5G power multi-service slice resource allocation method

Abstract

The invention discloses a 5G power multi-service slice resource allocation method, which comprises the following steps of: extracting historical network resource demand information of services from historical data, and performing local time scale-based time sequence feature extraction on the historical network resource demand information by adopting a deep learning-based data processing technology; the method comprises the following steps of: acquiring a fine-grained time sequence change trend of the service network resource demand by using a time sequence of the service network resource demand, and further performing dynamic propagation aggregation on network resource demand characteristics on each local time scale based on a time sequence attenuation characteristic of information propagation so as to form comprehensive understanding of the service network resource demand; and the network resource demand of the service in a future period of time is predicted based on this. Through the mode, the dynamic characteristic of the service demand changing along with time can be more accurately captured, the accuracy of network resource demand prediction is improved, and a more accurate decision basis is provided for resource allocation, so that the resource utilization rate is improved, and the operation cost is reduced.

Description

Technical Field

[0001] The present application relates to the field of intelligent prediction, and more specifically, to a 5G power multi-service slice resource allocation method. Background Art

[0002] With the rapid development of the fifth generation mobile communication technology (5G), its application in various industries is becoming more and more extensive, especially in the power industry. 5G technology, with its high speed, large capacity and low latency, provides strong technical support for the construction of smart grids. In the power system, different business scenarios (such as remote meter reading, video monitoring, fault warning and rapid response, etc.) have significant differences in the demand for network resources, and these demands often change dynamically over time. For example, fault detection and location require extremely low latency, while daily data collection may have higher requirements for bandwidth. Therefore, how to reasonably and effectively allocate 5G network resources to meet the needs of diversified services in the power system has become a key issue that needs to be solved urgently.

[0003] In the prior art, the invention patent with publication number CN117998594A proposes a 5G power multi-service slice resource allocation method. For the same type of service, it uses the historical data of this type of service, adopts time series analysis, linear regression model and other algorithms to analyze the historical data and predict the resource demand of this type of service in the future.

[0004] However, although the above scheme can improve the efficiency of resource allocation to a certain extent, it mainly focuses on trend prediction at the macro level and fails to fully consider the volatility and uncertainty of business demand on a fine-grained time scale, which may lead to inaccurate and inflexible forecasting of business resource demand.

[0005] Therefore, an optimized 5G power multi-service slice resource allocation method is expected. Summary of the invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a 5G power multi-service slice resource allocation method, which extracts the historical network resource demand information of the service from the historical data, and uses the data processing technology based on deep learning to extract the time series features of the historical network resource demand information based on the local time scale, so as to capture the fine-grained time series change trend of the network resource demand of the service, and then, based on the time series attenuation characteristics of information propagation, the network resource demand characteristics on each local time scale are dynamically propagated and aggregated to form a comprehensive understanding of the network resource demand of the service, and based on this, the network resource demand of the service in the future period of time is predicted. In this way, the dynamic characteristics of the business demand changing over time can be captured more accurately, the accuracy of the network resource demand prediction can be improved, and a more accurate decision-making basis can be provided for resource allocation, thereby helping to improve resource utilization and reduce operating costs.

[0007] According to one aspect of the present application, a 5G power multi-service slice resource allocation method is provided, which includes: obtaining historical data and current service requirements of the communication network of the power grid; predicting the network resource requirements of each service based on the historical data and current service requirements; obtaining an initial network resource allocation strategy based on the network resource requirements of each service, and forming a resource mapping table; receiving resource usage information sent by a local controller in real time; optimizing the initial network resource allocation strategy based on the resource usage information, and updating the resource mapping table; based on the updated resource mapping table, sending the optimized network resource allocation strategy to the local controller, and allocating resources of the communication network of the power grid, characterized in that predicting the network resource requirements of each service based on the historical data and current service requirements includes:

[0008] Extracting a time queue of historical network resource requirements of the first service from the historical data;

[0009] Performing local sampling timing analysis on the time queue of the historical network resource demand of the first service to obtain a queue of implicit feature vectors of the sampling window granularity of the historical network resource demand of the first service;

[0010] Performing dynamic propagation aggregation on the queue of the implicit feature vector of the sampling window granularity of the first service history network resource demand to obtain a time series propagation aggregation representation vector of the first service history network resource demand;

[0011] Based on the time series propagation aggregation representation vector of the historical network resource demand of the first service, the network resource demand of the first service is predicted.

