A mobile base station dynamic site selection optimization method targeting service reliability

Abstract

The application relates to a mobile base station dynamic site selection optimization method aiming at service reliability, and the steps comprise the following: constructing a simple model of a planning area and carrying out initial data collection; obtaining a propagation model of a fixed position base station by using a neural network method based on physical information, and obtaining an agent channel model by fusing multiple propagation models; taking user service reliability as an adaptability function of a genetic algorithm and carrying out site selection planning; when personnel in the area are transferred and the service reliability is decreased, the current optimal site selection scheme is given by re-planning. The method combines physical information and measured data, considers accuracy and efficiency, considers multiple uncertain factors influencing user service quality, and can give a site selection scheme with more practical reference significance.

Description

A Dynamic Location Optimization Method for Mobile Base Stations with Service Reliability as the Objective Technical Field

[0001] This invention relates to the field of mobile base station site selection technology, particularly A mobile base station with service reliability as its objective Dynamic location optimization method . Background Technology

[0002] A mobile base station is a communication base station equipment that can be quickly assembled, disassembled, and moved. It typically includes components such as antennas, transceivers, transmission equipment, and power supplies, and can be deployed to areas requiring communication support in a short time to provide temporary or emergency communication services. Mobile base stations are usually composed of vehicle-mounted or drone-based equipment, allowing for rapid relocation and deployment as needed, making them suitable for various special situations, such as natural disasters, major events, and military operations. With the support of mobile base stations, network services can be quickly restored, improving communication capabilities and efficiency. In special circumstances, mobile base stations are a reliable and effective communication solution; however, compared to fixed base stations, mobile base stations generally have higher failure rates and less efficient coverage calculations.

[0003] In engineering, the site selection scheme for base stations is usually planned as follows: First, the maximum allowable path loss is calculated based on equipment information and environmental information. Then, the effective coverage radius of the base station is calculated using a wireless propagation model, and the coverage area is obtained by drawing a circle. Finally, the location of the base station is planned according to different optimization methods.

[0004] Wireless propagation models used to predict the effective coverage range of base stations fall into two categories: the first is an empirical model composed of a large amount of manually measured data combined with certain correction coefficients. Empirical models have strong generalization capabilities but ignore environmental diversity and cannot adapt to complex and ever-changing application scenarios. The second is a deterministic model that calculates the electromagnetic field distribution by solving Maxwell's equations mathematically based on electromagnetic wave theory. This type of model has high accuracy but is computationally expensive. In mobile base station site selection applications, because the effective coverage range of the base station needs to be constantly recalculated, deterministic models based on numerical calculations or simulation experiments are too time-consuming. In practice, empirical models are usually used for rapid calculation, but this results in a certain loss of accuracy.

[0005] In base station site selection optimization methods, existing methods usually only consider user distribution and user coverage indicators, lacking consideration of dynamic scenarios and user service quality, making it difficult to meet the changing user distribution and task transfer requirements in mobile base station deployment scenarios.

[0006] Existing wireless propagation models cannot balance accuracy and efficiency, making them unsuitable for dynamic site selection optimization of mobile base stations. Site selection optimization methods for mobile base stations lack consideration of uncertainties.

[0007] Therefore, it is necessary to present a dynamic location optimization method for mobile base stations with service reliability as the objective. Summary of the Invention

[0008] The purpose of this invention is to address the problems existing in current mobile base station site selection optimization methods and propose a dynamic site selection optimization method for mobile base stations with service reliability as the objective. This method utilizes a neural network based on physical information to obtain the effective coverage area of ​​the base station, and with user service reliability as the optimization objective, it can efficiently find the optimal site selection scheme while maintaining a certain level of accuracy.

[0009] Step 1: Scene setup for the planning area. The size, dimensions, shape, number, and location of buildings within the planning area need to be determined based on the problem requirements.

[0010] Step 2: Initial Data Acquisition. This requires collecting electric field intensity information at different locations within the service area of ​​a single base station. This can be done through on-site data collection or simulation experiments.

[0011] Step 3: Obtain the propagation model of a fixed-location base station in the area using a physical information-based neural network method. Using initial data and base station transmitted signals as training samples, and Maxwell's equations as physical constraint equations, a neural network is constructed. A hard constraint method is used to control the boundary conditions and reflection conditions of the buildings. The network inputs spatial coordinates x and y and time t, and outputs the electric field intensity at the corresponding location and time in the service area.

