Utility maximization resource allocation method for intelligent power grid network slices

Abstract

The invention relates to a utility maximization resource allocation method for smart grid network slices. The method comprises the following steps: acquiring resource demand parameters of different services and resource quantities owned by different resource blocks; establishing a network slice resource allocation model by taking network utility maximization as a target according to resource demand parameters of different services and resource quantities owned by different resource blocks; processing constraint conditions of the network slice resource allocation model by using a model constructed by a penalty function method; solving the network slice resource allocation model after constraint condition processing based on an early stop genetic algorithm improved by a tanh function, and outputting an optimal network slice resource allocation scheme; and allocating resource blocks to the power grid terminal of the network slice service according to the optimal network slice resource allocation scheme. According to the method, the quality of the optimal solution obtained under the same algorithm operation time is realized, and excessive useless time waste is avoided by introducing an early stop strategy, so that higher network utility is brought to the system.

Description

Technical Field

[0001] The present invention relates to a utility maximization resource allocation method for smart grid network slicing, belonging to the technical field of resource allocation of smart power networks. Background Art

[0002] With the rapid development of the fifth generation mobile communication technology (5G), the intelligentization of power grids has become a new direction for the development of power grids. At the same time, the allocation of 5G network service slice resources is an important component that cannot be ignored in the future development of power grids. Under limited network resources, through a reasonable allocation plan, as many diverse power grid service needs as possible can be met, resource utilization can be improved, resource waste can be reduced, and efficient operation of the power grid can be achieved.

[0003] In recent years, with the advent of the fifth-generation mobile communication system era, most industries have undergone profound technological changes, and the power grid will inevitably participate in the wave of 5G technology innovation. With the continuous expansion of the distribution network, the coverage of power grid services has become more and more extensive, and the types and number of power grid services have also increased rapidly. This requires the power grid terminal communication equipment to process massive business data while ensuring the differentiated and diversified service quality requirements between various services. However, at present, the cache resources and computing resources of power grid terminal equipment are limited, and it is difficult to put all data on the terminal equipment for users to use, which increases costs and makes it difficult to guarantee service quality. At the same time, communication entities that establish business relationships with power grid terminals will also face similar problems. How to provide differentiated and diverse service quality for power grid services under limited resources has become a challenge. If the traditional physical networking solution continues to be used, it will bring inestimable costs and operation and maintenance difficulties.

[0004] In response to the above problems, network slicing technology can divide the integrated physical network into multiple functionally independent logical networks to provide specific services for different business scenarios. Network slicing technology can efficiently allocate resources according to the specific needs of different businesses. Compared with the 4th Generation Mobile Communication Technology (4G) era, which caused competition for network resources due to unreasonable resource allocation, 5G introduces network slicing to enable different businesses to take what they need, thereby achieving the normal operation of more businesses under the same network resources. Nowadays, there are many related cases of using network slicing technology, such as holographic technology, smart cities, and unmanned driving, which actually use network slicing technology to allocate appropriate resources for operation. Network slicing technology has certainly solved the problem of resource competition between businesses, but compared with the 4G era when network resource allocation was completely fixed, the use of network slicing technology means that a certain amount of time and energy needs to be invested to formulate an allocation model for specific applicable scenarios, which increases the complexity of network resource allocation to a certain extent and requires additional costs to improve the model. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a resource allocation method for maximizing the utility of smart grid network slices. According to the differences in the types of grid services and the types of resources required by different services, a network slice resource allocation model is established in the context of smart grids with the goal of maximizing network utility. The model constructed based on the penalty function method is used to process the constraints to maximize the comprehensive utility of the terminal and the resource allocation party. The double tanh function early stopping genetic algorithm is used for solution, which effectively avoids the problem of the algorithm falling into the local optimal tanh too early.

[0006] Terminology explanation:

[0007] 1. Resource block: It is the smallest unit for allocating spectrum resources in wireless communications and is widely used in 5G New Radio (NR) and Long Term Evolution (LTE) systems. Resource blocks are clearly divided in time and frequency to ensure effective wireless resource management.

[0008] 2. eMBB: The full English name is Enhanced Mobile Broadband, which uses 5G's better network coverage and higher transmission rate to provide users with better Internet access services, enabling wireless Internet access to have higher Internet speeds and more stable transmission.

[0009] 3. uRLLC: The full English name is Ultra Reliable & Low Latency Communication, which means low latency and high reliability. According to its characteristics, it can be widely used in AR / VR, industrial control systems, traffic and transportation (such as unmanned driving), smart grid and smart home management, interactive remote medical diagnosis, etc.

[0010] 4. mMTC: The full English name is Massive Machine Type Communication, which refers to massive IoT communications or large-scale IoT services. The characteristics of 5G's low power consumption, large connections, low latency and high reliability are well adapted to IoT-oriented services, and can focus on solving the problems of IoT and vertical industry applications that traditional mobile communications cannot support well.

