Network slice size adjustment method based on distributed photovoltaic power generation

Abstract

The application provides a network slice size adjustment method based on distributed photovoltaic power generation, and belongs to the technical field of photovoltaic power generation network slices. The method comprises the following steps: S10, acquiring environmental parameters of the current state of a distributed photovoltaic network slice system; S20, obtaining an initial system state according to the acquired environmental parameters, wherein the system state represents the power distribution of the network slice in each time slot; S30, establishing an optimization problem of a network slice size adjustment target according to the system state; S40, solving the optimization problem to obtain an optimal solution; S50, adjusting the state by using the obtained optimal solution; S60, taking the adjusted state as the system state, and repeating steps S30-S60 until a repeated execution end condition is reached, and taking the result of the last state adjustment as a target state; and S70, calculating a network slice size target value of the distributed photovoltaic network slice system according to the target state.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic power generation network slicing technology, and more specifically, relates to a method for adjusting the size of network slices based on distributed photovoltaic power generation. Background Technology

[0002] In recent years, with the continuous expansion of the global photovoltaic industry, the proportion of distributed photovoltaic power generation technology in global electricity supply has been steadily increasing. With the rapid development of the industry and the transformation towards an energy internet of interconnected things, leveraging the advantages of extensive and diverse infrastructure, new, diversified, and high-standard requirements have been placed on communication networks. The comprehensive deployment of 5G networks provides strong communication capability support for the construction of "three types and two networks." In response to these needs, network slicing technology has emerged. Through network slicing, operators can build multiple dedicated, virtualized, and isolated logical networks on top of a general physical network to meet the differentiated network capability requirements of different customers. As a customized dedicated logical network, network slicing can meet the specific needs of various business scenarios in the smart grid for communication network systems.

[0003] Network slicing is a new network architecture that provides multiple logical networks on the same shared network infrastructure, each serving a specific business type or industry user. Each network slice can flexibly define its own logical topology, SLA requirements, reliability, and security level to meet the differentiated needs of different businesses, industries, or users. Slicing refers to a way of dividing system resources according to certain rules or proportions. Resources can be power resources, memory, network bandwidth, etc., and slices are the logical units that divide these resources. Adjusting the network slice size is a crucial step in the slice lifecycle. By adjusting slices, the system can manage and schedule resources more finely. The system can allocate appropriate slice resources to each task based on the nature and requirements of the task, changing the slice size to ensure reasonable resource utilization and normal task execution. For network slices targeting different business types, a resource allocation strategy needs to be determined during deployment to guarantee the services provided by the network slices. Traditional resource allocation algorithms usually consider resource utilization, maximizing resource utilization to maximize operator revenue while ensuring fairness. However, for network slice resource allocation, the continuous changes in multi-service scenarios and the dynamic requests for slices by various types of services cause resource utilization to decrease over time. Meanwhile, QoE is also an important consideration for slice resource allocation. Latency and packet loss issues will increase as resource utilization increases, thus affecting user experience. Summary of the Invention

[0004] In view of this, the present invention provides a method for adjusting the network slice size based on distributed photovoltaic power generation, which can solve the technical problem that when adjusting the network slice size of distributed photovoltaic power generation in the existing technology, the latency and packet loss rate will increase as the resource utilization rate increases, thus affecting the user experience.

[0005] This invention is implemented as follows:

[0006] This invention provides a method for adjusting the network slice size based on distributed photovoltaic power generation, comprising the following steps:

[0007] S10. Obtain the environmental parameters of the current state of the distributed photovoltaic network slicing system;

[0008] S20. Obtain the initial system state based on the acquired environmental parameters. The system state represents the power distribution of the network slice in each time slot.

[0009] S30. Based on the system status, establish an optimization problem for adjusting the network slice size target;

[0010] S40. Solve the optimization problem to obtain the optimal solution;

[0011] S50. Adjust the state using the obtained optimal solution;

[0012] S60. Take the adjusted state as the system state and repeat steps S30-S60 until the repetition termination condition is met, and take the result of the last state adjustment as the target state.

[0013] S70. Calculate the target value of the network slice size of the distributed photovoltaic network slicing system based on the target state.

