Power distribution network communication network resource multi-time scale reservation method, system and device and medium

Abstract

The invention discloses a power distribution network communication network resource multi-time scale reservation method, system, device and medium, and belongs to the technical field of power communication scheduling, and the method comprises the steps: building a time sequence prediction model used for peak data transmission pressure prediction based on data, and predicting a peak data transmission pressure value sequence in a plurality of time steps in the future; in combination with the difference between conventional service data and peak service data in the power distribution network, the task age is adopted as a quantitative index, a resource reservation optimization model with task timeliness loss minimization as a target is constructed, and a basic resource reservation scheme is determined; and taking the basic resource reservation scheme as an initial configuration basis of the policy network, respectively generating a communication resource regulation and control parameter and a computing resource regulation and control parameter in a short time period, and updating a resource allocation policy. According to the invention, the timeliness of peak data processing and the adaptivity of resource configuration are improved, and the stability and transmission efficiency of the system under a high dynamic load are effectively enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power communication scheduling, and particularly relates to a power distribution network communication network resource multi-time scale reservation method, system, device and medium. BACKGROUND

[0002] With the wide access of renewable energy in the power distribution network, especially the increasing proportion of distributed photovoltaic, wind power and other new energy generation, the power distribution network communication network is facing increasing data flow pressure. Especially in the scene of fault tripping, transient process or new energy change, a large amount of burst data with burstiness, high priority and low delay tolerance will be generated in the network, such as recording wave signal and protection instruction, and the burst transmission demand puts forward strict challenges to the existing communication resource configuration mechanism.

[0003] At present, the power distribution network communication system mostly adopts static resource reservation mechanism, which cannot dynamically adjust resource allocation according to business type and network state, and cannot meet the difference requirements of conventional business and peak business in transmission timeliness, data volume and processing priority. In addition, the existing peak data transmission pressure prediction method is often based on single time sequence characteristics, and does not fully model the dynamic coupling between multi-source heterogeneous data, resulting in insufficient prediction accuracy and unable to effectively support subsequent resource planning.

[0004] On the other hand, the existing resource regulation strategy generally lacks time scale division, and there is lack of linkage mechanism between resource baseline reservation in long time period and elastic scheduling in short time period, which is easy to cause low resource utilization efficiency or strategy out of control. The resource allocation in short time still depends on experience rules, and the action decision space is large, which is difficult to realize low delay and fine response in high dynamic scene. SUMMARY

[0005] In view of the above problems, the present application is proposed.

[0006] Therefore, the technical problem solved by the present application is: how to construct a multi-time scale oriented resource management mechanism in the power distribution network communication network, aiming at the difference characteristics of conventional business data and peak business data in data generation frequency, communication delay tolerance and computing processing demand, to realize the cooperative integration of static reservation in long time period and dynamic regulation in short time period of communication resources and computing resources, so as to improve the transmission timeliness of task data, resource utilization efficiency and the running stability of system in high data pressure scene.

[0007] To solve the above technical problems, the present application provides the following technical scheme: a power distribution network communication network resource multi-time scale reservation method, comprising,

[0008] Collect multi-source operation data in a power distribution network communication network, build a time series prediction model for predicting peak data transmission pressure based on the data, and predict a peak data transmission pressure value sequence in a plurality of future time steps;

[0009] According to the prediction result, combining the difference between the conventional service data and the peak service data in the power distribution network, using task age as a quantitative index, a resource reservation optimization model is constructed to minimize the timeliness loss of the task, and a basic resource reservation scheme is determined;

[0010] The basic resource reservation scheme is used as the initial configuration basis of the strategy network, combined with multi-source operation data, to generate communication resource control parameters and computing resource control parameters in a short time period, and update the resource allocation strategy.

[0011] As a preferred scheme of the power distribution network communication network resource multi-time scale reservation method, wherein: the time series prediction model includes at least one recursive structure and at least one attention mechanism module, the recursive structure is used to process the historical state variables of each time step in the multi-source operation data, and the attention mechanism module is used to weight the hidden state of multiple time steps output by the recursive structure, calculate the representation vector, and output the peak data transmission pressure prediction value sequence corresponding to the future multiple time steps through linear transformation.