[0012] Compared with the prior art, the present application provides a 5G power multi-service slice resource allocation method, which extracts the historical network resource demand information of the service from the historical data, and uses the data processing technology based on deep learning to extract the time series features of the historical network resource demand information based on the local time scale, so as to capture the fine-grained time series change trend of the network resource demand of the service, and then, based on the time series attenuation characteristics of information propagation, the network resource demand characteristics on each local time scale are dynamically propagated and aggregated to form a comprehensive understanding of the network resource demand of the service, and based on this, the network resource demand of the service in the future period of time is predicted. In this way, the dynamic characteristics of the business demand changing over time can be captured more accurately, the accuracy of the network resource demand prediction can be improved, and a more accurate decision-making basis can be provided for resource allocation, which helps to improve resource utilization and reduce operating costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0014] Figure 1 It is a flowchart of a 5G power multi-service slice resource allocation method according to an embodiment of the present application;

[0015] Figure 2 A data flow diagram of a 5G power multi-service slice resource allocation method according to an embodiment of the present application;

[0016] Figure 3 This is a flowchart of sub-step S2 of the 5G power multi-service slice resource allocation method according to an embodiment of the present application;

[0017] Figure 4 This is a flowchart of sub-step S23 of the 5G power multi-service slice resource allocation method according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0019] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0020] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0021] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0022] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0023] Patent CN117998594A proposes a 5G power multi-service slice resource allocation method, which includes: obtaining historical data and current business needs of the power grid's communication network; based on the historical data and current business needs, predicting the network resource needs of each business; based on the network resource needs of each business, obtaining an initial network resource allocation strategy and forming a resource mapping table; receiving resource usage information sent by a local controller in real time; based on the resource usage information, optimizing the initial network resource allocation strategy and updating the resource mapping table; based on the updated resource mapping table, sending the optimized network resource allocation strategy to the local controller to allocate resources for the power grid's communication network.

[0024] However, although the above scheme can improve the efficiency of resource allocation to a certain extent, it mainly focuses on trend prediction at the macro level and fails to fully consider the volatility and uncertainty of business demand on a fine-grained time scale, which may lead to inaccurate and inflexible forecasting of business resource demand.

[0025] Based on this, in the technical solution of this application, a 5G power multi-service slice resource allocation method is proposed. Figure 1This is a flowchart of a 5G power multi-service slice resource allocation method according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the 5G power multi-service slice resource allocation method according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to the 5G power multi-service slice resource allocation method of the embodiment of the present application, the steps include: S1, obtaining the historical data and current business needs of the communication network of the power grid; S2, predicting the network resource needs of each business based on the historical data and current business needs; S3, based on the network resource needs of each business, obtaining the initial network resource allocation strategy and forming a resource mapping table; S4, receiving the resource usage information sent by the local controller in real time; S5, based on the resource usage information, optimizing the initial network resource allocation strategy and updating the resource mapping table; S6, based on the updated resource mapping table, sending the optimized network resource allocation strategy to the local controller to allocate resources of the communication network of the power grid.

[0026] In particular, the S1 obtains historical data and current business requirements of the communication network of the power grid.

[0027] It can be understood that by analyzing the historical data of the power grid communication network, the changing trend of network resource demand in different time periods can be understood. Information on current business demand can help identify the characteristics of different businesses. Combining historical data with current business demand can more accurately predict the network resource demand of each business in the future.

[0028] In particular, the S2 predicts the network resource requirements of each service based on the historical data and current service requirements. Figure 3 As shown, S2 includes: S21, extracting the time queue of the historical network resource demand of the first service from the historical data; S22, performing local sampling timing analysis on the time queue of the historical network resource demand of the first service to obtain a queue of implicit feature vectors at the sampling window granularity of the historical network resource demand of the first service; S23, performing dynamic propagation aggregation on the queue of implicit feature vectors at the sampling window granularity of the historical network resource demand of the first service to obtain a time series propagation aggregation representation vector of the historical network resource demand of the first service; S24, predicting the network resource demand of the first service based on the time series propagation aggregation representation vector of the historical network resource demand of the first service.

[0029] Specifically, the S21 extracts the time queue of the historical network resource demand of the first business from the historical data. It should be understood that the power system includes multiple different business scenarios, such as power monitoring, load forecasting, power trading, etc., and each business scenario has its specific network resource demand characteristics. In the technical solution of the present application, independent network resource demand analysis and prediction are performed for each business scenario, and the resource demand pattern and trend of the business are identified by extracting the time queue data of its historical network resource demand, thereby achieving accurate prediction of its future network resource demand.

[0030] Specifically, the S22 performs local sampling timing analysis on the time queue of the historical network resource demand of the first service to obtain a queue of implicit feature vectors of the sampling window granularity of the historical network resource demand of the first service. Specifically, first, a predetermined sampling window is used to perform sliding window sampling on the time queue of the historical network resource demand of the first service to obtain a queue of the sampling window of the historical network resource demand of the first service. Considering that the network resource demand of different service scenarios usually has different time characteristics, for example, for the remote meter reading service, its network resource demand may reach a peak at certain specific time periods of the day, and is relatively low in other time periods. For the video surveillance service, its demand may be more evenly distributed, but the bandwidth requirement is higher. Therefore, in order to avoid the loss of detailed information caused by the timing analysis of long-term span data, in the technical solution of the present application, a predetermined sampling window is used to perform sliding window sampling on the time queue of the historical network resource demand of the first service to refine the granularity of data analysis and obtain a queue of the sampling window of the historical network resource demand of the first service, thereby helping to more carefully capture the details of the changes in service demand over time and provide more accurate data support for subsequent network resource demand prediction. Next, each sampling window of the first business historical network resource demand in the queue of the sampling window of the first business historical network resource demand is input into the network resource demand sequence encoder based on the bidirectional LSTM model to obtain the queue of the implicit feature vector of the sampling window granularity of the first business historical network resource demand. Here, in order to fully extract the network resource demand characteristics in each sampling window, the present application further adopts a bidirectional LSTM model to construct a historical network resource demand sequence encoder, and performs feature extraction on each sampling window of the first business historical network resource demand in the queue of the sampling window of the first business historical network resource demand. A person of ordinary skill in the art should know that the bidirectional LSTM model can simultaneously process the forward and backward information of the sequence data through the LSTM layer in two directions, and more fully capture the long-term dependencies in the time series data, so as to enhance the deep understanding of the changes in network resource demand, and effectively extract the time series change pattern of the network resource demand from each sampling window, including trends, periodicity, and abnormal points, so as to obtain the queue of the implicit feature vector of the sampling window granularity of the first business historical network resource demand.