[0014] Based on the above technical solution, the network slice size adjustment method based on distributed photovoltaic power generation of the present invention can be further improved as follows:

[0015] The environmental parameters specifically include: power allocation array, downlink data volume of each slice in each time slot, slice latency requirements, total power generation of the access resource layer, and the QoE of the service.

[0016] Furthermore, the step of obtaining the initial system state based on the acquired environmental parameters specifically includes:

[0017] Calculate the power consumption of each slice in each time slot based on the downlink data volume of each slice in each time slot;

[0018] The remaining power of each slice is obtained by subtracting the actual power consumption of the slice from the power consumption of each slice, as obtained from the power metering data.

[0019] The slices are sorted according to the priority of power allocation, and a two-dimensional system state array representing the power allocation status is generated as the initial system state.

[0020] Furthermore, the step of establishing an optimization problem for adjusting the network slice size based on the system state specifically includes:

[0021] Establish a normalized comprehensive value with total power generation as the constraint, and maximize energy efficiency and user experience quality as the objective function;

[0022] Add power constraints, remaining power constraints, and delay constraints to the objective function;

[0023] Construct a Lagrange function, introduce multipliers to relax constraints, and obtain the optimization problem.

[0024] Furthermore, the objective function considers dynamically planning the power generation to simultaneously achieve optimal energy efficiency and maximize the user's QoE.

[0025] Furthermore, the method for solving the optimization problem to obtain the optimal solution is to use Q-learning to learn the optimal slicing strategy. Specifically, the state-action value function of Q-learning is updated asynchronously, that is, the current estimated value is updated with the state-action pair with the highest value in the next state.

[0026] Furthermore, the method for obtaining the optimal solution to the optimization problem is to use an iterative algorithm to solve it step by step or to use a heuristic algorithm to solve it quickly.

[0027] The step of adjusting the state using the obtained optimal solution specifically includes:

[0028] The state adjustment increment for each slice is calculated based on the obtained optimal solution and the initial system state. That is, the state adjustment increment is the optimal solution power value minus the current power value.

[0029] The termination condition for repeated execution is that the loop is executed 200 times, or the similarity between the optimal solutions obtained from two adjacent loops is greater than the termination threshold.

[0030] The beneficial effect of adopting the above-mentioned improvement scheme is that the termination threshold is 0.9, which can also be set according to the operator's experience.

[0031] Furthermore, the method for calculating the similarity of the optimal solutions obtained from two adjacent cycles is cosine similarity.

[0032] Compared with existing technologies, the network slice size adjustment method based on distributed photovoltaic power generation provided by this invention has the following advantages: This invention acquires photovoltaic power generation data in real time and, combined with factors such as slice load and QoS requirements, uses an optimization algorithm to dynamically calculate the optimal resource allocation scheme for different slices at different times. This avoids the problem of the separation between photovoltaic power generation and slice resources in existing technologies, achieving close matching and reasonable scheduling between the two, and ensuring the effective utilization of photovoltaic power generation. Simultaneously, this invention, while meeting slice QoS requirements, fully considers QoS indicators such as slice latency and packet loss rate, and uses rounding to ensure that the allocation result meets the constraints. This can significantly reduce latency jitter and disconnection problems caused by improper resource allocation, effectively improving user experience and increasing QoE.

[0033] In summary, this invention solves the technical problem that when adjusting the network slice size for distributed photovoltaic power generation in existing technologies, latency and packet loss rates often increase with the improvement of resource utilization, thereby affecting user experience. Attached Figure Description

[0034] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 A flowchart of the network slice size adjustment method based on distributed photovoltaic power generation provided by the present invention. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0037] like Figure 1 The diagram shows a flowchart of a network slice size adjustment method based on distributed photovoltaic power generation provided by this invention. This method includes the following steps:

[0038] S10. Obtain the environmental parameters of the current state of the distributed photovoltaic network slicing system;

[0039] S20. Obtain the initial system state based on the acquired environmental parameters. The system state represents the power distribution of the network slice in each time slot.