[0012] As a preferred scheme of the power distribution network communication network resource multi-time scale reservation method, wherein: the prediction value sequence includes peak data transmission pressure estimation values of a plurality of future discrete time steps, wherein each prediction value corresponds to a future time step, the prediction value sequence is used as an input variable to participate in the variable setting of the resource reservation optimization model, and is associated with the optimization objective function structure for characterizing data timeliness.

[0013] As a preferred scheme of the power distribution network communication network resource multi-time scale reservation method, wherein: in the resource reservation optimization model, task age is used as a freshness quantitative index of various types of service data, and based on the task age, the difference parameters of conventional service data and peak service data in update cycle, observation noise covariance, data volume and calculation intensity are combined to construct a weighted timeliness loss function for optimizing the basic allocation proportion of communication resources and computing resources.

[0014] As a preferred scheme of the power distribution network communication network resource multi-time scale reservation method, the resource allocation strategy is updated by a regulation and control structure constructed based on a strategy network, the regulation and control structure receives a combined input of a basic resource reservation scheme and the multi-source operation data, and generates communication resource regulation parameters and computing resource regulation parameters in a short time period based on the current time communication state, resource configuration records and spike pressure prediction results.

[0015] As a preferred scheme of the power distribution network communication network resource multi-time scale reservation method, the recursive structure includes a long short-term memory neural network, the attention mechanism module is connected with the long short-term memory neural network at the outputs of multiple time steps, is used for calculating attention coefficients of each time step, and performs weighted summation on the multiple time step outputs to form a representation vector for spike data transmission pressure prediction.

[0016] The beneficial effects of the preferred technical scheme are that: by using the long short-term memory neural network to extract long-term dependence features in the time series, and combining the attention mechanism to weight the output results of the key time steps, the prediction model has stronger feature extraction capability in capturing the spike data mutation characteristics, and has higher prediction accuracy.

[0017] As a preferred scheme of the power distribution network communication network resource multi-time scale reservation method, the strategy network includes a discrete strategy sub-network and a continuous strategy sub-network, the discrete strategy sub-network outputs a resource allocation direction label according to the communication state, the resource configuration record and the spike pressure prediction result included in the multi-source operation data;

[0018] The continuous strategy sub-network outputs corresponding resource allocation amplitude parameters based on the allocation direction;

[0019] The allocation direction label and the resource allocation amplitude parameter are used together to determine the communication resource regulation parameter and the computing resource regulation parameter.

[0020] The beneficial effects of the preferred technical scheme are that: by outputting the resource allocation direction and the resource adjustment amplitude by different sub-networks respectively, the coupling degree of the overall strategy generation is reduced, the flexibility and executability of the resource allocation strategy are improved while meeting the low delay response requirement.

[0021] The application provides a power distribution network communication network resource multi-time scale reservation system.

[0022] To solve the above technical problems, the application provides the following technical solutions: a power distribution network communication network resource multi-time scale reservation system, comprising: a data acquisition module, configured to acquire multi-source operation data in the power distribution network communication network, construct a time series prediction model for peak data transmission pressure prediction based on the data, and predict a peak data transmission pressure value sequence in a future plurality of time steps;

[0023] A reservation optimization module is configured to determine a basic resource reservation scheme according to the prediction result, in combination with the difference between conventional service data and peak service data in the power distribution network, adopt a task age as a quantitative index, and construct a resource reservation optimization model with the minimum loss of task timeliness as the target.

[0024] A strategy configuration module is configured to combine the basic resource reservation scheme as an initial configuration of a policy network with the multi-source operation data, respectively generate communication resource control parameters and computing resource control parameters in a short time period, and update a resource allocation strategy.

[0025] The application provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the power distribution network communication network resource multi-time scale reservation method when executing the computer program.

[0026] The application provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the power distribution network communication network resource multi-time scale reservation method when executed by a processor.

[0027] The application has the following beneficial effects: by introducing a prediction model combining an attention mechanism and a long short-term memory structure, the time dependence and key fluctuation characteristics of burst data can be more effectively captured, the accuracy of peak pressure prediction is significantly improved, and the lag of resource preparation is reduced.

[0028] The introduction of a task age index quantifies the different business needs for real-time performance, so that resource allocation is more in line with the urgency of the task, and the situation that high-priority business is delayed when resources are insufficient is avoided.