[0031] Specifically, the S23 performs dynamic propagation aggregation on the queue of implicit feature vectors at the granularity of the sampling window of the first business history network resource demand to obtain the time series propagation aggregation representation vector of the first business history network resource demand. In the technical solution of the present application, in order to understand and integrate the network resource demand characteristics of each sampling window from a global perspective, the present application further performs feature aggregation on the queue of implicit feature vectors at the granularity of the sampling window of the first business history network resource demand. In particular, considering that in the process of predicting future network resource demand, as time goes by, early network resource demand information may gradually lose its relevance and accuracy. Therefore, in order to avoid excessive accumulation of early information effects, the present application proposes a dynamic aggregation method based on the characteristic of feature propagation time series attenuation, which simulates the propagation process of information in time so that the recent network resource demand characteristics have a higher weight, while the early characteristics gradually weaken, thereby realizing dynamic tracking of the changing trend of business network resource demand. More specifically, in a specific example of the present application, such as Figure 4 As shown, the S23 includes: S231, performing time node division and principal component analysis on the queue of the implicit feature vector of the sampling window granularity of the first business historical network resource demand to obtain the current first business historical network resource demand principal component representation vector and the historical first business historical network resource demand principal component representation vector queue; S232, based on the time span of each historical first business historical network resource demand principal component representation vector in the queue of the historical first business historical network resource demand principal component representation vector relative to the current first business historical network resource demand principal component representation vector, performing propagation aggregation based on feature time series gradient attenuation on the queue of the implicit feature vector of the sampling window granularity of the first business historical network resource demand to obtain the first business historical network resource demand timing propagation aggregation representation vector.

[0032] More specifically, the S231 performs time node division and principal component analysis on the queue of the first service history network resource demand sampling window granularity implicit feature vector to obtain the current first service history network resource demand principal component representation vector and the historical first service history network resource demand principal component representation vector queue. Specifically, first, extract the current first service history network resource demand sampling window granularity implicit feature vector from the queue of the first service history network resource demand sampling window granularity implicit feature vector, and perform principal component extraction based on eigenvalues ​​on the current first service history network resource demand sampling window granularity implicit feature vector to obtain the current first service history network resource demand principal component representation vector; and define other first service history network resource demand sampling window granularity implicit feature vectors in the queue of the first service history network resource demand sampling window granularity implicit feature vectors as historical first service history network resource demand sampling window granularity implicit feature vectors, and perform principal component extraction based on eigenvalues ​​on each historical first service history network resource demand sampling window granularity implicit feature vector to obtain the queue of the historical first service history network resource demand principal component representation vector. That is, the last sampling window in the queue of the implicit feature vector of the sampling window granularity of the first business historical network resource demand is regarded as the current time node, and the other sampling windows are regarded as historical time nodes. PCA analysis is performed on the implicit feature vector of the sampling window granularity of the first business historical network resource demand at each time node to remove noise interference in the data, highlight key information, and extract the most representative principal components.

[0033] Among them, the specific process of performing eigenvalue-based principal component extraction on the implicit eigenvector of the sampling window granularity of the current first business historical network resource demand includes: calculating the covariance matrix of the implicit eigenvector of the sampling window granularity of the current first business historical network resource demand to obtain the covariance matrix of the current first business historical network resource demand; performing eigenvalue decomposition on the covariance matrix of the current first business historical network resource demand to obtain a set of principal component eigenvalues ​​of the current first business historical network resource demand and a set of corresponding principal component eigenvectors of the current first business historical network resource demand; selecting the principal component eigenvector of the current first business historical network resource demand corresponding to the maximum eigenvalue in the set of principal component eigenvalues ​​of the current first business historical network resource demand as the principal component representation vector of the current first business historical network resource demand.