[0040] S30. Based on the system status, establish an optimization problem for adjusting the network slice size target;

[0041] S40. Solve the optimization problem to obtain the optimal solution;

[0042] S50. Adjust the state using the obtained optimal solution;

[0043] S60. Take the adjusted state as the system state and repeat steps S30-S60 until the repetition termination condition is met, and take the result of the last state adjustment as the target state.

[0044] S70. Calculate the target value of the network slice size of the distributed photovoltaic network slicing system based on the target state.

[0045] In the above technical solution, the environmental parameters specifically include: power allocation array, downlink data volume of each slice in each time slot, slice latency requirements, total power generation of the access resource layer, and the QoE of the service.

[0046] Below are two specific embodiments of the above method.

[0047] Example 1

[0048] The specific steps for S20 to obtain the environmental parameters of the current state of the distributed photovoltaic network slicing system and obtain the initial system state are as follows:

[0049] 1. Define parameters:

[0050] D i,j Downlink data volume of the j-th slice in the i-th time slot

[0051] C i,j Power consumption of the j-th slice in the i-th time slot

[0052] C′ i,j The actual power consumption of the j-th slice in the i-th time slot.

[0053] E i,j The remaining power in the j-th slice of the i-th time slot.

[0054] P j Priority weight of the j-th slice

[0055] 2. Calculate power consumption:

[0056] For all i, j:

[0057] C i,j =f(D i,j )

[0058] 3. Calculate the remaining battery power:

[0059] For all i, j:

[0060] E i,j =C i,j -C′ i,j

[0061] 4. Slice sorting:

[0062] rank j = Sort(P) j )

[0063] Obtain the sorted slice index rank j

[0064] 5. Generate the state array S:

[0065]

[0066] S30, based on the system state, establishes an optimization problem for adjusting the network slice size objective and finds the optimal solution. The specific steps are as follows:

[0067] 1. Define parameters:

[0068] TE i Total power generation in the i-th time slot

[0069] QoE j : The QoE weight of the service corresponding to the j-th slice

[0070] a j Energy efficiency coefficient of the j-th slice

[0071] Delay_need j Delay requirement for the j-th slice

[0072] 2. Establish the following optimization model:

[0073]

[0074] st

[0075]

[0076] E i,j -R i,j ≥0,

[0077] R i,j -Delay_need j ×C i,j ≥0,

[0078] The objective function represents the weighted sum of energy efficiency coefficient and QoE; the first constraint represents the power generation limit for each time slot; the second constraint represents that the allocation cannot exceed the remaining power; and the third constraint represents that the allocated power must meet the delay requirements.

[0079] Delay_need j It can be determined by the type of service; for example, voice services require relatively low latency. This indicates the unit of energy efficiency calculation.

[0080] The problem can be transformed into an unconstrained optimization problem using the Lagrange relaxation method. By introducing Lagrange multipliers λ, μ, and η, the Lagrange function is constructed:

[0081]

[0082] Take the pair R i,j The optimal solution can be obtained when the partial derivative is equal to 0.

[0083] Several other methods can be considered to find the optimal solution:

[0084] 1) Solve step by step using an iterative algorithm.

[0085] The optimization problem is discretized, with both time slots and slices discretized. Starting from an initial feasible solution, the power allocation of each slice within each time slot is gradually adjusted to achieve a better objective function value. Specifically, a coordinate descent method can be used, where only one variable R is adjusted at a time. i,j With other variables fixed, find the R-value that increases the objective function. i,j The value of the objective function is then repeated until it is no longer possible to improve the objective function value.

[0086] 2) Solve quickly using heuristic algorithms

[0087] Due to the large variable space, heuristic algorithms such as genetic algorithms and ant colony algorithms can be considered. These algorithms can start from multiple different initial solutions and iterate using a strategy similar to biological evolution to quickly search for a better solution in the complex solution space.

[0088] The specific steps for S50 to adjust the state using the obtained optimal solution are as follows:

[0089] 1. Receive the optimal solution output by S40, which is in the format of a two-dimensional array, representing the optimal power allocation value for each time slot and each slice.

[0090] 2. Based on the current system status, calculate the increment for each adjustment. The increment is the optimal power value minus the current power value.

[0091] 3. Optionally, check whether the increment meets the constraints, such as the increment cannot exceed the remaining power or violate priority. If it does, the increment needs to be corrected to meet the constraints.