[0012] Step 4: Integrate multiple propagation models to obtain proxy channel models for base stations at different locations within the service area. Use training data generated from multiple neural networks constructed in Step 3 to train the neural networks. This network can directly input base station locations to obtain the electric field intensity information generated by the corresponding base station within the service area.

[0013] Step 5: User Service Reliability Calculation. Using the Monte Carlo method, this step comprehensively considers factors such as whether the user is covered by a base station, the level of signal interference at the user's location, the capacity of the communication base station, the user's maximum allowable downtime, and the signal transmission distance. Under conditions of random fluctuations in base station signal strength and load, the probability that the user can successfully obtain stable network service within the specified time is calculated.

[0014] Step 6: Genetic Algorithm Searches for the Optimal Site Selection Scheme. Using user service reliability as the optimization objective, the base station location information is encoded, and a genetic algorithm is used to search for the optimal base station deployment scheme within a specified number of base stations in the service area.

[0015] Step 7: Dynamic planning based on service reliability. After providing a site selection plan, continuously monitor the service reliability within the area. When the service reliability falls below a threshold, invoke the genetic algorithm to replan based on changes in the scenario.

[0016] Through the above steps, a method for dynamic location optimization of mobile base stations with service reliability as the objective is presented.

[0017] The superior performance of this method is as follows: In the field of mobile base station site selection technology, a neural network method based on physical information is used for electromagnetic field modeling to predict electromagnetic data around the base station, thereby obtaining the corresponding wireless propagation model and further calculating the effective coverage area of ​​the base station. While maintaining a certain level of accuracy, it avoids the high cost of repeatedly performing full-domain simulation calculations. Furthermore, this model has better interpretability and higher accuracy than empirical models and data-driven neural network models. In the mobile base station site selection optimization strategy, the randomness of signal fluctuations and user location changes within the service area is considered. With service reliability as the objective, the site selection optimization problem is guided, and the proposed site selection scheme has greater practical reference value. Attached Figure Description

[0018] Figure 1 shows the logic of the dynamic location optimization method for mobile base stations with service reliability as the objective.

[0019] Figure 2. Actual topographic map (left) and simplified topographic map (right)

[0020] Figure 3 Electromagnetic wave situation at the intermediate moment

[0021] Figure 4 Service Reliability Calculation Logic

[0022] Figure 5. Location selection scheme output by the genetic algorithm

[0023] Figure 6. Service reliability variation graph

[0024] Figure 7 Adjusted Site Selection Scheme Detailed Implementation

[0025] To provide a clearer understanding of the features and advantages of the present invention, a detailed description is provided below in conjunction with examples and accompanying drawings:

[0026] Step 1: Taking the terrain around Incirli, 15 km from the epicenter of the February 6th Turkish earthquake, as an example, this area has multiple mountain ranges. Establishing fixed base stations after the earthquake is difficult and time-consuming; therefore, a mobile base station strategy is suitable for restoring regional communication capabilities. As shown in Figure 2, this is simplified to a two-dimensional problem. Assume that five mountain areas completely reflect electromagnetic waves, and that personnel and base station locations are not included within these mountain areas. Let x and y represent the coordinates in the two directions, and for time-varying electromagnetic fields, there is also a time dimension t; the boundary of this region adopts a PML complete absorption boundary condition. There are three densely populated town areas within the region (the coordinates of the dots in Figure 2).

[0027] Step 2: Initial data sampling was conducted by placing base stations around the three town centers. Multiple sampling points were set up throughout the area to record the change of electric field strength over time. In this example, a simulation experiment was used to obtain the initial data from the above nine base station locations. The simulation experiment used Matlab software and the finite-difference time-domain method (FDTD) to calculate the field strength at each location. The simulation image at a certain intermediate moment is shown in the left image of Figure 3.

[0028] The base station transmits a stable sinusoidal plane wave, and the sinusoidal signal satisfies the following formula:

[0029] e=E0sin(2πftdt) (1)

[0030] In this simulation, E0 is set to 1V, f is the signal frequency, t is the number of time iterations (400 in total), dt is the time interval satisfying the stability condition (dt = dx / 2c0), dx is the simulation grid width (one-fortieth of the signal wavelength), and c0 is the speed of light. The grid size is set to 120×120, the grid width is dx, the PML layer thickness is 10, and the electromagnetic wave transmittance of the building area is set to 0. The FDTD method iteratively solves the problem based on the initial field values ​​and the signal source, performing simulations for 400 time intervals. This yields the Ez, Hx, and Hy field values ​​for all grid points within the problem domain at 400 time points. Points within the PML layer on the boundary are removed, and one point is retained for every five points in the x and y directions. Therefore, each set of data contains 20×20×400 = 16000 data points.