[0029] A stable resource configuration baseline is constructed in a long period, and resource allocation is adjusted in time according to network state changes in a short period, so that the system has stronger load adaptability and flexibility.

[0030] The strategy calculation complexity is reduced, and the decision-making efficiency is improved: the strategy structure separates the control direction and amplitude, reduces the action space dimension, makes the reinforcement learning strategy easier to train and deploy, and adapts to the demand of the power distribution network for second-level response speed.

[0031] By constructing a multi-objective reward mechanism that takes into account delay improvement, load balancing and policy diversity, the system can maintain efficient and balanced resource usage levels when responding to large-scale data burst scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0033] Figure 1 The overall flowchart of a power distribution network communication network resource multi-time scale reservation method provided by an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0034] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0035] Embodiment 1, refer to Figure 1 For an embodiment of the present application, the embodiment provides a power distribution network communication network resource multi-time scale reservation method, comprising:

[0036] Step 101: Collect multi-source operation data in the power distribution network communication network, build a time series prediction model for peak data transmission pressure prediction based on the data, and predict the peak data transmission pressure value sequence in the future multiple time steps;

[0037] Step 102: According to the prediction result, combining the difference between the conventional business data and the peak business data in the power distribution network, using task age as a quantitative index, building a resource reservation optimization model with the goal of minimizing task timeliness loss, and determining the basic resource reservation scheme;

[0038] Step 103: Take the basic resource reservation scheme as the initial configuration of the policy network, combine it with the multi-source operation data, generate communication resource control parameters and computing resource control parameters in a short time period respectively, and update the resource allocation strategy.

[0039] In this embodiment, the "multi-source operation data" described in step 101 includes but is not limited to historical data transmission volume, bandwidth occupancy rate, server utilization rate, user connection number and historical peak data transmission pressure label in the power distribution network communication network. These data have typical time series characteristics and are affected by various external disturbance factors, showing nonlinear and multi-scale coupling characteristics. Therefore, this embodiment uses a hybrid model containing a long short-term memory structure (LSTM) and an attention mechanism for modeling: the LSTM is used to capture the long-time dependent input feature changes, and the attention mechanism assigns weights to the historical input states at different time steps, highlighting the contribution of key influencing factors during peak periods, thereby outputting a peak pressure prediction value sequence in multiple future time steps, providing accurate data support for subsequent resource allocation.

[0040] In step 102, to further improve the pertinence and adaptability of the resource reservation strategy, the Age of Task (AoT) is introduced as a quantitative indicator in the embodiment. This indicator is based on the time difference from the generation to the processing of each type of business data, reflecting its timeliness decay characteristics. According to the predicted peak pressure sequence, combined with the update period of regular business data, observation noise covariance, calculation intensity and data size, etc. Business characteristics, construct the weight term in the resource reservation optimization model. In the optimization model, the objective function is set to minimize the weighted value of task timeliness loss, and the decision variable is the reservation ratio of basic communication resources and computing resources. By solving the optimization model, the basic resource allocation strategy suitable for the current load characteristics and business demand in a long time period can be obtained.

[0041] In step 103, the basic resource reservation scheme determined in the previous step is used as the initialization parameter of the reinforcement learning strategy network in this embodiment, and combined with the multi-source operation data collected at the current time, such as network utilization rate, data queue length, historical allocation effect record and other feature information, is input into the strategy network for dynamic calculation. The strategy network is composed of a discrete direction sub-network and a continuous amplitude sub-network, the former is used to judge the resource allocation direction (increase, decrease, maintain), and the latter outputs the allocation amplitude of the corresponding node. This structure can effectively realize the separate modeling of resource allocation behavior, reducing the complexity of the high-dimensional policy search space. The finally generated communication resource control parameters and computing resource control parameters are used to guide real-time resource allocation actions in a short period, so that the system has the ability to adaptively adjust in response to peak flow disturbances, ensuring the continuity of data business and the stability of network operation.