[0011] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0012] A utility maximization resource allocation method for smart grid network slicing, comprising:

[0013] Obtain resource requirement parameters for different services and the amount of resources owned by different resource blocks;

[0014] According to the resource demand parameters of different services and the resource amounts of different resource blocks, a network slice resource allocation model is established with the goal of maximizing network utility;

[0015] The model constructed using the penalty function method handles the constraints of the network slice resource allocation model;

[0016] The early stopping genetic algorithm improved based on the tanh function solves the network slice resource allocation model after the constraints are processed, and outputs the optimal network slice resource allocation solution;

[0017] Resource blocks are allocated to the grid terminals served by the network slice according to the optimal network slice resource allocation scheme.

[0018] Furthermore, resource requirement parameters of different power grid services include:

[0019] Assuming that each terminal corresponds to a terminal service, represents the energy consumed by the i-th terminal service, represents the delay of the i-th terminal service, Represents the data transmission rate required by the i-th terminal service.

[0020] Furthermore, according to the resource demand parameters of different services and the resource amounts of different resource blocks, a network slice resource allocation model is established with the goal of maximizing network utility; including:

[0021] Assume that the smart grid area is covered by N different base stations, there are I terminals in the area, and the different needs of terminal services in the area are served by different network slices. The number of network slice resource blocks is H.

[0022] Defined in base station g n Received from terminal b i Signal-to-interference-noise ratio for:

[0023]

[0024] in, From terminal b i Transmit to base station g n The transmission power, and Respectively represent terminal b i and terminal b k With base station g n The channel gain between , σ represents the additive white Gaussian noise.

[0025] Channel Gain The calculation formula is defined as:

[0026]

[0027] in, is the antenna gain, d e is the path attenuation exponent, is the carrier frequency, d i,n d k,l is the distance between the terminal and the base station, Obeys Rayleigh distribution;

[0028] After obtaining the signal-to-interference-to-noise ratio, the terminal b is obtained by the Shannon formula i Data transfer rate

[0029]

[0030] in, For terminal b i The allocated bandwidth;

[0031] Assume terminal b i The size of the packet to be transmitted is Terminal b i The amount of computation required is This gives terminal b i Delay

[0032]

[0033] in, For terminal b i Allocated computing resources;

[0034] Then we get terminal b i Energy consumption of transmitting data packets for:

[0035]

[0036] The utility function includes the utility of the power terminal and the utility of the control center;

[0037] Power terminal side utility function U B for:

[0038]

[0039] in, and are the penalty factors when the actual energy consumption, transmission rate, bandwidth, and delay cannot reach the optimal level. and Optimal energy consumption, transmission rate, bandwidth, and latency for terminal services. and The energy consumption, transmission rate, bandwidth, and latency of the terminal after it is actually allocated resource blocks;

[0040] The utility function U of the control center C for:

[0041]

[0042] in, represents the allocation relationship between terminals and resource blocks, α W , α c is the benefit of the allocated resource block unit bandwidth and computing resources, is the resource block operation cost;

[0043] Then, the objective function of the network slice resource allocation model with the goal of maximizing network utility is expressed as:

[0044] U=Max(β C U C +β B U B )(1.9)

[0045] Among them, β C With β B are the weights that the control center and the terminal place in the system utility respectively.

[0046] Furthermore, the constraints of the network slice resource allocation model include:

[0047] Constraint 1: Indicates the transmission speed after the terminal is allocated a resource block Must be greater than the minimum transmission speed tolerated by the business

[0048] Constraint 2: Indicates the bandwidth after the terminal is allocated to the resource block Must be greater than the minimum bandwidth tolerated by the service

[0049] Constraint 3: Indicates the energy consumption of the terminal after being allocated to the resource block Must be less than the maximum energy consumption tolerated by the business

[0050] Constraint 4: Indicates the delay after the terminal is allocated a resource block Must be less than the maximum delay tolerated by the service

[0051] Constraint 5: Indicates that each resource block can only be allocated to one terminal;

[0052] Constraint 6: Indicates that each terminal can only be allocated one resource block;

[0053] Constraint 7: Indicates that the bandwidth resources allocated to all terminals do not exceed all available bandwidth resources W in the system;

[0054] Constraint 8: It means that the computing resources allocated to all terminals do not exceed all the computing resources C available in the system.