[0034] More specifically, in S232, based on the time span of each historical first business historical network resource demand principal component representation vector in the queue of the historical first business historical network resource demand principal component representation vector relative to the current first business historical network resource demand principal component representation vector, the queue of the first business historical network resource demand sampling window granularity implicit feature vector is subjected to propagation aggregation based on feature time series gradient attenuation to obtain the first business historical network resource demand time series propagation aggregation representation vector. Specifically, firstly, the time series propagation absolute attenuation factor of each historical first business historical network resource demand principal component representation vector in the queue of the historical first business historical network resource demand principal component representation vector relative to the current first business historical network resource demand principal component representation vector is calculated to obtain the queue of the historical first business historical network resource demand characteristic time series propagation absolute attenuation factor; that is, in order to capture the trend of the network resource demand characteristics of each historical time node evolving over time and the information loss, the corresponding time series propagation absolute attenuation factor is further calculated based on the principal component change rate of the network resource demand characteristics of each historical time node relative to the current time node, so as to enhance the model's ability to understand information transmission over long time intervals. Next, based on the time span between the principal component representation vectors of each historical first-service historical network resource demand and the principal component representation vector of the current first-service historical network resource demand, the queue of the absolute attenuation factor of the characteristic time series propagation of the historical first-service historical network resource demand is modulated by time span to obtain a queue of the time span modulated propagation attenuation factor. The above-mentioned time series propagation absolute attenuation factor is adjusted by introducing the time span factor to simulate the gradual forgetting process of data in the time dimension, so that the model can pay more attention to the most recent user network resource demand characteristics and reduce excessive dependence on early data. Furthermore, based on the queue of the time span modulated propagation attenuation factor, the queue of the implicit feature vector of the sampling window granularity of the first-service historical network resource demand is subjected to feature time series transmission aggregation to obtain the first-service historical network resource demand time series propagation aggregation representation vector. Here, based on the gating mechanism, the information of the absolute attenuation factor of the time series propagation after time span modulation is screened, the most influential propagation path is identified and the corresponding weight is assigned accordingly to improve the distinctiveness of the features, and the main component features of the historical network resource demand of the first business at each time node are weighted and aggregated based on the generated weights. In this way, while capturing the long-term change trend of the historical network resource demand of the first business, the short-term dynamic pattern of the network resource demand is effectively emphasized, and the time series propagation aggregation representation vector of the historical network resource demand of the first business is generated, thereby achieving a comprehensive understanding of the changes in the business network resource demand in the historical period.

[0035] Among them, the specific process of calculating the absolute attenuation factor of the time series propagation of each historical first business historical network resource demand principal component representation vector in the queue of the historical first business historical network resource demand principal component representation vector relative to the current first business historical network resource demand principal component representation vector includes: calculating the dot division vector between the historical first business historical network resource demand principal component representation vector and the current first business historical network resource demand principal component representation vector, and calculating the base-two logarithmic value of the absolute value of each eigenvalue in the dot division vector to obtain the absolute standard representation vector of time series propagation attenuation; calculating the dot product vector between the absolute standard representation vector of time series propagation attenuation and the historical first business historical network resource demand principal component representation vector, and calculating the exponential function value with base e and the sum of the eigenvalues ​​of the dot product vector as the exponent to obtain the absolute attenuation factor of the time series propagation of the historical first business historical network resource demand characteristics.

[0036] More specifically, the specific process of time span modulating the queue of the absolute attenuation factor of the time series propagation of the historical first business historical network resource demand characteristics based on the time span between each historical first business historical network resource demand principal component representation vector in the queue of the historical first business historical network resource demand principal component representation vector and the current first business historical network resource demand principal component representation vector includes: calculating the timestamp difference between the historical first business historical network resource demand principal component representation vector and the current first business historical network resource demand principal component representation vector, rounding down the timestamp difference to obtain the time span factor; dividing the time series propagation absolute attenuation factor of the historical first business historical network resource demand principal component representation vector by the natural exponential function value of the time span factor to obtain the time span modulated propagation attenuation factor.

[0037] More specifically, based on the queue of the time span modulated propagation attenuation factor, the specific process of performing feature timing transmission aggregation on the queue of the implicit feature vector of the sampling window granularity of the first business historical network resource demand includes: inputting the queue of the time span modulated propagation attenuation factor into the information transmission screening module based on the gating function to obtain the queue of the historical first business historical network resource demand feature timing gradient attenuation weight; based on the queue of the historical first business historical network resource demand feature timing gradient attenuation weight, calculating the weighted sum of the queue of the historical first business historical network resource demand principal component representation vector to obtain the historical first business historical network resource demand principal component significant transmission aggregation representation vector; calculating the positional sum of the historical first business historical network resource demand principal component significant transmission aggregation representation vector and the current first business historical network resource demand principal component representation vector to obtain the first business historical network resource demand timing propagation aggregation representation vector. Among them, the specific steps of inputting the queue of time span modulated propagation attenuation factors into the information transmission screening module based on the gating function include: based on a preset threshold, masking the queue of time span modulated propagation attenuation factors to obtain a queue of masked time span modulated propagation attenuation factors; inputting the queue of masked time span modulated propagation attenuation factors into the Sigmoid function for normalization processing to obtain a queue of the timing gradient attenuation weights of the historical first service historical network resource demand characteristics.