[0092] 4. Optionally, adjust the power allocation value in the system status array incrementally, and update relevant parameters such as remaining power.

[0093] 5. Optionally, check the validity of the adjusted system state, such as whether the time slices are balanced and whether the minimum requirements are met. If invalid, make minor adjustments until the requirements are met.

[0094] 6. Optionally, the adjusted system state can be updated back into the environment parameters as input for the next solution.

[0095] In S70, the specific steps for calculating the target slice size based on the last adjusted system state are as follows:

[0096] 1. In the system state array, calculate the sum of the time slot power for each slice to obtain the total power for each slice;

[0097] 2. Based on the total power consumption and average power consumption, the amount of physical resources required for each slice can be estimated;

[0098] 3. After adding the redundancy factor, the physical resource target of each slice is obtained;

[0099] 4. Convert physical resources into targets of the corresponding slice size;

[0100] 5. Returns the target array of slice sizes.

[0101] Example 2

[0102] The difference between Example 2 and Example 1 lies in the different requirements for each environmental parameter and the different methods for finding the optimal solution. In the description, the specific variable names used in Example 1 are irrelevant to those in Example 2, as detailed below:

[0103] Consider power allocation. Assuming the number of users in each service slice remains constant within a time slot, the power allocated to the slice within each time slot under a resource allocation strategy is fixed. All service types at the access layer can be represented as a set:

[0104] K = {k1,k2,...,k} N}

[0105] There are N types of services in total, so the system allocates N network slices, and each slice is represented by n∈N.

[0106] Consider the downlink data volume of each slice in each time slot. Assuming there are T time slots in total, the downlink data volume of each slice in each time slot can be expressed as:

[0107] V = {v1, v2, ..., v} N}

[0108] t=0,1,...,T-1

[0109] Consider the latency requirements for each slice. The latency requirement for each slice is:

[0110] D = {d1, d2, ..., d} N}

[0111] Consider the total power generation at the access resource layer. Assuming the total power generation is E, and En represents the power allocated to each service slice, then the following condition is met:

[0112]

[0113] Due to the limitations of energy resources and the wide range of connected devices, energy efficiency needs to be kept within a reasonable range.

[0114] S15. The QoE of the service needs to be considered. Affected metrics include transmission rate and packet loss rate. Power slicing latency is also a significant factor; failure to guarantee adequate latency will lead to a severe decrease in QoE. The optimization objective is to dynamically plan power generation to simultaneously optimize energy efficiency and user QoE. Considering that the energy efficiency value lies between [0,1], and its value varies depending on the QoE metric considered, a simple weighted sum cannot reflect the individual impact of these two metrics on the optimization objective. Therefore, these two values ​​are first normalized to one order of magnitude, and their performance improvement effects are considered before constructing the optimization objective.

[0115]

[0116] in and This represents the maximum value of QoE and energy efficiency EE, and (·)! represents the factorial operation.

[0117] The optimal strategy is to adjust the approach based on the specific needs of different business types, maximizing the combined QoE and EE value for each slice. Considering that the total power consumption of each business slice is fixed, this is a constrained optimization problem.

[0118] S16. The objective equation can be expressed as:

[0119]

[0120] The constraints take into account different slice delays and power resource requirements.

[0121] In this embodiment, Q-learning is used to learn the optimal slicing strategy. The state-action value function of Q-learning is updated asynchronously, that is, the current estimate is updated with the state-action pair with the highest value in the next state. The iteration of the estimate of the current state-action pair can be expressed as:

[0122]

[0123] Where α represents the learning rate.

[0124] In this embodiment, both SARSA and Q-learning adopt a greedy strategy for action selection, which can be expressed as:

[0125]

[0126] Where ε∈(0,1), that is, each action selection has a probability of ε of randomly selecting an action from the action set, and a probability of 1-ε of selecting the optimal action given the known information. In view of the exploratory nature of reinforcement learning, the value can be set larger in the early stage to allow it to explore more location environments; the value can be set smaller in the later stage to allow it to focus on utilizing the information that has already been explored.