[0055] Furthermore, the penalty function method is used to deal with the constraints of the network slice resource allocation model; including:

[0056] Execute the dynamic penalty function coefficient strategy, the penalty function factor is:

[0057]

[0058] Among them, ε(t) is the penalty function factor, ε min is the minimum value of the penalty function factor, ε max is the maximum value of the penalty function factor, t represents the current number of iterations, and T represents the maximum number of iterations;

[0059] The penalty function Φ(x) is defined as:

[0060]

[0061] Among them, constraint s (x) is used to determine whether constraints are violated or satisfied:

[0062] Used to determine whether constraint 1 is violated or satisfied;

[0063] Used to determine whether constraint 2 is violated or satisfied;

[0064] Used to determine whether constraint 3 is violated or satisfied;

[0065] Used to determine whether constraint 4 is violated or satisfied;

[0066] Used to determine whether constraint 5 is violated or satisfied;

[0067] Used to determine whether constraint 6 is violated or satisfied;

[0068] Used to determine whether constraint 7 is violated or satisfied;

[0069] Used to determine whether constraint conditions are violated or satisfied 8;

[0070] If constraint s (x) returns a non-negative value, indicating a constraint violation. s (x) If the return value is 0, it means that the constraint is satisfied;

[0071] Then the network slice resource allocation optimization problem of the network slice resource allocation model is expressed as:

[0072] Min(Φ(x)-Max(β C U C +β B U B ))(1.12)

[0073] Furthermore, the network slice resource allocation model after constraint processing is solved based on the double tanh early stopping genetic algorithm, including:

[0074] Calculate the crossover rate p of the double tanh function early stopping genetic algorithm crossoperate (t):

[0075]

[0076] Among them, p crossoperate represents the crossover rate at the iteration number t, k1 and k2 represent the slope control constants of the two tanh functions, which are used to control the steepness of the two tanh functions, and p crossoperate_max Indicates the maximum crossover rate, p crossoperate_min Indicates the minimum crossover rate;

[0077] Calculate the mutation rate p of the double tanh function early stopping genetic algorithm variation (t):

[0078]

[0079] Among them, p variation represents the mutation rate at the iteration number t, k3 and k4 represent the slope control constants of the two tanh functions, which are used to control the steepness of the two tanh functions, and p variation_max Indicates the maximum mutation rate, p variation_min Indicates the minimum mutation rate;

[0080] The solution is run using a modified double tanh early stopping genetic algorithm, including:

[0081] Initialize the algorithm parameters. Initialize the gene length of the population individuals to be the same as the product of the number of network slice resource blocks and the number of terminals, both H×I. Each individual in the population corresponds to a slice resource allocation scheme. Set the maximum number of iterations T, the global best performing individual (GBPI), the best performing individual (BPI) of the current generation, set the population size x, and set the maximum crossover rate p. crossoperate_max and the minimum crossover rate p crossoperate_min , set the tanh function slope control constants k1, k2, and set the maximum mutation rate p variation_max , minimum mutation rate p variation_min , set the tanh function slope control constants k3 and k4;

[0082] Randomly generate individuals in the population, each individual corresponds to a resource allocation scheme, and initialize the global best performing individual and the best phenotype individual of the current generation;

[0083] Iterate all individuals in a loop, substitute the corresponding parameters of each individual into the objective function for calculation, and use the result as the fitness of the current individual. Then update the individual and select the best phenotype individual in the current generation based on the fitness.

[0084] Each individual modifies its configuration parameters through mutation and exchange to form a new generation of offspring. Mutation means randomly changing a configuration parameter of an individual, and exchange means exchanging a random number of configuration parameters between two individuals. After the parameter modification is completed, the fitness of the offspring is calculated. The individual with the lowest fitness of the offspring is replaced with the individual with the best phenotype in the current generation, and then the mutation rate and crossover rate are updated.

[0085] The global best phenotype individual is updated. Whether to update the global best phenotype individual is determined by evaluating the global optimal solution of the current iteration. If the best phenotype individual of the current iteration is better than the global best phenotype individual, the global best phenotype individual is replaced by the best phenotype individual of the current iteration.

[0086] The total number of iterations is recorded in real time. When the iteration number t reaches 85% of the preset iteration number, that is, the current iteration number is greater than or equal to 0.85×T, the calculation of the early stopping strategy begins: if it continues after 0.05×T, the double tanh early stopping genetic algorithm is stopped in advance; after the iteration is completed, the network slice resource allocation plan corresponding to the global best phenotype individual is used as the network slice resource allocation result.

[0087] The beneficial effects achieved by the present invention are:

[0088] The present invention aims at the resource allocation problem of smart grid network slices, combines network slicing technology, takes into account the different requirements of different service bandwidths, delays and transmission rates, and aims to maximize network utility, and constructs a utility maximization resource allocation method for smart grid network slices. A double tanh early stopping genetic algorithm is proposed, which enhances the global search capability of the algorithm, effectively solves the problem that traditional genetic algorithms are prone to fall into local optimal solutions and lose the previous generation of better solutions, achieves the quality of the optimal solution obtained under the same algorithm operation time, and avoids excessive useless time waste by introducing an early stopping strategy, thereby bringing higher network utility to the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 This is a schematic diagram of the application scenario of the network slicing resource allocation model;

[0090] Figure 2 Establish a flow chart for the network slicing resource allocation model;

[0091] Figure 3 This is the flow chart for solving the double tanh early stopping genetic algorithm. DETAILED DESCRIPTION

[0092] The present invention is further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical method of the present invention, and cannot be used to limit the protection scope of the present invention.