[0038] In summary, the queue of the implicit feature vector of the sampling window granularity of the first business history network resource demand is dynamically propagated and aggregated to obtain the time series propagation aggregation representation vector of the first business history network resource demand, including: processing the queue of the implicit feature vector of the sampling window granularity of the first business history network resource demand with the following time series dynamic propagation aggregation formula to obtain the time series propagation aggregation representation vector of the first business history network resource demand, wherein the time series dynamic propagation aggregation formula is:

[0039] X={x1,x2,...,x i ,...,x p}

[0040] h i = PCA(x i )

[0041]

[0042] Wherein, X represents the queue of the implicit feature vector of the sampling window granularity of the first service history network resource demand, p is the number of vectors in the queue of the implicit feature vector of the sampling window granularity of the first service history network resource demand, x p represents the implicit feature vector of the sampling window granularity of the current first service historical network resource demand, x1, x2 and xi The first, second and i-th historical first service historical network resource demand sampling window granularity implicit feature vectors in the queue respectively represent the historical first service historical network resource demand sampling window granularity implicit feature vectors, PCA(·) represents the principal component analysis function, h i represents the principal component representation vector of the historical network resource demand of the i-th historical first service, h p represents the principal component representation vector of the current historical network resource demand of the first service, (·) T represents the transpose of the vector, n represents the length of the implicit feature vector of the sampling window granularity of the historical first service historical network resource demand, C i The covariance matrix of the implicit feature vector of the sampling window granularity of the historical network resource demand of the i-th historical first service, that is, the covariance matrix of the historical network resource demand of the historical first service, U i represents the matrix composed of the set of principal component eigenvectors of the historical first service historical network resource demand obtained by performing eigenvalue decomposition on the covariance matrix of the historical first service historical network resource demand, Λ i represents a diagonal matrix consisting of a set of principal component eigenvalues ​​of the historical first service historical network resource demand obtained by performing eigenvalue decomposition on the covariance matrix of the historical first service historical network resource demand of the i-th historical first service, diag is a diag function used to construct a diagonal matrix, λ i1 , i2 , ik and λ im represents the first, second, kth and mth historical first service historical network resource demand principal component eigenvalues ​​in the set of the i-th historical first service historical network resource demand principal component eigenvalues, m is the number of eigenvalues ​​in the set of the historical first service historical network resource demand principal component eigenvalues, u i1 、u i2 、u iz and u im represents the first, second, zth and mth historical first service historical network resource demand principal component eigenvectors in the set of the i-th historical first service historical network resource demand principal component eigenvectors, argmax(·) represents the index corresponding to the maximum value, z represents the index of the maximum eigenvalue in the set of the historical first service historical network resource demand principal component eigenvalues, h ij represents the jth eigenvalue of the principal component representation vector of the historical network resource demand of the i-th historical first service, h pjrepresents the j-th eigenvalue of the principal component representation vector of the current first service history network resource demand, |·| represents the absolute value, log represents the logarithmic function with base 2, l represents the length of the principal component representation vector of the current first service history network resource demand, exp(·) represents the exponential function with base e, d i represents the absolute attenuation factor of the time series propagation of the historical network resource demand characteristics of the i-th historical first service, t p and t i respectively represent the timestamps of the current first service history network resource demand sampling window granularity implicit feature vector and the i-th history first service history network resource demand sampling window granularity implicit feature vector, Indicates rounding down, e i represents the modulation propagation attenuation factor of the i-th time span, mask(·) represents mask processing, sigmoid represents the sigmoid function, θ represents the preset threshold, and w i represents the temporal gradient attenuation weight of the historical network resource demand feature of the i-th historical first service, and V represents the temporal propagation aggregation representation vector of the historical network resource demand of the first service.

[0043] Specifically, the S24 predicts the network resource demand of the first business based on the time series propagation aggregation representation vector of the historical network resource demand of the first business. In a specific example of the present application, the time series propagation aggregation representation vector of the historical network resource demand of the first business is input into a decoder-based network resource demand prediction module to obtain the network resource demand of the first business. Specifically, the historical network resource demand prediction module adopts a sequence-to-sequence (Seq2Seq) model architecture, in which the decoder part is composed of a unidirectional LSTM network, which learns the time dependency in the time series propagation aggregation representation vector of the historical network resource demand of the first business to capture the evolution law of business demand over time, thereby gradually predicting the network resource demand at each time point in the future, and generating a prediction result of the network resource demand of the first business. In this way, an effective basis can be provided for the resource scheduling and optimization of the power system, thereby improving the operating efficiency and economic benefits of the power system.

