource allocation scheme over time includes the following steps:

[0019] Resource training model initialization; the power safety test range server sends the model parameters to the first test range simulator. On the test range simulator, two steps are performed simultaneously: one part updates all model parameters using private data; the other part updates the personalized layers to build a personalized model based on the base layer number, which is fixed, and uploads all the updated model parameters to the power safety test range server. The power safety test range server then merges the model parameters and moves them to the next test range simulator, and so on.

[0020] As a preferred embodiment of the present invention, the real-time fusion processing of the output resource allocation scheme over time includes the following steps:

[0021] Resource training model initialization; the power safety test range server sends the model parameters to the first test range simulator. On the test range simulator, two steps are performed simultaneously: one part updates all model parameters using private data; the other part updates the personalized layers to build a personalized model based on the base layer number, which is fixed, and uploads all the updated model parameters to the power safety test range server. The power safety test range server then merges the model parameters and moves them to the next test range simulator, and so on.

[0022] As a preferred embodiment of the present invention: the real-time fusion processing further includes, in the deep learning scenario, representing the K target range simulators as (N1, N2, ..., N...). K In the target range simulator, gradient descent is used to update the local model parameters, as shown in the following formula: The power safety test range server uses a model knowledge transfer algorithm to aggregate the updated model parameters from the first two test range simulators and transfers the updated parameters to the next newly appearing test range simulator, as shown below:

[0023]

[0024] in For round t, N k Model parameters of the target range simulator For round t, N k-1 Model parameters of the target range simulator, N k-1 For N k-1 The resource size of the target range simulator, N k-2 For N k-2 The size of the resources in the target range simulator.

[0025] As a preferred technical solution of the present invention: the training dataset is optimized by the discrete points of the time series, and its trend change pattern is judged by scatter plot, autocorrelation and partial correlation graphs to identify whether it belongs to stationary or non-stationary series. If it belongs to non-stationary time series, it is stationarized and heteroscedasticity is eliminated to make the above parameters approach zero, so as to meet the analysis requirements of stationary series.

[0026] As a preferred technical solution of the present invention: in the process of optimization through discrete points of time series, the penalty factor and kernel function parameters of the support vector machine are optimized by the gravitational search algorithm.

[0027] As a preferred embodiment of the present invention, the support vector machine is specifically:

[0028]

[0029] Where K(x) i ,x)=Φ(x i )Φ(x), where Φ(x) i Φ(x) is the kernel function that satisfies the Mercer condition, a i , Let the Lagrange multiplier in the quadratic programming problem be a set of l training samples: {(x... i y i ), i = 1, 2, ..., l}, b is a constant;

[0030] The kernel function Φ(x) i Φ(x) is specifically:

[0031] K(x i ,x)=exp(-g||xx i || 2 )

[0032] Where g > 0.

[0033] As a preferred embodiment of the present invention: the optimization of the penalty factor and kernel function parameters of the support vector machine using the universal gravitation search algorithm specifically involves:

[0034]

[0035] Among them, y i Let be the measured value of the i-th sample. Let be the simulated value of the i-th sample.

[0036] This invention provides a resource allocation method based on a power safety test range system. It improves upon traditional power test range resource models using machine learning technology, proposing a dynamic allocation model for deployable resources. Integrating big data technology, the model is solved through training with large datasets and cross-validation. Furthermore, an emergency resource threshold replaces uncertain resource allocation data in the calculation, avoiding model uncertainty and making the model more reasonable. Finally, testing and comparative analysis were conducted. Under the condition of constant model resource quantity, resource overflow is reduced and resource utilization efficiency is improved. Detailed Implementation

[0037] It should be noted that, unless otherwise specified, the embodiments and features described in this embodiment can be combined with each other. The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. 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.

[0038] A preferred embodiment of the present invention provides a resource allocation method based on a power safety test range system, comprising the following steps:

[0039] A virtual resource training model is established based on the resource data of the power test site. The model is trained in conjunction with the current allocation items. Under the premise that the available emergency resources are greater than the emergency resource threshold, a resource allocation plan is output. The output resource allocation plan is then integrated and processed in real time over time.