[0093] Example 1

[0094] A utility maximization resource allocation method for smart grid network slicing, comprising:

[0095] Obtain resource requirement parameters for different services and the amount of resources owned by different resource blocks;

[0096] According to the resource demand parameters of different services and the resource amounts of different resource blocks, a network slice resource allocation model is established with the goal of maximizing network utility;

[0097] The model constructed using the penalty function method handles the constraints of the network slice resource allocation model;

[0098] The early stopping genetic algorithm improved based on the tanh function solves the network slice resource allocation model after the constraints are processed, and outputs the optimal network slice resource allocation solution;

[0099] Resource blocks are allocated to the grid terminals served by the network slice according to the optimal network slice resource allocation scheme.

[0100] Example 2

[0101] The utility maximization resource allocation method for smart grid network slicing according to Embodiment 1 is different in that:

[0102] like Figure 1 As shown, the present invention considers a smart grid scenario located in a residential community or office building. The scenario is covered by N different base stations, there are I terminals in the area, and different terminal service needs in the area are served by different network slices. The number of network slice resource blocks is H, such as Figure 1 As shown in the figure, service terminals, resource blocks, slices, base stations and control centers constitute the smart grid system. Network slicing defines a logical partition, each of which has its own specific performance and service requirements. Base stations are one of the physical infrastructures for implementing network slicing. Each base station can support multiple network slices and dynamically allocate wireless resources to each slice based on different service requirements. For example, a base station can simultaneously support low-energy collection service network slices and low-latency control service network slices. Resource blocks are specific resources allocated by base stations to network slices. Base stations allocate physical resource blocks to different slices based on the needs of each network slice. For example, for drone inspection service slices that require high bandwidth, the base station will allocate more resource blocks, while for precise load control services with low latency requirements, the base station will optimize resource allocation to ensure real-time and reliability. Resource blocks are finally allocated to specific users by network slices.

[0103] Resource requirement parameters for different power grid services, including:

[0104] Assuming that each terminal corresponds to a terminal service, represents the energy consumed by the i-th terminal service, represents the delay of the i-th terminal service, Represents the data transmission rate required by the i-th terminal service.

[0105] According to the resource demand parameters of different services and the amount of resources owned by different resource blocks, a network slice resource allocation model is established with the goal of maximizing network utility; Figure 2 As shown, including:

[0106] Assume that the smart grid area is covered by N different base stations, there are I terminals in the area, and the different needs of terminal services in the area are served by different network slices. The number of network slice resource blocks is H.

[0107] In the smart grid communication process, orthogonal frequency division multiplexing technology is used to allocate orthogonal frequencies to terminals connected to the same base station. Therefore, there is interference between terminals connected to different base stations. Therefore, the definition of base station g n Received from terminal b i Signal-to-interference-noise ratio for:

[0108]

[0109] in, From terminal b i Transmit to base station g n The transmission power, and Respectively represent terminal b i and terminal b k With base station g n The channel gain between , σ represents the additive white Gaussian noise.

[0110] Channel Gain The channel model of Rayleigh fading is used for calculation, and the specific calculation formula is defined as:

[0111]

[0112] in, is the antenna gain, d e is the path attenuation exponent, is the carrier frequency, d i,n d k,l is the distance between the terminal and the base station, Obeys Rayleigh distribution;

[0113] Generally speaking, when a terminal is allocated more network resources, its transmission rate will increase accordingly. It is a following After increasing the parameter and obtaining the signal-to-interference-noise ratio, the terminal b is obtained by the Shannon formula i Data transfer rate

[0114]

[0115] in, For terminal b i The allocated bandwidth; Indicates that at base station g n Received from terminal b i signal-to-interference-noise ratio.

[0116] Orthogonal frequency division multiplexing technology is used in the communication process of smart grids. Orthogonal frequencies are allocated to terminals connected to the same base station. Therefore, there is interference between terminals connected to different base stations, which is the signal-to-interference-noise ratio. The definition is as follows:

[0117]

[0118] in, From terminal b i Transmit to base station g n The transmission power, and Respectively represent terminal b i and terminal b k With base station g n The channel gain between σ and σ represents the power of additive white Gaussian noise. The transmit power and channel gain are related to the distance between the terminal and the base station and the environment. They change according to the actual situation of the terminal, thus realizing dynamic adaptation of network resource allocation.