[0044] Preferably, inputting the time series propagation aggregation representation vector of the first service history network resource demand into a decoder-based network resource demand prediction module to obtain the network resource demand of the first service includes:

[0045] Based on the association value and distance value of the eigenvalue of the first service history network resource demand timing propagation aggregation representation vector, the first service history network resource demand timing propagation aggregation association response matrix and the first service history network resource demand timing propagation aggregation distance response matrix are determined, that is:

[0046] M1(i,j)=v i ×v j

[0047]

[0048] where v i ,v j ∈V∈R 1×L , M1 is the first service history network resource demand time series propagation aggregation association response matrix, M2 is the first service history network resource demand time series propagation aggregation distance response matrix;

[0049] Performing matrix multiplication of the first service history network resource demand time series propagation aggregation representation vector and the first service history network resource demand time series propagation aggregation association response matrix to obtain a first service history network resource demand time series propagation aggregation association response vector;

[0050] Performing matrix multiplication of the first service history network resource demand timing propagation aggregation distance response matrix and the transposed vector of the first service history network resource demand timing propagation aggregation representation vector to obtain the first service history network resource demand timing propagation aggregation distance response vector;

[0051] After performing a dot multiplication on the first service history network resource demand timing propagation aggregation association response matrix and the first service history network resource demand timing propagation aggregation distance response matrix, perform a matrix multiplication on the transposed vector of the first service history network resource demand timing propagation aggregation representation vector to obtain a first service history network resource demand timing propagation aggregation logic bias vector;

[0052] The first service history network resource demand timing propagation aggregation association response vector, the first service history network resource demand timing propagation aggregation distance response vector and the first service history network resource demand timing propagation aggregation logic offset vector are added to obtain an optimized first service history network resource demand timing propagation aggregation representation vector; and

[0053] The optimized first service historical network resource demand time series propagation aggregation representation vector is input into a decoder-based network resource demand prediction module to obtain the network resource demand of the first service.

[0054] Here, the optimized first service historical network resource demand time series propagation aggregation representation vector V' is expressed as:

[0055]

[0056] In V1∈R 1×L , and V2∈R L×1In the case of:

[0057]

[0058] Wherein, V is the time series propagation aggregation representation vector of the historical network resource demand of the first service, V T is the transposed vector of the first service history network resource demand timing propagation aggregation representation vector, V1 is the response vector obtained by adding the first service history network resource demand timing propagation aggregation association response vector and the first service history network resource demand timing propagation aggregation distance response vector by position, V2 is the first service history network resource demand timing propagation aggregation logic offset vector, is added by position, is matrix multiplication, ⊙ is position point multiplication, and V′ is the optimized first service historical network resource demand time series propagation aggregation representation vector.

[0059] That is, when each first business historical network resource demand sampling window granularity implicit feature vector in the queue of the first business historical network resource demand sampling window granularity implicit feature vector respectively represents the first business historical network resource demand sampling window granularity local temporal correlation feature, when performing feature temporal dynamic propagation based on the node feature principal component gradient attenuation, the local temporal semantic feature gap of each first business historical network resource demand sampling window granularity will also cause the node feature principal component and gradient attenuation to become sparse, resulting in large-scale simple repetition in the feature temporal dynamic propagation, affecting the semantic logical dependency of the feature temporal dynamic propagation, and reducing the accuracy of the decoding results based on the semantic temporal distribution.

[0060] Therefore, by using the self-correlation matrix and self-distance matrix of the first business history network resource demand time-series propagation aggregation representation vector as the statistical reference-free distribution response framework of the first business history network resource demand time-series propagation aggregation representation vector, based on the retrieval-enhanced reverse response construction for the first business history network resource demand time-series propagation aggregation representation vector, simple repetition of feature distribution is avoided, and by ensuring the intrinsic mapping logic of the first business history network resource demand time-series propagation aggregation representation vector based on the retrieval-response context, the surface combinatorial mapping of the first business history network resource demand time-series propagation aggregation representation vector is avoided, so as to achieve the logical dependency mapping of the first business history network resource demand time-series propagation aggregation representation vector to the decoding target domain while maintaining the intuitive response relevance, thereby improving the accuracy of the network resource demand of the first business obtained by the network resource demand prediction module based on the decoder input of the first business history network resource demand time-series propagation aggregation representation vector.

[0061] In particular, the S3 obtains an initial network resource allocation strategy based on the network resource requirements of each business and forms a resource mapping table. It should be understood that the resource mapping table can be used to determine how to most effectively allocate network resources based on the specific requirements of each business. The resource mapping table is a mechanism that matches network resources with specific business requirements. It defines which resources should be allocated to which business, as well as the amount and priority of the allocation. In this way, resource allocation can be performed quickly and accurately according to this table in actual operation. In one example, when the system receives new resource usage information, it can optimize resource allocation by adjusting the allocation strategy in the mapping table to better adapt to changing business needs.

[0062] In particular, S4 receives resource usage information sent by the local controller in real time. By receiving resource usage information in real time, the actual operating status and resource consumption of each node in the power grid communication network can be timely understood. Based on the real-time resource usage information, the system can more accurately determine whether the current resource allocation is reasonable.

[0063] In particular, the S5 optimizes the initial network resource allocation strategy based on the resource usage information and updates the resource mapping table. The initial network resource allocation strategy is based on historical data and current business demand forecasts, and the actual situation may change. Real-time resource usage information reflects the actual consumption of current network resources. By analyzing this information, it is possible to identify situations of excess or shortage of resources, thereby optimizing resource allocation and improving overall resource utilization. Therefore, in the technical solution of the present application, the resource allocation strategy can be dynamically adjusted through real-time resource usage information to make it more in line with actual needs.

