A path loss modeling method, apparatus, device, and storage medium

Abstract

This application discloses a path loss modeling method, apparatus, device, and storage medium, comprising: determining an initial model and an environmental parameter dataset; the environmental parameter dataset including at least one candidate environmental parameter; optimizing the initial model using a genetic algorithm to obtain a target calibration model; adjusting the target calibration model according to each candidate environmental parameter in the environmental parameter dataset to obtain the adjusted error value of the target calibration model corresponding to the candidate environmental parameter; determining the target environmental parameter based on the error value using a preset greedy algorithm; and optimizing the target calibration model according to the target environmental parameter to obtain a path loss model. In this way, by using both genetic and greedy algorithms for model optimization and parameter selection, considering the influence of environmental parameters on the path loss model, and selecting the environmental parameter with a greater impact on the model to optimize the model, the modeling complexity is reduced, while the modeling accuracy and applicability are improved.

Description

Technical Field

[0001] This application relates to the field of channel propagation modeling technology, and in particular to a path loss modeling method, apparatus, device, and storage medium. Background Technology

[0002] With the development of 5G (5th Generation Mobile Communication Technology), the frequency range (FR) of 5G New Radio (NR) is defined as two different FRs: FR1 and FR2. FR1 refers to the 5G Sub-6GHz band, while FR2 is the 5G millimeter-wave band. Compared to the Sub-6GHz band, the millimeter-wave band offers faster transmission rates, wider bandwidth, the ability to support more connections, and better directionality. However, when using the millimeter-wave band for communication, its high propagation loss and high penetration loss (affected by the propagation medium), as well as attenuation due to rainfall, result in a higher overall path loss between transceivers compared to lower frequency bands.

[0003] In related technologies, traditional millimeter-wave channel path loss modeling algorithms can be divided into two categories: one is statistical channel modeling based on actual environmental measurement data, i.e., empirical models; the other is deterministic channel modeling based on electromagnetic theory calculations, i.e., deterministic models. For empirical millimeter-wave channel path loss models, since they are essentially based on pure statistical analysis, their drawback is that the model built from each measurement is only applicable to the specific environment at that time, resulting in poor universality. For deterministic millimeter-wave channel path loss models, their computational complexity is high, and there is no unified standard for the electrical parameters of different environmental materials; the determination of electrical parameters is often based on experience, further reducing their versatility. Therefore, millimeter-wave channel path loss modeling for scenarios with different environmental parameters (air density, water vapor partial pressure, air temperature, atmospheric pressure, humidity, transmitter altitude, oxygen content, and rainfall) currently lacks universality and suffers from high computational complexity. Summary of the Invention

[0004] This application provides a path loss modeling method, apparatus, device, and storage medium, which can not only reduce the modeling complexity of path loss models, but also improve modeling accuracy and applicability.

[0005] The technical solution of this application is implemented as follows:

[0006] In a first aspect, embodiments of this application provide a path loss modeling method, the method comprising:

[0007] Determine the initial model and environmental parameter dataset; wherein the environmental parameter dataset includes at least one candidate environmental parameter;

[0008] The initial model is optimized using a genetic algorithm to obtain the target correction model;

[0009] The target calibration model is adjusted according to each candidate environmental parameter in the environmental parameter dataset to obtain the adjusted error value of the target calibration model corresponding to the candidate environmental parameter. The target environmental parameter is then determined based on the error value using a preset greedy algorithm.

[0010] The target correction model is optimized based on the target environmental parameters to obtain the path loss model.

[0011] Secondly, embodiments of this application provide a path loss modeling apparatus, which includes a first determining unit, a first optimizing unit, a second determining unit, and a second optimizing unit, wherein...

[0012] The first determining unit is configured to determine an initial model and an environmental parameter dataset; wherein the environmental parameter dataset includes at least one candidate environmental parameter;

[0013] The first optimization unit is configured to use a genetic algorithm to optimize the initial model to obtain the target correction model;

[0014] The second determining unit is configured to adjust the target correction model according to each candidate environmental parameter in the environmental parameter dataset, obtain the adjusted error value of the target correction model corresponding to the candidate environmental parameter, and determine the target environmental parameter according to the error value using a preset greedy algorithm;

[0015] The second optimization unit is configured to optimize the target correction model based on the target environment parameters to obtain a path loss model.

[0016] Thirdly, embodiments of this application also provide an electronic device, which includes a memory and a processor, wherein,

[0017] The memory is used to store computer programs that can run on the processor;

[0018] The processor is configured to execute the path loss modeling method as described in the first aspect when running the computer program.

[0019] Fourthly, embodiments of this application provide a computer storage medium storing a computer program that, when executed by at least one processor, implements the path loss modeling method as described in the first aspect.