[0127] For the slice power resource allocation strategy studied in this invention, the state transition probability is 1. The state set, action set and reward function are defined below.

[0128] State: The power allocation of a network slice in each time slot, which is related to the current network state. Therefore, assuming a fixed total number of users, each state consists of two elements: the service type of each time slot and the downlink data volume of each service type. The set of states within all time slots is defined as the state set in Q-learning.

[0129] Action: Allocate bandwidth to different service slices according to power consumption. QoE requirements: Each slice division can be regarded as an action. Therefore, the action set can be defined as a set of all power resource allocation schemes.

[0130] Reward function: The normalized sum of the QoE and EE of each service type slice after each power allocation is used as the reward for Q-learning. The reward function is the reward generated by the network resource allocation method within a certain time slot, and can be used to evaluate the quality of performing the current action in that state.

[0131] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for adjusting the size of network slices based on distributed photovoltaic power generation, characterized in that: Includes the following steps: S10. Obtain the environmental parameters of the current state of the distributed photovoltaic network slicing system; S20. Obtain the initial system state based on the acquired environmental parameters. The system state represents the power distribution of the network slice in each time slot. S30. Based on the system status, establish an optimization problem for adjusting the network slice size target; S40. Solve the optimization problem to obtain the optimal solution; S50. Adjust the state using the obtained optimal solution; S60. Take the adjusted state as the system state and repeat steps S30-S60 until the repetition termination condition is met, and take the result of the last state adjustment as the target state. S70. Calculate the target value of the network slice size of the distributed photovoltaic network slicing system based on the target state; The environmental parameters specifically include: power allocation array, downlink data volume of each slice in each time slot, slice latency requirements, total power generation of the access resource layer, and QoE of the service; The steps for establishing an optimization problem based on the system state to adjust the network slice size specifically include: Establish a normalized comprehensive value with total power generation as the constraint, and maximize energy efficiency and user experience quality as the objective function; Add power constraints, remaining power constraints, and delay constraints to the objective function; Construct a Lagrange function, introduce multipliers to relax constraints, and obtain the optimization problem.

2. The method for adjusting the network slice size based on distributed photovoltaic power generation according to claim 1, characterized in that, The step of obtaining the initial system state based on the acquired environmental parameters specifically includes: Calculate the power consumption of each slice in each time slot based on the downlink data volume of each slice in each time slot; The remaining power of each slice is obtained by subtracting the actual power consumption of the slice from the power consumption of each slice, as obtained from the power metering data. The slices are sorted according to the priority of power allocation, and a two-dimensional system state array representing the power allocation status is generated as the initial system state.

3. The method for adjusting the network slice size based on distributed photovoltaic power generation according to claim 1, characterized in that, The objective function considers dynamically planning the power generation to simultaneously achieve optimal energy efficiency and maximize the user's QoE.

4. The method for adjusting the network slice size based on distributed photovoltaic power generation according to claim 1, characterized in that, The method for solving the optimization problem and obtaining the optimal solution is to learn the optimal slicing strategy using Q-learning. Specifically, the state-action value function learned by Q-learning is updated asynchronously, that is, the current estimated value is updated with the state-action pair with the highest value in the next state.

5. The method for adjusting the network slice size based on distributed photovoltaic power generation according to claim 1, characterized in that, The method for obtaining the optimal solution to the optimization problem is to use an iterative algorithm to solve it step by step or to use a heuristic algorithm to solve it quickly.

6. The method for adjusting the network slice size based on distributed photovoltaic power generation according to claim 1, characterized in that, The step of adjusting the state using the obtained optimal solution specifically includes: The state adjustment increment for each slice is calculated based on the obtained optimal solution and the initial system state. That is, the state adjustment increment is the optimal solution power value minus the current power value.

7. The method for adjusting the network slice size based on distributed photovoltaic power generation according to claim 1, characterized in that, The termination condition for repeated execution is that the loop is executed 200 times, or the similarity between the optimal solutions obtained from two adjacent loops is greater than the termination threshold.

8. The method for adjusting the network slice size based on distributed photovoltaic power generation according to claim 7, characterized in that, The method for calculating the similarity of the optimal solutions obtained from two adjacent cycles is cosine similarity.

Images