[0064] In particular, the S6, based on the updated resource mapping table, sends the optimized network resource allocation strategy to the local controller to allocate the resources of the communication network of the power grid. It should be understood that the optimized resource allocation strategy has been adjusted according to the latest resource usage information and can more accurately reflect the actual needs of the current network. By sending the optimized strategy to the local controller, dynamic management and flexible scheduling of resources can be achieved to ensure that network resources are always in the optimal configuration state.

[0065] In summary, the 5G power multi-service slice resource allocation method according to the embodiment of the present application is explained, which extracts the historical network resource demand information of the service from the historical data, and uses the data processing technology based on deep learning to extract the time series features of the historical network resource demand information based on the local time scale, so as to capture the fine-grained time series change trend of the network resource demand of the service, and then, based on the time series attenuation characteristics of information propagation, the network resource demand characteristics on each local time scale are dynamically propagated and aggregated to form a comprehensive understanding of the network resource demand of the service, and based on this, the network resource demand of the service in the future period of time is predicted. In this way, the dynamic characteristics of the business demand changing over time can be captured more accurately, the accuracy of the network resource demand prediction can be improved, and a more accurate decision-making basis can be provided for resource allocation, which helps to improve resource utilization and reduce operating costs.

[0066] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A 5G power multi-service slice resource allocation method, comprising: Obtain historical data and current business needs of the power grid’s communication network; Based on the historical data and current business needs, predict the network resource requirements of each business; Based on the network resource requirements of each service, an initial network resource allocation strategy is obtained to form a resource mapping table; resource usage information sent by a local controller is received in real time; based on the resource usage information, the initial network resource allocation strategy is optimized and the resource mapping table is updated; based on the updated resource mapping table, the optimized network resource allocation strategy is sent to the local controller to allocate resources of the communication network of the power grid, characterized in that the network resource requirements of each service are predicted based on the historical data and the current service requirements, including: Extracting a time queue of historical network resource requirements of the first service from the historical data; Performing local sampling timing analysis on the time queue of the historical network resource demand of the first service to obtain a queue of implicit feature vectors of the sampling window granularity of the historical network resource demand of the first service; Performing dynamic propagation aggregation on the queue of the implicit feature vector of the sampling window granularity of the first service history network resource demand to obtain a time series propagation aggregation representation vector of the first service history network resource demand; Based on the time series propagation aggregation representation vector of the historical network resource demand of the first service, the network resource demand of the first service is predicted. Among them, the queue of implicit feature vectors at the granularity of the sampling window of the first business historical network resource demand is dynamically propagated and aggregated, including: performing time node division and principal component analysis on the queue of implicit feature vectors at the granularity of the sampling window of the first business historical network resource demand to obtain a queue of principal component representation vectors of the current first business historical network resource demand and principal component representation vectors of the historical first business historical network resource demand; based on the time span of each principal component representation vector of the historical first business historical network resource demand in the queue of the principal component representation vector of the historical first business historical network resource demand relative to the principal component representation vector of the current first business historical network resource demand, the queue of implicit feature vectors at the granularity of the first business historical network resource demand is propagated and aggregated based on characteristic time series gradient attenuation to obtain the time series propagation aggregation representation vector of the first business historical network resource demand.

2. The 5G power multi-service slice resource allocation method according to claim 1 is characterized in that: Performing local sampling timing analysis on the time queue of the historical network resource demand of the first service to obtain a queue of implicit feature vectors of the sampling window granularity of the historical network resource demand of the first service, including: Using a predetermined sampling window, sliding window sampling is performed on the time queue of the historical network resource demand of the first service to obtain a queue of the sampling window of the historical network resource demand of the first service; Each sampling window of the first business historical network resource demand in the queue of the sampling window of the first business historical network resource demand is input into the network resource demand sequence encoder based on the bidirectional LSTM model to obtain a queue of the first business historical network resource demand sampling window granularity implicit feature vector.

3. The 5G power multi-service slice resource allocation method according to claim 2 is characterized in that: The queue of the implicit feature vector of the sampling window granularity of the first service history network resource demand is divided into time nodes and principal component analysis is performed to obtain a queue of the principal component representation vector of the current first service history network resource demand and the principal component representation vector of the historical first service history network resource demand, including: Extracting the current first service history network resource demand sampling window granularity implicit feature vector from the queue of the first service history network resource demand sampling window granularity implicit feature vector, and performing principal component extraction based on the eigenvalue on the current first service history network resource demand sampling window granularity implicit feature vector to obtain the current first service history network resource demand principal component representation vector; The other first business historical network resource demand sampling window granularity implicit feature vectors in the queue of the first business historical network resource demand sampling window granularity implicit feature vectors are defined as historical first business historical network resource demand sampling window granularity implicit feature vectors, and principal component extraction based on the eigenvalue is performed on each historical first business historical network resource demand sampling window granularity implicit feature vector to obtain the queue of the historical first business historical network resource demand principal component representation vectors.