[0119] Generally speaking, the total delay is composed of two parts: one is the time spent by the terminal to process the service, and the other is the transmission time of the uploaded data. The relationship function between the delay and the transmission rate is defined. Assume that the terminal b i The size of the packet to be transmitted is Terminal b i The amount of computation required is This gives terminal b i Delay

[0120]

[0121] The size of the transmitted data packet is Terminal b in the process i The amount of computation required, For terminal b i Allocated computing resources;

[0122] During the communication between the terminal and the base station, the main energy consumption comes from the transmission power consumption of the terminal signal. i Energy consumption of transmitting data packets for:

[0123]

[0124] Considering that the allocation of slice resources is a two-way operation process between the power terminal and the control center, the utility function includes the utility of the power terminal and the utility of the control center;

[0125] The three application scenarios in the 5G era are eMBB, uRLLC and mMTC. For these three scenarios, the utility function U of the power terminal side is designed. B for:

[0126]

[0127] in, and are the penalty factors when the actual energy consumption, transmission rate, bandwidth, and delay cannot reach the optimal level. and Optimal energy consumption, transmission rate, bandwidth, and latency for terminal services. and The energy consumption, transmission rate, bandwidth, and latency of the terminal after the resource block is actually allocated to the terminal; modify the resource block according to the different application scenarios of the service. and To increase or decrease the weight of different indicators in the utility function, and then realize the division of different application scenarios.

[0128] The control center needs to ensure that as much bandwidth and computing resources as possible are allocated to avoid idle resources. Therefore, the utility function U of the control center is C for:

[0129]

[0130] in, represents the allocation relationship between terminals and resource blocks, α W , α c is the benefit of the allocated resource block unit bandwidth and computing resources, is the resource block operation cost; α W , α c It needs to be a positive number to ensure that the control center allocates resource blocks with positive returns.

[0131] Considering that the allocation of slice resources is a two-way operation process between the power terminal and the control center, the terminal hopes to obtain sufficient resources for the normal operation of the business, and the control center hopes to allocate as many resources as possible to avoid idleness. Therefore, the utility function is composed of the power terminal utility U B U with Control Center Utility C It consists of two parts. Then, the objective function of the network slice resource allocation model with the goal of maximizing network utility is expressed as:

[0132] U=Max(β C U C +β B U B )(1.9)

[0133] Among them, β C With β B are the weights placed by the control center and the terminal in the system utility respectively. C With β B , you can change the dominant object. It is mainly to achieve high-value terminal business and improve β B To improve the priority of meeting terminal business needs; or to focus on higher resource utilization and improve β C To increase the need to allocate more idle resources.

[0134] The constraints of the network slicing resource allocation model include:

[0135] Constraint 1: Indicates the transmission speed after the terminal is allocated a resource block Must be greater than the minimum transmission speed tolerated by the business

[0136] Constraint 2: Indicates the bandwidth after the terminal is allocated to the resource block Must be greater than the minimum bandwidth tolerated by the service

[0137] Constraint 3: Indicates the energy consumption of the terminal after being allocated to the resource block Must be less than the maximum energy consumption tolerated by the business

[0138] Constraint 4: Indicates the delay after the terminal is allocated a resource block Must be less than the maximum delay tolerated by the service

[0139] Constraint 5: Indicates that each resource block can only be allocated to one terminal;

[0140] Constraint 6: Indicates that each terminal can only be allocated one resource block;

[0141] Constraint 7: Indicates that the bandwidth resources allocated to all terminals do not exceed all available bandwidth resources W in the system;

[0142] Constraint 8: It means that the computing resources allocated to all terminals do not exceed all the computing resources C available in the system.

[0143] The establishment of the above mathematical model has the following characteristics:

[0144] 1. Comprehensively consider differentiated needs: Different services may have different latency and transmission rate requirements, and different smart grid services have different priorities. For example, emergency power distribution adjustment may have a higher priority.

[0145] 2. Improve fairness: Maximizing the weighted sum of service priorities can improve the fairness of resource allocation. Terminal services with higher priorities can receive more attention in resource allocation, thus avoiding being ignored due to low demand for resources.

[0146] 3. Considering the influence of possible noise between terminals connected to different base stations when using OFDM, by introducing the signal-to-interference-noise ratio To calculate parameters such as energy consumption and delay, it is more in line with the scenario of information transmission between terminals and base stations in reality. The Rayleigh fading channel model is used for channel gain, which is consistent with the dense urban environment involved in this invention.

[0147] Using the penalty function method to handle the constraints of the network slice resource allocation model: Selecting the appropriate penalty function coefficient is often a big challenge, because this value is highly dependent on the specific optimization problem and is not easy to determine. If the penalty function coefficient is too large, it is easy to cause the early evolution direction of the population to be too fixed, which seriously affects the search range. If the penalty function coefficient is too small, the solution obtained is likely to fail to meet the constraints. Including:

[0148] The present invention adopts a dynamic penalty function coefficient strategy: the penalty function coefficient is adjusted through the evolution process of the population. In the early stage of evolution, the penalty function coefficient is small, which can effectively promote the exploration ability of the population. In the late stage of evolution, the individual phenotypes of the population are relatively fixed, and increasing the penalty function coefficient can enhance the utilization ability and convergence speed of the population. In this way, a balance between exploration and utilization is achieved. When the dynamic penalty function coefficient strategy is implemented, the penalty function factor is:

[0149]

[0150] Among them, ε(t) is the penalty function factor, ε min is the minimum value of the penalty function factor, ranging from [0.001, 1], ε max is the maximum value of the penalty function factor, ranging from [100, 10000], t represents the current number of iterations, and T represents the maximum number of iterations;