[0042] Embodiment 2, an embodiment of the present application, based on the previous embodiment provides a power distribution network communication network resource multi-time scale reservation method, comprising:

[0043] In step 101, as the data transmission volume presents a sharp growth, the peak data transmission pressure brings great challenges to the stability and reliability of the network system. The peak data transmission pressure is influenced by a variety of complex factors, has a high degree of nonlinearity and dynamics, and accurate prediction of the peak data transmission pressure can take corresponding resource allocation and optimization strategies in advance, avoid network congestion, data loss and other problems, and guarantee the efficiency and stability of data transmission.

[0044] In a preferred embodiment of the present application, a peak data transmission pressure prediction method based on an improved LSTM-Attention model is proposed, which fully utilizes the time series characteristics of multi-factor data, combines the long-term dependence capturing ability of LSTM and the focusing ability of attention mechanism on key information, and improves the accuracy and reliability of peak data transmission pressure prediction.

[0045] The specific implementation is: first, collect historical data transmission volume Network bandwidth occupancy rate Server utilization User concurrent connection number And historical transmission pressure label y t-i (i = 1, 2, …, m) and process the missing values and outliers in the data to ensure the integrity and accuracy of the data.

[0046] Let the input vector be:

[0047] The predicted output vector is: The transmission pressure peak value vector of the next n time steps is predicted, where represents the predicted transmission pressure value of the t+i time step.

[0048] It should be noted that the LSTM layer is specially used to process sequence data and capture long-term dependencies. The change of peak data transmission pressure has time series characteristics, and there is a close relationship between data at different times. LSTM can selectively remember or forget historical information through a unique gating mechanism. The forgetting gate determines which information to discard from the cell state, the input gate determines the new information to be saved to the cell state, the cell state is responsible for integrating historical information and current input information, and the output gate controls which part of the cell state will be used for the output at the current time.

[0049] At time step t, the current decoder state s t The attention score e j of the jth time step hidden state h t,j of the encoder:

[0050] e t,j =W a tanh(W h h j +W s s t )

[0051] wherein, W a is a weight matrix for calculating attention scores, W h is a weight matrix for linearly transforming the encoder hidden state; and W s is a weight matrix for linearly transforming the current state of the decoder.

[0052] The attention scores are converted into a probability distribution by using a softmax function:

[0053] a t,j =softmax(e t,j )

[0054] The context vector is defined as c t , representing the weighted sum of the encoder hidden states:

[0055]

[0056] The spike data transmission pressure value is predicted according to the context vector and the current hidden state

[0057]

[0058] wherein, W o is a weight matrix of the output layer, and b o is a bias vector of the output layer. The spike data transmission pressure prediction method based on the improved LSTM-Attention model can effectively capture the complex change law of the spike data transmission pressure by fully utilizing the time series characteristics of multi-factor data, combining the LSTM network and the attention mechanism.

[0059] The preferred technical scheme has the beneficial effects that: in the power distribution network communication network environment facing high dynamic and burst traffic impact, the scheme can provide stable and reliable prediction support for subsequent resource reservation strategies through continuous prediction of multi-time step spike pressure, and significantly improve the sensitive response capability and resource allocation efficiency of the system in the burst scenario.

[0060] In an optional embodiment of the present application, the time series prediction model is replaced by a multi-head attention prediction model based on the Transformer structure. The model adopts a position encoding mechanism to process the input time series, and simultaneously captures short-term fluctuations and long-term dependencies through a multi-head attention layer, and is suitable for complex scenarios with dramatic spike data mode distribution and highly non-stationary time series.

[0061] In another optional embodiment of the present application, the time series prediction model uses a combination structure of graph convolutional neural network (GCN) and gated recurrent unit (GRU) to model the spatial correlation between the communication nodes of the power distribution network, while extracting the dynamic evolution characteristics of the node sequence state, and to realize the spike pressure prediction in the space-time joint domain.

[0062] In a preferred embodiment of the present application, in step 102, since all slices observe the same physical process in the joint estimation scenario, they share the same state equation, and the dynamics of this hierarchical slice architecture can be represented by the following discrete-time system:

[0063] x(t+1)=Ax(t)+Bu(t)+w(t)

[0064] where x(t) represents the state of the power distribution network at time slot t, u(t) is the control input of the power distribution network. The state transition matrix is represented as A, the control input matrix is represented as B, and the system noise w(t) obeys a Gaussian distribution with a mean of zero.