4. The 5G power multi-service slice resource allocation method according to claim 3 is characterized in that: The method of performing principal component extraction based on the eigenvalue on the sampling window granularity implicit eigenvector of the current first service history network resource demand to obtain the principal component representation vector of the current first service history network resource demand includes: Calculating the covariance matrix of the current first service historical network resource demand sampling window granularity implicit feature vector to obtain the current first service historical network resource demand covariance matrix; Performing eigenvalue decomposition on the covariance matrix of the current first service historical network resource demand to obtain a set of principal component eigenvalues ​​of the current first service historical network resource demand and a set of corresponding principal component eigenvectors of the current first service historical network resource demand; The current first service historical network resource demand principal component eigenvalue vector corresponding to the maximum eigenvalue in the set of the current first service historical network resource demand principal component eigenvalues ​​is selected as the current first service historical network resource demand principal component representation vector.

5. The 5G power multi-service slice resource allocation method according to claim 4 is characterized in that: Based on the time span of each historical first service historical network resource demand principal component representation vector in the queue of the historical first service historical network resource demand principal component representation vector relative to the current first service historical network resource demand principal component representation vector, the queue of the first service historical network resource demand sampling window granularity implicit feature vector is subjected to propagation aggregation based on feature time series gradient attenuation to obtain the first service historical network resource demand time series propagation aggregation representation vector, including: Calculate the time series propagation absolute attenuation factor of each historical first service historical network resource demand principal component representation vector in the queue of the historical first service historical network resource demand principal component representation vector relative to the current first service historical network resource demand principal component representation vector to obtain a queue of the time series propagation absolute attenuation factor of the historical first service historical network resource demand feature; Based on the time span between the principal component representation vectors of each historical first service historical network resource demand and the principal component representation vector of the current first service historical network resource demand, the queue of the historical first service historical network resource demand characteristic time series propagation absolute attenuation factor is modulated by time span to obtain a queue of the time span modulated propagation attenuation factor; Based on the queue of the time span modulated propagation attenuation factor, the queue of the implicit feature vector of the sampling window granularity of the first business history network resource demand is aggregated with feature timing transmission to obtain the first business history network resource demand timing propagation aggregation representation vector.

6. The 5G power multi-service slice resource allocation method according to claim 5 is characterized in that: Calculating the time series propagation absolute attenuation factor of each historical first service historical network resource demand principal component representation vector in the queue of the historical first service historical network resource demand principal component representation vector relative to the current first service historical network resource demand principal component representation vector to obtain a queue of the historical first service historical network resource demand characteristic time series propagation absolute attenuation factor, including: Calculate the dotted division vector between the historical first service historical network resource demand principal component representation vector and the current first service historical network resource demand principal component representation vector, and calculate the base-two logarithm of the absolute value of each eigenvalue in the dotted division vector to obtain the absolute standard representation vector of time series propagation attenuation; Calculate the dot product vector between the absolute standard representation vector of the time series propagation attenuation and the principal component representation vector of the historical first business historical network resource demand, and calculate the exponential function value with e as the base and the sum of the eigenvalues ​​of the dot product vector as the exponent to obtain the absolute attenuation factor of the characteristic time series propagation of the historical first business historical network resource demand.

7. The 5G power multi-service slice resource allocation method according to claim 6 is characterized in that: Based on the time span between each historical first service historical network resource demand principal component representation vector in the queue of the historical first service historical network resource demand principal component representation vector and the current first service historical network resource demand principal component representation vector, the queue of the historical first service historical network resource demand characteristic time series propagation absolute attenuation factor is modulated by time span to obtain a queue of time span modulated propagation attenuation factors, including: Calculate the timestamp difference between the historical first service historical network resource demand principal component representation vector and the current first service historical network resource demand principal component representation vector, and round down the timestamp difference to obtain a time span factor; The time span modulation propagation attenuation factor is obtained by dividing the time series propagation absolute attenuation factor of the principal component representation vector of the historical first service historical network resource demand by the natural exponential function value of the time span factor.

8. The 5G power multi-service slice resource allocation method according to claim 7 is characterized in that: Based on the queue of the time span modulated propagation attenuation factor, the queue of the implicit feature vector of the sampling window granularity of the first service history network resource demand is subjected to feature timing transmission aggregation to obtain the first service history network resource demand timing propagation aggregation representation vector, including: Inputting the queue of the time span modulated propagation attenuation factor into the information transmission screening module based on the gating function to obtain the queue of the time series gradient attenuation weight of the historical first service historical network resource demand feature; Based on the queue of the time series gradient attenuation weight of the historical first service historical network resource demand feature, calculate the weighted sum of the queue of the historical first service historical network resource demand principal component representation vector to obtain the historical first service historical network resource demand principal component significant transfer aggregation representation vector; The position-wise sum of the historical first business historical network resource demand principal component significant transfer aggregation representation vector and the current first business historical network resource demand principal component representation vector is calculated to obtain the first business historical network resource demand time series propagation aggregation representation vector.

9. The 5G power multi-service slice resource allocation method according to claim 8 is characterized in that: Predicting the network resource demand of the first service based on the time series propagation aggregation representation vector of the historical network resource demand of the first service includes: The time series propagation aggregation representation vector of the historical network resource demand of the first service is input into a decoder-based network resource demand prediction module to obtain the network resource demand of the first service.

Images