[0151] The penalty function Φ(x) is defined as:

[0152]

[0153] Among them, constraint s (x) is used to determine whether constraints are violated or satisfied:

[0154] Used to determine whether constraint 1 is violated or satisfied;

[0155] Used to determine whether constraint 2 is violated or satisfied;

[0156] Used to determine whether constraint 3 is violated or satisfied;

[0157] Used to determine whether constraint 4 is violated or satisfied;

[0158] Used to determine whether constraint 5 is violated or satisfied;

[0159] Used to determine whether constraint 6 is violated or satisfied;

[0160] Used to determine whether constraint 7 is violated or satisfied;

[0161] Used to determine whether constraint conditions are violated or satisfied 8;

[0162] If constraint s (x) returns a non-negative value, indicating a constraint violation. s (x) If the return value is 0, it means that the constraint is satisfied;

[0163] Since the objective function of the present invention is to maximize the network utility, after the penalty function method is used for the factor, the objective condition should take the minimum value, then the network slice resource allocation optimization problem of the network slice resource allocation model is expressed as:

[0164] Min(Φ(x)-Max(β C U C +β B UB ))(1.12)

[0165] The network slice resource allocation model after constraint processing is solved based on the double tanh early stopping genetic algorithm, such as Figure 3 As shown, including:

[0166] Tanh function is smoother than traditional linear function, and can provide more suitable transition parameters in the process of converting global search to local search, thereby effectively improving the convergence speed of the algorithm and the quality of the solution, thereby solving the problem of genetic algorithm falling into local optimum too early. However, tanh function may cause parameters to change too quickly during iteration. Based on this, the present invention proposes a double tanh function to optimize inertia weight.

[0167] Calculate the crossover rate p of the double tanh function early stopping genetic algorithm crossoperate (t):

[0168]

[0169] Among them, p crossoperate represents the crossover rate at the iteration number t, k1 and k2 represent the slope control constants of the two tanh functions, which are used to control the steepness of the two tanh functions, and p crossoperate_max Indicates the maximum crossover rate, p crossoperate_min Represents the minimum crossover rate; the first tanh function can be set to be relatively gentle to encourage global search. This helps the algorithm to explore the solution space more widely and avoid falling into local optimality too early. In the later stages, the steepness of the second Tanh function can be increased to encourage more detailed search (local search). This helps the algorithm to explore the area more deeply after finding a potential good area.

[0170] Calculate the mutation rate p of the double tanh function early stopping genetic algorithm variation (t):

[0171]

[0172] Among them, p variation represents the mutation rate at the iteration number t, k3 and k4 represent the slope control constants of the two tanh functions, usually between [0.1, 2], and are used to control the steepness of the two tanh functions. variation_max Indicates the maximum mutation rate, p variation_min Indicates the minimum mutation rate.

[0173] Change the population composition: According to the mutation rate, randomly modify the genotype of each individual in the population, and each 01 position has p variationThen, according to the crossover rate, the partial expression of the genotype of any two individuals is randomly exchanged, and each 01 position has p crossoperate The probability of (t) is to exchange the 01 bit in the same position with that of another individual.

[0174] Algorithm elite strategy: After each round of evolution, the fitness of the population is calculated, and the individual with the highest fitness is selected and registered as the best phenotype individual BPI of the current generation and compared with the global best phenotype individual GBPI. If the fitness is higher than GBPI, GBPI is replaced with the individual with the highest fitness. At the same time, after the next round of evolution, the individual with the lowest fitness in the current population is selected and replaced with BPI first, so as to eliminate the offspring that do not meet the requirements and realize pruning, and then repeat the process of S2.

[0175] Early stopping strategy: When the number of iterations t reaches the maximum of 0.85×T, the early stopping strategy calculation begins. If the global best phenotype individual GBPI remains unchanged after 0.05×T, the double tanh genetic algorithm is stopped in advance, and the GBPI at this time is output as the best allocation plan to avoid unnecessary waste of solution time.

[0176] The solution is run using a modified double tanh early stopping genetic algorithm, including:

[0177] Initialize the algorithm parameters. Initialize the gene length of the population individuals to be the same as the product of the number of network slice resource blocks and the number of terminals, both H×I. Each individual in the population corresponds to a slice resource allocation scheme. Set the maximum number of iterations T, the global best phenotype individual GBPI, the current generation best phenotype individual BPI, set the population size x, and set the maximum crossover rate p crossoperate_max and the minimum crossover rate p crossoperate_min , set the tanh function slope control constants k1, k2, and set the maximum mutation rate p variation_max , minimum mutation rate p variation_min , set the tanh function slope control constants k3 and k4;

[0178] Randomly generate individuals in the population, each individual corresponds to a resource allocation scheme, and initialize the global best performing individual and the best phenotype individual of the current generation;

[0179] Iterate all individuals in a loop, substitute the corresponding parameters of each individual into the objective function for calculation, and use the result as the fitness of the current individual. Then update the individual and select the best phenotype individual in the current generation based on the fitness.