[0031] Step 3: Construct a neural network model based on physical information. Optionally, the model uses a fully connected neural network with three input variables: coordinates x, y, and time t, and three output variables: Ez, Hx, and Hy. Optionally, the neural network is set to 6 layers with a structure of [3, 50, 100, 100, 50, 3], and four hidden layers with a width of 50 or 100. The training data is one set from the nine initial data sets.

[0032] Optionally, the hyperbolic tangent activation function Tanh is used. W and b are neural network structure parameters.

[0033] The loss function of the PINNs neural network model is defined as a weighted sum of the following classes: boundary conditions L BC Initial condition L IC Sampling true data L data Partial differential equation L PDE Since the PML fully absorbing boundary condition is used, no additional constraints are needed at the region boundary. Therefore, this problem does not require setting boundary condition L. BC Initial conditions L IC The signal data is obtained directly from the signal formula. The simulation data obtained in step 2 is regarded as the sampled true data L. data .

[0034] When calculating the electromagnetic field value, the partial differential equations are Maxwell's equations:

[0035]

[0036]

[0037] Where E is the electric field strength, in V / m; H is the magnetic field strength, in A / m; and D is the electric flux density, in C / m. 2 B represents magnetic flux density, in units of Wb / m. 2 J represents current density, in A / m. 2 J m Magnetic flux density, unit V / m 2 .

[0038] Discretizing the above equations using coordinates yields the following formula:

[0039]

[0040]

[0041]

[0042] Where ε is the dielectric constant of the medium, and μ is the magnetic permeability. The above three equations are the formulas for calculating L in this problem. PDE The formula.

[0043] Optionally, during neural network training, the parameters of the neural network are randomly initialized, and the Adam optimizer is used for training with a learning rate of 0.005. Within the problem domain, 20,000 points are sampled using Latin hypercube sampling to calculate the partial differential equation term L in the loss function. PDE 2000 points were randomly sampled to construct signal data for calculating L. IC Optionally, the three data points x, y, and t of the sampled true data are standardized to 0-1, and 4000 data points are randomly sampled from the 16000 data points for calculating L. data .

[0044] During the neural network training process, a hard constraint condition is applied to the mountain-blocking area, causing the neural network output to be multiplied by the electromagnetic transmittance. An example of the neural network output image is shown in the right image of Figure 3.

[0045] Step 4: Construct the neural network model from Step 3 using the 9 sets of initial data, and convert the output electric field intensity into power density for subsequent base station coverage calculation, as shown in the following formula:

[0046]

[0047] PL is the maximum allowable path loss, P r To calculate the power density at the point, P t Let represent the power density at the base station location. The relationship between power density and electric field strength is as follows:

[0048]

[0049] Where η0 is the impedance of free space, taken as 377Ω. Combining the two equations, we have:

[0050]

[0051] The neural network outputs information about the change of electric field strength over time. After sampling, the root mean square of the average electric field strength is substituted into the above formula to determine whether the path loss of the location point is greater than the allowable value, thereby determining whether it is within the coverage area.

[0052] The base station locations and corresponding global average electric field intensities corresponding to multiple models are combined as training data to construct a new neural network. After training, this network can automatically output the average electric field intensity within a region based on the input base station location. Optionally, the neural network uses the ReLU activation function, contains three hidden layers with a width of 50, and employs hard constraints to ensure the output is 0 at locations obstructed by mountains.

[0053] Step 5: User Service Reliability Calculation. User service reliability is defined as the probability that a user can successfully complete a network transaction at any time within the service area and during the service period. The Monte Carlo method is used to calculate user service reliability. Specifically, the user's location information, access time, usage time, and maximum allowable interruption time are first randomly generated. The user's location information is randomly obtained from a distribution model consisting of a normal random distribution at the center of three town areas and a uniform random distribution across the entire area. Considering fluctuations in base station strength and capacity load, the path loss at the user's location, the network resources, the interference from other base stations, network latency, and the maximum network interruption time are comprehensively considered to determine whether the user successfully completes a network service transaction. Finally, the ratio of the number of successfully completed transactions to the total number of users is used as the service reliability. The user service reliability calculation logic is shown in Figure 4.