[0040] The steps for training the virtual resource model are as follows:

[0041] S1: Model initialization;

[0042] S2: Input model parameters and perform the i-th iteration of training on the training dataset (i = 1, 2, ..., n);

[0043] S3: Cross-validate the model on the validation dataset and determine if it meets the requirements;

[0044] S4: If the requirements are not met, adjust the model and return to S1;

[0045] S5: If the requirements are met, output the model that performs best in each cross-validation group, and perform aggregate training on the two models;

[0046] S6: Output the final model.

[0047] In addition, the constraints for training the virtual resource model are:

[0048] The total amount of resources available for allocation within any time period cannot exceed the maximum resource utilization of each module in the system during that time period, and the maximum resource utilization cannot exceed twice the amount of resources available for emergency allocation. Specifically:

[0049]

[0050] Among them, S t0 S represents the total amount of resources available for allocation within the time period t0. i S represents the maximum resource utilization of each module in the system during the time period t0. y The amount of resources that can be urgently allocated, where n is the number of modules in the system.

[0051] The objective function of the virtual resource training model is:

[0052]

[0053] in, σ represents the total amount of available resources in the model that performed best in the previous cross-validation, and σ is the model parameter.

[0054] First, input the emergency resource threshold and corresponding model parameters into the model to complete model initialization. Then, iteratively train on the training dataset. After training, perform cross-validation. Adjust the emergency resource threshold and model parameters input into the model based on the results and continue training on the training dataset until the model with the best performance in each cross-validation process is trained. Finally, perform a final aggregation training on the two models to obtain the optimal model.

[0055] The real-time fusion processing of output resource allocation schemes over time includes the following steps:

[0056] Resource training model initialization; the power safety test range server sends the model parameters to the first test range simulator. On the test range simulator, two steps are performed simultaneously: one part updates all model parameters using private data; the other part updates the personalized layers to build a personalized model based on the base layer number, which is fixed, and uploads all the updated model parameters to the power safety test range server. The power safety test range server then merges the model parameters and moves them to the next test range simulator, and so on.

[0057] Real-time fusion processing also includes, in deep learning scenarios, representing K target range simulators as (N1, N2, ..., N...). K In the target range simulator, gradient descent is used to update the local model parameters, as shown in the following formula:

[0058] The power safety test range server uses a model knowledge transfer algorithm to aggregate the updated model parameters from the first two test range simulators and transfers the updated parameters to the next newly appearing test range simulator, as shown below:

[0059]

[0060] in For round t, N k Model parameters of the target range simulator For round t, N k-1 Model parameters of the target range simulator, N k-1 For N k-1 The resource size of the target range simulator, n k-2 For N k-2 The size of the resources in the target range simulator.

[0061] A fusion algorithm based on model knowledge transfer is proposed. When a new target simulator appears, the knowledge learned from the target simulator is transferred to the server to fuse the models. Experiments show that the algorithm significantly outperforms traditional deep learning frameworks when the data distribution among the target simulators is either independent and identically distributed (i.i.d.) or non-independent and identically distributed (i.i.d.).

[0062] Secondly, the training dataset is optimized by using discrete points of the time series. The trend changes are judged by scatter plots, autocorrelation and partial correlation graphs to identify whether it is a stationary or non-stationary series. If it is a non-stationary time series, it is stationary by performing heteroscedasticity elimination on the data to make the above parameters approach zero, thus meeting the analysis requirements of stationary series.

[0063] During the optimization process using discrete points of the time series, the penalty factor and kernel function parameters of the support vector machine are optimized using the gravitational search algorithm.

[0064] Support Vector Machine (SVM) is specifically:

[0065]

[0066] Where K(x) i ,x)=Φ(x i )Φ(x), where Φ(x) i Φ(x) is the kernel function that satisfies the Mercer condition, a i , Let the Lagrange multiplier in the quadratic programming problem be a set of l training samples: {(x... i y i ), i = 1, 2, ..., l}, b is a constant;

[0067] The kernel function Φ(x) i Φ(x) is specifically:

[0068] K(x i ,x)=exp(-g||xx i || 2 )

[0069] Where g > 0.