[0180] Each individual modifies its configuration parameters through mutation and exchange to form a new generation of offspring. Mutation means randomly changing a configuration parameter of an individual, and exchange means exchanging a random number of configuration parameters between two individuals. After the parameter modification is completed, the fitness of the offspring is calculated. The individual with the lowest fitness of the offspring is replaced with the individual with the best phenotype in the current generation, and then the mutation rate and crossover rate are updated.

[0181] The global best phenotype individual is updated. Whether to update the global best phenotype individual is determined by evaluating the global optimal solution of the current iteration. If the best phenotype individual of the current iteration is better than the global best phenotype individual, the global best phenotype individual is replaced by the best phenotype individual of the current iteration.

[0182] The total number of iterations is recorded in real time. When the iteration number t reaches 85% of the preset iteration number, that is, the current iteration number is greater than or equal to 0.85×T, the calculation of the early stopping strategy begins: if after 0.05×T, the global best phenotype individual GBPI is not replaced or remains unchanged, it means that the model result is close to or reaches the global optimal solution, and there is no need to spend extra time to solve it, then the double tanh early stopping genetic algorithm is stopped in advance; after the iteration, the network slice resource allocation plan corresponding to the global best phenotype individual is used as the network slice resource allocation result.

Claims

1. A utility maximization resource allocation method for smart grid network slicing, characterized in that: include: Obtain resource requirement parameters for different services and the amount of resources owned by different resource blocks; According to the resource demand parameters of different services and the resource amounts of different resource blocks, a network slice resource allocation model is established with the goal of maximizing network utility; The model constructed using the penalty function method handles the constraints of the network slice resource allocation model; The early stopping genetic algorithm improved based on the tanh function solves the network slice resource allocation model after the constraints are processed, and outputs the optimal network slice resource allocation solution; Resource blocks are allocated to the grid terminals served by the network slice according to the optimal network slice resource allocation scheme.

2. The utility maximization resource allocation method for smart grid network slicing according to claim 1, characterized in that: Resource requirement parameters for different power grid services, including: Assuming that each terminal corresponds to a terminal service, represents the energy consumed by the i-th terminal service, represents the delay of the i-th terminal service, Represents the data transmission rate required by the i-th terminal service.

3. The utility maximization resource allocation method for smart grid network slicing according to claim 1 is characterized in that: According to the resource demand parameters of different services and the resource amounts of different resource blocks, a network slice resource allocation model is established with the goal of maximizing network utility; including: Assume that the smart grid area is covered by N different base stations, there are I terminals in the area, and different terminal service needs in the area are served by different network slices. The number of network slice resource blocks is H; Defined in base station g n Received from terminal b i Signal-to-interference-noise ratio for: in, From terminal b i Transmit to base station g n The transmission power, and Respectively represent terminal b i and terminal b k With base station g n The channel gain between , σ represents additive Gaussian white noise; Channel Gain The calculation formula is defined as: in, is the antenna gain, d e is the path attenuation exponent, is the carrier frequency, d i,n ,d k,l is the distance between the terminal and the base station, Obeys Rayleigh distribution; After obtaining the signal-to-interference-to-noise ratio, the terminal b is obtained by the Shannon formula i The data transmission rate r bi : in, For terminal b i The allocated bandwidth; Assume terminal b i The size of the packet to be transmitted is Terminal b i The amount of computation required is This gives terminal b i Delay in, For terminal b i Allocated computing resources; Then we get terminal b i Energy consumption of transmitting data packets for: The utility function includes the utility of the power terminal and the utility of the control center; Power terminal side utility function U B for: in, and are the penalty factors when the actual energy consumption, transmission rate, bandwidth, and delay cannot reach the optimal level. and Optimal energy consumption, transmission rate, bandwidth, and latency for terminal services. and The energy consumption, transmission rate, bandwidth, and latency of the terminal after it is actually allocated resource blocks; The utility function U of the control center C for: in, represents the allocation relationship between terminals and resource blocks, α W , α c is the benefit of the allocated resource block unit bandwidth and computing resources, is the resource block operation cost; Then, the objective function of the network slice resource allocation model with the goal of maximizing network utility is expressed as: U=Max(β C U C +b B U B (1.9) Among them, β C With β B are the weights that the control center and the terminal place in the system utility respectively.