[0054] In this example, we assume that the base station strength fluctuations follow the following formula:

[0055] E t =E+0.1sin(4πt) (10)

[0056] Where E is the electric field intensity output by the proxy channel model, E t The electric field strength generated by the base station at a user's location at a certain time t. The path loss at the user's location can be calculated using this electric field strength. Optionally, the maximum path loss should not exceed 40dB (considering only spatial loss and obstruction loss during propagation), using the formula in step 4:

[0057]

[0058] Similarly, assume that the base station capacity load follows the following fluctuations:

[0059] C t =|sin(2πt)| (12)

[0060] Optionally, if the load exceeds 0.9 during the user's network usage time, the network is considered congested and the user cannot obtain sufficient network resources.

[0061] When a user is under the influence of multiple base stations, since there is only one base station with which the user communicates, the electric field strength generated by the other base stations at the user's location can be regarded as interference signal. The signal-to-noise ratio (SNR) is used to measure the magnitude of the interference. The SNR formula is as follows:

[0062]

[0063] Among them, P s It is the power of the signal, P nIt is the power of the noise, E s It is the field strength generated by the communication base station, E n It is the maximum field strength generated by other sources. Optionally, when the signal-to-noise ratio (SNR) is below 70 dB, the network service quality is considered unqualified.

[0064] Ray tracing can be used to obtain the shortest distance from the base station to the user's location. This distance is the signal propagation distance. Dividing the distance by the speed of light gives the network latency in milliseconds (ms). A latency higher than 80ms is considered to be unqualified in terms of network service quality.

[0065] During the duration of a user's network usage, network outages may occur temporarily due to signal and capacity fluctuations. However, from the user's perspective, network service is not immediately canceled after an outage. Users generally have a maximum acceptable downtime, or patience. When the network outage time does not exceed the user's patience, the user will wait until the network is restored. Therefore, network outage time is recorded during the user's usage time, and when the maximum outage time exceeds the user's patience, the network service quality is considered unsatisfactory.

[0066] Taking all the above factors into account, the number of users who can successfully complete a network service transaction is counted. The ratio of this number to the total number of users is the probability that a user can successfully complete a network transaction at any time during the service period and within the service area, i.e., the service reliability.

[0067] Step 6: Use a genetic algorithm to find the optimal site selection scheme. Using the service reliability R obtained in Step 5 as the fitness function, encode the location information of a specified number of base stations, randomly generate an initial population, calculate the fitness, select individuals with high fitness for crossover, and randomly select individuals to undergo mutation. After several generations, the genetic algorithm provides the optimal base station site selection scheme. Optionally, the initial population size is set to 200, the number of base stations is set to 3, and the mutation probability is set to 0.05. The site selection scheme given by the genetic algorithm after 50 generations is shown in Figure 5, where the coordinates of solid circles represent the base station site locations, and the coordinates of hollow circles represent the user locations.

[0068] Step 7: Dynamic Planning Based on Service Reliability. When personnel relocate within the service area due to task requirements, the user distribution changes, and the service reliability of the existing mobile base station site selection scheme shows a downward trend. When the service reliability falls below the threshold of 0.7, the scheme is considered unsuitable. The genetic algorithm is then invoked again to update the user random distribution model, providing the current optimal site selection scheme, and the mobile base station is redeployed. In this example, assuming that personnel from the town in the upper left corner of the map relocate to the upper right corner, the scheme is replanned when the service reliability falls below 0.7. The image showing the change in service reliability with the location of 50 points during the relocation process is shown in Figure 6. The new site selection scheme after replanning is shown in Figure 7, where solid circles represent base station locations and hollow circles represent user locations.

[0069] The above description represents the preferred embodiment of the present invention. For those skilled in the art, modifications or equivalent substitutions can be made to the specific embodiments of the present invention without departing from the overall concept of the present invention, and these modifications or substitutions should also be considered within the scope of protection of the present invention.