[0020] This application provides a path loss modeling method, apparatus, device, and storage medium. The method involves determining an initial model and an environmental parameter dataset, wherein the environmental parameter dataset includes at least one candidate environmental parameter. A genetic algorithm is used to optimize the initial model to obtain a target calibration model. The target calibration model is then adjusted based on each candidate environmental parameter in the dataset to obtain the adjusted error value corresponding to each candidate environmental parameter. A pre-defined greedy algorithm is used to determine the target environmental parameter based on the error value. Finally, the target calibration model is optimized based on the target environmental parameter to obtain a path loss model. In this way, when establishing a path loss model, both genetic and greedy algorithms are used for model optimization and environmental parameter selection, fully considering the influence of each environmental parameter on the path loss model. The environmental parameter with the greater impact on the model is selected for calibration, thereby reducing the modeling complexity of the path loss model and improving its modeling accuracy and applicability. Attached Figure Description

[0021] Figure 1 A flowchart illustrating a path loss modeling method provided in an embodiment of this application;

[0022] Figure 2 A detailed flowchart illustrating a path loss modeling method provided in this application embodiment;

[0023] Figure 3 This is a schematic diagram of a BP neural network structure provided in an embodiment of this application;

[0024] Figure 4 A flowchart illustrating the optimization of weights and thresholds in a BP neural network, provided as an embodiment of this application;

[0025] Figure 5 A detailed flowchart illustrating another path loss modeling method provided in this application embodiment;

[0026] Figure 6 A schematic diagram of the composition structure of a path loss modeling device provided in an embodiment of this application;

[0027] Figure 7 A schematic diagram of the composition structure of an electronic device provided in an embodiment of this application;

[0028] Figure 8 This is a schematic diagram of the composition structure of another electronic device provided in an embodiment of this application. Detailed Implementation

[0029] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining the relevant application and not for limiting the application. Furthermore, it should be noted that, for ease of description, only the parts related to the relevant application are shown in the accompanying drawings.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0031] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0032] It should be noted that the terms "first, second, and third" used in the embodiments of this application are merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, and third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0033] As is well known, wireless communication signals rely on electromagnetic waves for propagation, and the most valuable resource in this process is the frequency band. To prevent mutual interference between wireless television broadcasting, mobile communication networks, and military frequency bands, each country has strictly allocated frequency bands. Based on the propagation characteristics of electromagnetic waves in the air, radio waves below 6 GHz are considered high-quality frequency band resources due to their advantages such as low power attenuation and strong penetration in the air. Many applications that rely on radio wave propagation are concentrated in this part of the frequency band, making it very congested.

[0034] With the development of technology and the improvement of people's living standards, users' data demand for mobile communication networks has exploded, especially for wireless applications that require the transmission of large amounts of real-time data, such as high-definition teleconferencing, live video streaming, and virtual reality (VR) games, which pose a severe test to the capacity of communication networks. At the same time, "mission-critical machine communication" (such as industrial automation and vehicle communication) places extremely stringent requirements on latency and reliability. Therefore, the industry has reached a consensus on the next generation of wireless communication: 1000 times the network capacity of the Fourth Generation Mobile Communication (4G) system and an extremely low latency of 1 millisecond. Faced with scarce wireless frequency bands and strong user demand, the arrival of the 5G era can gradually solve the problem of how to ensure the lowest network speed in this era of technological explosion, building a high-speed information transmission channel for society.

[0035] According to Shannon's theorem, the most direct and effective way to increase channel capacity is to increase the bandwidth of the communication system. Currently, the low-frequency bandwidth used for mobile communication in China is only about 600MHz. Based on the requirements of 5G technology, approximately 1GHz of bandwidth is needed for the formal commercial use of mobile communication. Therefore, new spectrum resources need to be developed to meet the development needs of 5G technology. The potentially large bandwidth of the millimeter-wave band (26-300GHz) is considered a new spectrum resource, capable of providing higher data transmission rates over short distances, and is seen as a potential development trend for 5G technology. Although the millimeter-wave band has the aforementioned advantages, there are still many technical challenges in realizing millimeter-wave cellular network communication. According to Friis's law of radio wave propagation, free-space omnidirectional path loss is proportional to the square of the frequency. When using the millimeter-wave band for communication, due to its high propagation loss and high penetration loss (affected by the propagation medium), and attenuation due to rainfall, the total path loss between transceivers becomes higher than that of low-frequency bands. Therefore, it is necessary to reduce the high cost of channel measurement and provide models that can more accurately grasp the propagation characteristics of the channel, thereby providing a more accurate channel model standard for 5G.

[0036] In related technologies, traditional millimeter-wave channel path loss modeling algorithms can be divided into two categories: one is statistical channel modeling based on actual environmental measurement data, i.e., empirical models; the other is deterministic channel modeling based on electromagnetic theory calculations, i.e., deterministic models.

[0037] Empirical models are based on measurements taken in actual indoor and outdoor environments, specifically by conducting real-world measurements of radio wave propagation to obtain data that reflects channel characteristics. By processing and analyzing this measured data, general channel patterns under different environments can be revealed. Currently, mainstream statistical channel modeling methods are based on geometric statistical modeling. However, millimeter-wave channel path loss empirical models, being fundamentally based on pure statistical analysis, suffer from poor universality because the models built from each measurement are only applicable to the specific environment in which they are performed.