[0065] It should be noted that the data obtained by the power distribution network can be divided into two categories, namely conventional data y D (t) and spike data y B (t). The conventional data is affected by the observation noise v D (t), has a larger covariance, has lower measurement accuracy but higher timeliness requirement. The spike data has a small observation noise covariance, but the data volume is large and requires high bandwidth support, and its observation equation is:

[0066] y D (t)=x(t)+v D (t)

[0067] y B (t)=x(t)+v B (t)

[0068] It should be noted that since the measurement freshness has an important influence on the estimation performance, it is important to characterize the timeliness of the measurement. AoT can capture the freshness of the task until the receiving end obtains the state information from all observation devices and completes the corresponding processing. According to the definition of task age, its value can be obtained by subtracting the generation time slot of the last received state in the latest observation from the current time slot:

[0069] Δ D (t)=t-τ D (t)

[0070] Δ B (t)=t-τ B (t)

[0071] where Δ D (t) represents the time interval of normal data generation to processing, Δ B (t) represents the time interval of peak data generation to processing, and τ D (t) and τ B (t) represent the normal data and peak data generation time, respectively.

[0072] Specifically, in order to improve the adaptability of power distribution network monitoring tasks to system dynamic changes and resource constraints, the transmission and computing resource reservation ratios need to be jointly optimized under the association slice constraints. The goal is to minimize the monitoring data timeliness loss represented by data age within a long time scale reservation window, while meeting the resource capacity constraints and system stability requirements:

[0073]

[0074] where ξ D and ξ B are weight factors, and satisfy ξ D +ξ B =1, which reflects the system's priority consideration for different types of data. When the power distribution network is in normal operation, more attention is paid to the timeliness of normal data, and the value of ξ D is relatively large; when the power distribution network fails, peak data becomes more important, and the value of ξ B increases accordingly to ensure that peak data can be processed first and transmitted with low delay. T represents the length of the optimization time window.

[0075] A series of constraints are also set to ensure the rationality of resource allocation and the stability of the system:

[0076]

[0077] The above formula limits the communication resource reservation ratio β and the computing resource reservation ratio γ, which ensures that the total communication and computing resources allocated to normal data and peak data do not exceed the total resource capacity of the system, and in order to ensure that the transmission delay and the computing delay of normal data are within the maximum allowed delay with a probability of no less than Here, ρ(A) is the spectral radius of the system matrix A, and when ρ(A)>1, it indicates that the system is unstable, so the delay needs to be strictly controlled to maintain the stability of the system.

[0078] It should be noted that in order to achieve optimal allocation of resources, the reservation ratios of communication resources and computing resources can be obtained as follows:

[0079]

[0080] The above formula is used to calculate the communication resource reservation ratio Comprehensive consideration of multiple factors, wherein ξ s Reflects the priority of data, the higher the priority, the more communication resources that can be allocated; Lambda s Indicates the size of the data packet; S s Is the number of devices within the slice, the more devices, the more data generated, and the greater the demand for communication resources; Then reflects the noise characteristics, noise characteristics will affect the reliability of data transmission, the greater the noise, in order to ensure accurate data transmission, more communication resources are needed.

[0081]

[0082] The above formula is used to calculate the resource reservation ratio Further introduces the computing intensity X s This factor. Different types of data have different demands for computing resources when processing, and peak data may need complex data analysis and processing, requiring higher computing intensity X s Therefore, when allocating computing resources, consider comprehensively to ensure that computing resources can be reasonably allocated to meet the needs of data processing.

[0083] It should be noted that the method constructs a resource reservation baseline under a long time scale by dynamic priority mapping and multi-dimensional parameter fusion, the weight factor is adaptively switched according to the power grid operation state, scene adaptation of resource preemption rules is realized, and stable guarantee is provided for short-period elastic scheduling.

[0084] The beneficial effects of the preferred technical solution are: under the background of highly dynamic changes of the power grid state, computing and communication resources can be dynamically allocated according to the time urgency of the task, the estimation inaccuracy risk caused by delay of high-priority data can be significantly reduced; and system stability constraints are embedded in the reservation scheme, so that the resource configuration strategy can still maintain the estimation performance without degradation when responding to sudden growth of data, and the resilience and adaptability of the distribution network communication system are strengthened.