Claims

1. A method for dynamic location optimization of mobile base stations with service reliability as the objective, characterized in that: The process includes the following steps: Step 1: Scene setup for the planning area; determine the size, dimensions, shape, number, and location of buildings within the planning area; simplify it to a two-dimensional problem; Step 2: Initial data acquisition; collect electric field intensity information at different locations within the service area under the condition of a single base station, which can be obtained through on-site data collection or simulation experiments; the simulation experiment uses the time-domain finite element difference method to calculate the field strength at each location; Step 3: Obtain the propagation model of a fixed-location base station in the area using a neural network method based on physical information; use the initial data and the base station's transmitted signal as training samples, and Maxwell's equations as physical constraint equations to build a neural network, employing hard constraint... The method controls the boundary and reflection conditions of the buildings. The network takes spatial coordinates x and y and time t as input and outputs the electric field intensity at the corresponding location and time in the service area. During the neural network training process, for mountain blocking areas, hard constraints are used to multiply the neural network output by the electromagnetic transmittance. Step 4: Fuse multiple propagation models to obtain proxy channel models for base stations at different locations in the service area. Use the neural networks constructed in step 3 to generate training data and train the proxy model. This network can directly input the base station location to obtain the electric field intensity information generated by the corresponding base station in the service area. The output electric field intensity is converted into power density for subsequent base station coverage calculation. The formula is: PL is the maximum allowable path loss, P r To calculate the power density at the point, P t Let be the power density at the base station location; the relationship between power density and electric field strength is as follows: in, Let the impedance of free space be taken as... Combining the two formulas, we have: The neural network outputs information on the change of electric field strength over time. After sampling, the root mean square of the average electric field strength is substituted into the above formula to determine whether the path loss of the location point is greater than the allowable value, thereby determining whether it is within the coverage area. Step 5: User service reliability calculation; The user service reliability is defined as the probability that the user can successfully complete a network transaction at any time within the service area during the service time. The Monte Carlo method is used to calculate the user service reliability. The specific method is as follows: First, the user's location information, access time, usage time, and maximum allowable interruption time are randomly generated. Among them, the user location information is randomly obtained by a distribution model composed of a regional central normal random distribution and a global uniform random distribution. Under the conditions of base station strength fluctuation and base station capacity load fluctuation, the path loss of the user's location, the network resources, the interference from other base stations, the network latency, and the maximum network interruption time are comprehensively considered to determine whether the user can successfully complete the network service transaction. Finally, the ratio of the number of people who successfully complete the transaction to the total number of people is used as the service reliability. Step 6: Genetic algorithm searches for the optimal Location Scheme: Using user service reliability as the optimization objective, base station location information is encoded, and a genetic algorithm is used to search for the optimal base station location deployment scheme for a specified number of base stations within the service area. Using the service reliability obtained in step 5 as the fitness function, the location information of a specified number of base stations is encoded, an initial population is randomly generated, fitness is calculated, individuals with high fitness are selected for crossover, and individuals are randomly selected to mutate. After multiple generations, the genetic algorithm provides the optimal base station location scheme. Step 7: Dynamic Planning Based on Service Reliability: After providing a location scheme, the service reliability within the area is monitored. When the service reliability is lower than the threshold, the genetic algorithm is called to re-plan based on the scenario change. Specifically: when personnel are transferred within the service area according to task needs, the user distribution changes, and the service reliability of the existing mobile base station location scheme shows a downward trend. When the service reliability is lower than the threshold of 0.7, the scheme is considered unsuitable, the genetic algorithm is called again, the user random distribution model is updated, the current optimal location scheme is given, and the mobile base stations are redeployed.

2. The method for dynamic location optimization of mobile base stations with service reliability as the objective, as described in claim 1, is characterized in that: In step 5, considering the path loss at the user's location, the network resources obtained, the interference from other base stations, network latency, and the maximum network interruption time, the system determines whether the user successfully completes the network service transaction. The specific implementation plan is as follows: Base station strength fluctuations follow the formula below: Where E is the electric field intensity output by the proxy channel model, E t The electric field strength generated by a base station at a user's location at a certain time t; the path loss at the user's location can be calculated using the electric field strength, with the maximum path loss required to not exceed 40dB; when a user is affected by multiple base stations, since there is only one base station communicating with it, the electric field strength generated by other base stations at the user's location is considered as interference signal, and the signal-to-noise ratio is used to measure the magnitude of the interference, as shown in the following formula: Among them, P s It is the power of the signal, P n It is the power of the noise, E s It is the field strength generated by the communication base station, E n The maximum field strength generated by other sources; when the signal-to-noise ratio (SNR) is below 70dB, the network service quality is considered unqualified; the shortest distance from the base station location to the user location can be obtained using ray tracing, which is the signal propagation distance. The network delay can be obtained by dividing the distance by the speed of light, in milliseconds. When the delay is higher than 80ms, the network service quality is considered unqualified; the network outage time is counted during the user's usage time. When the maximum outage time exceeds the user's patience, the network service quality is considered unqualified.

Images