4. The utility maximization resource allocation method for smart grid network slicing according to claim 1, characterized in that: The constraints of the network slice resource allocation model include: Constraint 1: Indicates the transmission speed after the terminal is allocated a resource block Must be greater than the minimum transmission speed tolerated by the business Constraint 2: Indicates the bandwidth after the terminal is allocated to the resource block Must be greater than the minimum bandwidth tolerated by the service Constraint 3: Indicates the energy consumption of the terminal after being allocated to the resource block Must be less than the maximum energy consumption tolerated by the business Constraint 4: Indicates the delay after the terminal is allocated a resource block Must be less than the maximum delay tolerated by the service Constraint 5: Indicates that each resource block can only be allocated to one terminal; Constraint 6: Indicates that each terminal can only be allocated one resource block; Constraint 7: b i ∈B, indicating that the bandwidth resources allocated to all terminals do not exceed all available bandwidth resources W in the system; Constraint 8: b i ∈B, which means that the computing resources allocated to all terminals do not exceed all available computing resources C in the system.

5. The utility maximization resource allocation method for smart grid network slicing according to claim 4 is characterized in that: The penalty function method is used to deal with the constraints of the network slice resource allocation model, including: Execute the dynamic penalty function coefficient strategy, the penalty function factor is: Among them, ε(t) is the penalty function factor, ε min is the minimum value of the penalty function factor, ε max is the maximum value of the penalty function factor, t represents the current number of iterations, and T represents the maximum number of iterations; The penalty function Φ(x) is defined as: Among them, constraint s (x) is used to determine whether constraints are violated or satisfied: Used to determine whether constraint 1 is violated or satisfied; Used to determine whether constraint 2 is violated or satisfied; Used to determine whether constraint 3 is violated or satisfied; Used to determine whether constraint 4 is violated or satisfied; Used to determine whether constraint 5 is violated or satisfied; Used to determine whether constraint 6 is violated or satisfied; Used to determine whether constraint 7 is violated or satisfied; Used to determine whether constraint conditions are violated or satisfied 8; If constraint s (x) returns a non-negative value, indicating a constraint violation. s (x) If the return value is 0, it means that the constraint is satisfied; Then the network slice resource allocation optimization problem of the network slice resource allocation model is expressed as: Min(Φ(x)-Max(β C U C +b B U B ))(1.12) 6. A utility maximization resource allocation method for smart grid network slicing according to any one of claims 1 to 5, characterized in that: The network slice resource allocation model after constraint processing is solved based on the double tanh early stopping genetic algorithm, including: Calculate the crossover rate p of the double tanh function early stopping genetic algorithm crossoperate (t): Among them, p crossoperate represents the crossover rate at the iteration number t, k1 and k2 represent the slope control constants of the two tanh functions, which are used to control the steepness of the two tanh functions, and p crossoperate_max Indicates the maximum crossover rate, p crossoperate_min Indicates the minimum crossover rate; Calculate the mutation rate p of the double tanh function early stopping genetic algorithm variation (t): Among them, p variation represents the mutation rate at the iteration number t, k3 and k4 represent the slope control constants of the two tanh functions, which are used to control the steepness of the two tanh functions, and p variation_max Indicates the maximum mutation rate, p variation_min Indicates the minimum mutation rate.

7. The utility maximization resource allocation method for smart grid network slicing according to claim 6, characterized in that: The solution is run using a modified double tanh early stopping genetic algorithm, including: Initialize the algorithm parameters. Initialize the gene length of the population individuals to be the same as the product of the number of network slice resource blocks and the number of terminals, both H×I. Each individual in the population corresponds to a slice resource allocation scheme. Set the maximum number of iterations T, the global best phenotype individual GBPI, the current generation best phenotype individual BPI, set the population size x, and set the maximum crossover rate p crossoperate_max and the minimum crossover rate p crossoperate_min , set the tanh function slope control constants k1, k2, and set the maximum mutation rate p variation_max , minimum mutation rate p variation_min , set the tanh function slope control constants k3 and k4; Randomly generate individuals in the population, each individual corresponds to a resource allocation scheme, and initialize the global best performing individual and the best phenotype individual of the current generation; Iterate all individuals in a loop, substitute the corresponding parameters of each individual into the objective function for calculation, and use the result as the fitness of the current individual. Then update the individual and select the best phenotype individual in the current generation based on the fitness. Each individual modifies its configuration parameters through mutation and exchange to form a new generation of offspring. Mutation means randomly changing a configuration parameter of an individual, and exchange means exchanging a random number of configuration parameters between two individuals. After the parameter modification is completed, the fitness of the offspring is calculated. The individual with the lowest fitness of the offspring is replaced with the individual with the best phenotype in the current generation, and then the mutation rate and crossover rate are updated. The global best phenotype individual is updated. Whether to update the global best phenotype individual is determined by evaluating the global optimal solution of the current iteration. If the best phenotype individual of the current iteration is better than the global best phenotype individual, the global best phenotype individual is replaced by the best phenotype individual of the current iteration. The total number of iterations is recorded in real time. When the iteration number t reaches 85% of the preset iteration number, that is, the current iteration number is greater than or equal to 0.85×T, the calculation of the early stopping strategy begins: if it continues after 0.05×T, the double tanh early stopping genetic algorithm is stopped in advance; after the iteration is completed, the network slice resource allocation plan corresponding to the global best phenotype individual is used as the network slice resource allocation result.

Images