[0085] In an optional embodiment of the present application, the resource reservation optimization model does not use the task age index, but introduces the product of data waiting time and resource occupancy rate as a scheduling urgency measure. Specifically, for each type of data task, record its waiting time in the communication / computing queue and combine historical resource occupancy statistics to construct a product index as a scheduling urgency; under this index, a linear objective function is constructed to minimize the weighted sum of waiting and occupancy in the whole system, to determine the allocation weight of resources.

[0086] In another optional embodiment of the present application, the basic resource reservation scheme is replaced by an online sampling heuristic proportional rule instead of the analytical optimization method: by periodically counting the total transmission amount and average processing delay of the peak data and regular data in the last scheduling period, an empirical mapping rule is constructed, for example, when the average delay of the peak data exceeds a certain threshold, the reservation resource proportion is automatically increased, otherwise it is reduced, forming a feedback adjustment model, realizing a lightweight resource reservation mechanism without modeling.

[0087] In a preferred embodiment of the present application, step 103, the communication network resources of the power distribution network need to be dynamically adjusted in a short time scale to cope with real-time data transmission pressure. This method constructs an adaptive optimization mechanism through a reinforcement learning framework, the core idea of which is to transform the resource allocation problem into a dynamic decision-making process, and through the continuous interaction between the agent and the environment, the optimal strategy is gradually learned. The objective function needs to balance the minimization of delay and load balancing:

[0088]

[0089] where w t =[w1,w2,...,w n ] represents the resource allocation weight vector of each node communication network at time t, satisfying ∑w i =1. D(w t ) is the average transmission delay of the network, which is negatively correlated with the resource allocation of the communication network, L(w t ) represents the node load balancing index, the smaller the value, the more balanced. α∈[0,1] is a dynamic weight parameter, which adjusts the priority according to the real-time pressure, when there is a data peak, α tends to 1, choosing to reduce the delay first.

[0090] State space construction, the state needs to include the real-time state of the network and the prediction result of the pressure:

[0091]

[0092] where u t =[u1,u2,...,u n ] represents the bandwidth utilization of each node, u i ∈[0,1]; q t =[q1,q2,...,q n ] represents the data transmission queue length; represents the predicted peak flow intensity of the next step; h t-k:t represents the historical resource allocation effect of the past k time windows.

[0093] Specifically, the action space is divided into direction decision and amplitude decision, both of which jointly affect the resource weight adjustment. The short-term resource allocation of distribution network needs fast response, involving high real-time requirements, so it is necessary to reduce the complexity of action space, and the direction decision allocates a discrete direction label to each node: when , the node releases resources; when , the current weight is kept unchanged; when , the node is allocated more resources.

[0094] Further, the amplitude decision allocates a continuous adjustment amplitude, δ max , to each node, and the maximum adjustment step size is set as

[0095] To balance multiple objectives, a hierarchical reward function is constructed, and the main reward and auxiliary reward are designed:

[0096] R(s t ,a t )=λ1R d +(1-λ1)R l +ηR c

[0097] wherein, is the delay improvement rate, D base is the baseline delay; is the load balancing improvement rate, both of which are the main reward items. The exploration reward item encourages trying low-frequency allocation combinations, and N(w t ) is the number of occurrences of w t in history, and η is the exploration coefficient, which decays with the training.

[0098] The strategy network is decomposed into resource flow strategy and adjustment amplitude strategy:

[0099]

[0100] The resource flow strategy π d outputs the probability distribution of the adjustment direction of each node; the adjustment amplitude strategy π q generates the Gaussian distribution parameters of the amplitude after the direction is given.

[0101] The conditional entropy constraint is to improve diversity, and the entropy of the amplitude strategy is constrained:

[0102]

[0103] wherein, Max Entropy of amplitude decision, β∈[0,1] represents dynamic adjustment, which can reduce the entropy to stabilize the strategy when the network load is high.

[0104] A dynamic weight adjustment mechanism is designed according to the spike pressure prediction Real-time update of main target weight α:

[0105]

[0106] Wherein, Sigmoid function, which maps the input to [0,1]; y th Traffic threshold, exceeding which is determined as spike pressure; k slope parameter, controlling the sensitivity of weight change.