[0070] The specific optimization of the penalty factor and kernel function parameters of the support vector machine using the gravitational search algorithm is as follows:

[0071]

[0072] Among them, y i Let be the measured value of the i-th sample. Let be the simulated value of the i-th sample.

[0073] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of equivalents of the claims be included within the present invention.

[0074] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A resource allocation method based on a power safety test range system, characterized in that: Includes the following steps: A virtual resource training model is established based on resource data from the power test site. This model is then trained using current allocation items. While ensuring that the available emergency resources exceed the emergency resource threshold, a resource allocation plan is output. This output resource allocation plan is then fused in real-time over time. The real-time fusion of the output resource allocation plan over time includes the following steps: Resource training model initialization; The power safety test range server sends the model parameters to the first test range simulator. Two steps are performed simultaneously on the test range simulator: one part updates all model parameters using private data; the other part updates the personalized layers to build a personalized model based on the base layer number, which is fixed, and uploads all the updated model parameters to the power safety test range server. The power safety test range server then merges the model parameters and moves them to the next test range simulator, and so on. The real-time fusion processing also includes, in deep learning scenarios, […]. Each range simulator is represented as In the target range simulator, gradient descent is used to update the local model parameters, as shown in the following formula: ; The power safety test range server uses a model knowledge transfer algorithm to aggregate the updated model parameters from the first two test range simulators and transfers the updated parameters to the next newly appearing test range simulator, as shown below: ； in for round Model parameters of the target range simulator for round Model parameters of the target range simulator for The size of the target range simulator's resources. for The size of the target range simulator's resources. for round Model parameters of the target range simulator.

2. The resource allocation method based on the power safety test range system according to claim 1, characterized in that: The steps for training the virtual resource model are as follows: S1: Model initialization; S2: Input model parameters, perform the first step on the training dataset. The next iteration of training ( ); S3: Cross-validate the model on the validation dataset and determine if it meets the requirements; S4: If the requirements are not met, adjust the model and return to S1; S5: If the requirements are met, output the model that performs best in each cross-validation group, and perform aggregate training on the two models; S6: Output the final model.

3. The resource allocation method based on the power safety test range system according to claim 1, characterized in that: The constraints of the virtual resource training model are: The total amount of resources available for allocation within any time period cannot exceed the maximum resource utilization of each module in the system during that time period, and the maximum resource utilization cannot exceed twice the amount of resources available for emergency allocation. Specifically: ； in, for The total amount of resources available for allocation within a given time period. for Maximum resource utilization of each module in the system within a given time period. In order to enable the emergency allocation of resources, This represents the number of modules in the system.

4. The resource allocation method based on the power safety test range system according to claim 1, characterized in that: The objective function of the virtual resource training model is: ； in, This represents the total amount of available resources in the model that performed best in the previous cross-validation set. These are the model parameters.

5. The resource allocation method based on the power safety test range system according to claim 2, characterized in that: The training dataset is optimized using discrete points of the time series. Its trend changes are determined by scatter plots, autocorrelation, and partial correlation graphs to identify whether it is a stationary or non-stationary series. If it is a non-stationary time series, it is stationary by performing heteroscedasticity elimination on the data to make the above parameters approach zero, thus meeting the analysis requirements of stationary series.

6. The resource allocation method based on the power safety test range system according to claim 5, characterized in that: In the process of optimization using discrete points of the time series, the penalty factor and kernel function parameters of the support vector machine are optimized using the gravitational search algorithm.

7. The resource allocation method based on the power safety test range system according to claim 6, characterized in that: The support vector machine is specifically: ； in, ,in, A kernel function that satisfies the Mercer condition. , For Lagrange multipliers in a quadratic programming problem, containing The set of training samples is: , It is a constant; The kernel function Specifically: ； in, .

8. The resource allocation method based on the power safety test range system according to claim 7, characterized in that: The optimization of the penalty factor and kernel function parameters of the support vector machine using the gravitational search algorithm specifically involves: ； in, For the first Measured values ​​of a sample For the first Simulated values ​​for a sample.

Images