[0107] It should be noted that through the hierarchical decision architecture and the dynamic weight self-adaptive mechanism, the agility and stability of short-term resource scheduling are unified, the direction decision module ensures the second-level response speed with low complexity, and the amplitude decision module supports fine resource adaptation through continuous action space; the dynamic reward function and the entropy constraint balance the multi-objective optimization and the strategy robustness requirement. In addition, this architecture enables the power distribution network communication network to not only instantaneously trigger resource tilt to suppress queue backlog when coping with spike pressure, but also maintain the smoothness of load migration through amplitude control and entropy constraint.

[0108] The beneficial effects of the preferred embodiment are: by introducing a reinforcement learning strategy framework, a double-strategy network system is constructed, the decoupling generation of communication resource and computing resource control direction and amplitude is realized, and the resource adjustment action is more flexible and has higher control precision. The state space design integrates the current state of the network, historical feedback and spike pressure prediction, which significantly improves the real-time response ability and foresight of the strategy; the hierarchical reward function combines the multi-objective index design, which can dynamically balance the dual requirements of delay reduction and load balancing; and the entropy constraint and strategy diversity incentive mechanism effectively prevent the strategy from falling into local optimum. The scheme realizes fine and adaptive resource scheduling of the power distribution network communication network in a high dynamic scene, and improves the service robustness and resource use efficiency of the system in a spike pressure scene.

[0109] In an optional embodiment of the present application, the short-term resource regulation does not use the reinforcement learning structure, but introduces a fuzzy control system based on rules. The controller uses the spike flow prediction value, the degree of communication congestion and the historical scheduling effect as input, and outputs the resource adjustment strategy through the pre-defined membership function and rule base. For example, if the predicted spike flow is high and the current node is severely congested, the system tends to increase the resource proportion of the node; otherwise, the resource quota is reduced. This method has low computational overhead and is simple to implement, and is suitable for edge scenarios with high scheduling frequency and limited computing power.

[0110] In another optional embodiment of the present application: an optimization method based on Evolution Strategy (ES) is used to replace the reinforcement learning model. The system maintains a population of resource allocation parameter vectors, and uses a mutation and selection mechanism to evolve based on short-term scheduling effects (such as average latency and load standard deviation) to update the allocation scheme for the next period. This method does not require gradient, and is suitable for high nonlinear, non-differentiable environment modeling scenarios, and has stronger robustness and stability in scenarios with sparse data or delayed sample feedback.

[0111] Embodiment 3 is an embodiment of the present application, which provides a power distribution network communication network resource multi-time scale reservation system, comprising:

[0112] A data acquisition module is configured to acquire multi-source operation data in the power distribution network communication network, construct a time series prediction model for peak data transmission pressure prediction based on the data, and predict a peak data transmission pressure value sequence in a plurality of future time steps.

[0113] A reservation optimization module is configured to determine a basic resource reservation scheme by constructing a resource reservation optimization model with task timeliness loss minimization as the target according to the prediction results and the difference between the conventional service data and the peak service data in the power distribution network, and using task age as a quantitative index.

[0114] A strategy configuration module is configured to combine the multi-source operation data with the basic resource reservation scheme as the initial configuration of the policy network, generate communication resource control parameters and computing resource control parameters in a short time period respectively, and update the resource allocation strategy.

[0115] The embodiment also provides an electronic device suitable for a power distribution network communication network resource multi-time scale reservation method, which comprises a memory and a processor; the memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions to implement a power distribution network communication network resource multi-time scale reservation method as described in the above embodiment.

[0116] The embodiment also provides a storage medium having a computer program stored thereon, which is executed by a processor to implement a power distribution network communication network resource multi-time scale reservation method as described in the above embodiment.

[0117] The storage medium provided in the embodiment and the power distribution network communication network resource multi-time scale reservation method provided in the above embodiment belong to the same inventive concept, and the technical details not described in the embodiment can be referred to the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.

[0118] From the above description of the embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software and necessary universal hardware, and of course can also be implemented by hardware, but in many cases the former is a better implementation. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a floppy disk, a ROM, a RAM, a FLASH, a hard disk, or an optical disc, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of various embodiments of the present application.

[0119] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application, and although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application, and all of them should be covered in the scope of the claims of the present application.

Claims

1. A method for reserving resources across multiple time scales in a power distribution network communication network, characterized in that: include, Collect multi-source operation data from the power distribution network communication network, and construct a time series prediction model based on the data for predicting peak data transmission pressure, predicting the peak data transmission pressure value sequence in multiple future time steps; Based on the prediction results and the differences between routine business data and peak business data in the distribution network, the task age is used as a quantitative indicator to construct a resource reservation optimization model with the goal of minimizing task timeliness loss, and to determine the basic resource reservation scheme. The basic resource reservation scheme is used as the initial configuration basis for the policy network. Combined with multi-source operation data, communication resource control parameters and computing resource control parameters are generated for short time periods, and the resource allocation strategy is updated.

2. The method for reserving resources across multiple time scales in a power distribution network as described in claim 1, characterized in that: The time-series prediction model includes at least one recursive structure and at least one attention mechanism module. The recursive structure is used to process the historical state variables of each time step in the multi-source running data. The attention mechanism module is used to assign weights to the hidden states of multiple time steps output by the recursive structure, calculate the representation vector, and output the predicted value sequence of peak data transmission pressure corresponding to multiple future time steps through linear transformation.

3. The method for reserving resources across multiple time scales in a power distribution network as described in claim 2, characterized in that: The predicted value sequence includes peak data transmission pressure estimates for multiple future discrete time steps, where each predicted value corresponds to a future time step. The predicted value sequence is used as an input variable in the resource reservation optimization model and is associated with the optimization objective function structure used to characterize data timeliness.

4. The method for reserving resources across multiple time scales in a power distribution network as described in claim 3, characterized in that: The resource reservation optimization model uses task age as a measure of the freshness of various types of business data. Based on the task age and the differences between regular business data and peak business data in terms of update cycle, observation noise covariance, data volume and computational intensity, a weighted timeliness loss function is constructed to optimize the basic allocation ratio of communication resources and computing resources.

5. The method for reserving resources across multiple time scales in a power distribution network as described in claim 4, characterized in that: The resource allocation strategy is updated by a control structure built on a policy network. The control structure receives a combination of basic resource reservation scheme and multi-source operation data as input, and generates communication resource control parameters and computing resource control parameters for a short period of time based on the current communication status, resource configuration records and peak pressure prediction results.

6. The method for reserving resources across multiple time scales in a power distribution network as described in claim 5, characterized in that: The recursive structure includes a long short-term memory neural network. The attention mechanism module is connected to the output of the long short-term memory neural network at multiple time steps to calculate the attention coefficient at each time step and perform a weighted summation of the outputs at multiple time steps to form a representation vector for predicting peak data transmission pressure.

7. The method for reserving resources across multiple time scales in a power distribution network as described in claim 6, characterized in that: The strategy network includes a discrete strategy sub-network and a continuous strategy sub-network. The discrete strategy sub-network outputs a resource allocation direction label based on the communication status, resource configuration records, and peak pressure prediction results included in the multi-source operation data. The continuous strategy sub-network outputs the corresponding resource allocation amplitude parameters based on the allocation direction; The allocation direction label and resource allocation amplitude parameter are used together to determine the communication resource control parameters and the computing resource control parameters.

8. A multi-timescale reservation system for distribution network communication network resources, employing the multi-timescale reservation method for distribution network communication network resources as described in any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to collect multi-source operation data in the power distribution network communication network, and to build a time series prediction model for peak data transmission pressure prediction based on the data, predicting the peak data transmission pressure value sequence in multiple future time steps. The reservation optimization module is used to construct a resource reservation optimization model with the goal of minimizing task timeliness loss based on the prediction results and the difference between routine business data and peak business data in the distribution network, using task age as a quantitative indicator, and to determine the basic resource reservation scheme. The strategy configuration module is used to take the basic resource reservation scheme as the initial configuration basis of the strategy network, combine it with multi-source operation data, generate communication resource control parameters and computing resource control parameters for short time periods, and update the resource allocation strategy.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the multi-timescale reservation method for distribution network communication network resources according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the multi-timescale reservation method for distribution network communication network resources according to any one of claims 1 to 7.

Images