[0038] Deterministic models use electromagnetic field theory and statistical methods to analyze the propagation characteristics of radio waves in the channel. This approach provides more detailed channel information, such as the angle of arrival (AoA), angle of departure (AoD), time of arrival (ToA), and Doppler shift. This method requires obtaining geometric information of scatterers in the wireless propagation environment and electrical parameters of the materials constituting that environment from a comprehensive environmental database. Based on the location of the base station and users and the antenna configuration, the channel response is calculated using Maxwell's equations or approximate equations for radio wave propagation. A typical deterministic channel modeling method is ray tracing (RT). However, deterministic models of millimeter-wave channel path loss suffer from high computational complexity, and the lack of a unified standard for electrical parameters across different environmental materials, coupled with the fact that parameter determination is often based on experience, limits the universality of deterministic models.

[0039] Furthermore, the Back Propagation Neural Network (BPNN) is a common multilayer feedforward neural network. Its key characteristic is data forward propagation and error backward propagation. Due to this error backpropagation property, it can theoretically fit any functional relationship, exhibiting good nonlinear mapping capabilities, superior adaptive and self-learning abilities, good generalization, and fault tolerance. Therefore, it can be used to solve the problem of millimeter-wave channel path loss modeling under various environmental parameters (water vapor partial pressure, air temperature, atmospheric pressure, humidity, carrier frequency, transmitter altitude, and rainfall, etc.). While BPNNs offer good performance and wide applications, they also have limitations, such as slow learning convergence speed, inability to guarantee convergence to the global minimum, and difficulty in determining the network structure. Moreover, the network structure, initial connection weights, and threshold selection significantly impact network training, yet these cannot be accurately determined.

[0040] Based on this, this application provides a path loss modeling method. The basic idea of ​​this method is as follows: determine an initial model and an environmental parameter dataset; wherein, the environmental parameter dataset includes at least one candidate environmental parameter; optimize the initial model using a genetic algorithm to obtain a target calibration model; adjust the target calibration model according to each candidate environmental parameter in the environmental parameter dataset to obtain the adjusted error value of the target calibration model corresponding to the candidate environmental parameter; determine the target environmental parameter based on the error value using a preset greedy algorithm; optimize the target calibration model according to the target environmental parameter to obtain the path loss model. In this way, when establishing the path loss model, both genetic algorithms and greedy algorithms are used for model optimization and environmental parameter selection, fully considering the influence of each environmental parameter on the path loss model, and selecting environmental parameters with a greater impact on the model for calibration, thereby not only reducing the modeling complexity of the path loss model, but also improving the modeling accuracy and applicability of the model.

[0041] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0042] In one embodiment of this application, see [link to embodiment]. Figure 1 This illustrates a flowchart of a path loss modeling method provided in an embodiment of this application. Figure 1 As shown, the method may include:

[0043] S101. Determine the initial model and environmental parameter dataset.

[0044] It should be noted that the path loss modeling method provided in this application embodiment is used to create a path loss model, which enables the path loss model to accurately determine the path loss caused by radio waves during propagation, especially the channel path loss in the millimeter wave band, by combining environmental parameters and other factors.

[0045] It should also be noted that this method can be applied to devices for creating path loss models, or electronic devices, systems, etc., that integrate such devices. Here, electronic devices can be such as computers, smartphones, tablets, laptops, handheld computers, personal digital assistants (PDAs), navigation devices, etc., and this application embodiment does not specifically limit them.

[0046] It should also be noted that, during the propagation of wireless communication signals, the difference between the transmitter's transmit power (also simply called transmit power) and the receiver's receive power (also simply called receive power) is defined as path loss, which is the power attenuation during signal propagation. The purpose of this application's embodiments is to establish an accurate path loss model, so that, given the transmitter's transmit power and other input parameters, the path loss model can accurately output the receiver's receive power, enabling a comprehensive analysis of channel propagation path loss from multiple input parameter dimensions under different environmental conditions.

[0047] In this embodiment of the application, an initial model and an environmental parameter dataset are first determined, the environmental parameter dataset including at least one candidate environmental parameter.

[0048] In some embodiments, determining the initial model may include:

[0049] Obtain the sample dataset;

[0050] The sample dataset is normalized to obtain the normalized sample dataset;

[0051] The preset neural network model is trained using the normalized sample dataset to obtain the initial model.

[0052] It should be noted that the sample dataset used to train the preset neural network model in this application embodiment mainly includes: the sample dataset corresponding to the input parameters that undoubtedly have a significant impact on path loss.

[0053] For example, the carrier frequency and the straight-line distance between the transmitter and receiver (i.e., the transmission distance) will inevitably affect the received power. Therefore, in this embodiment, the carrier frequency, the straight-line distance between the transmitter and receiver, the transmit power, and the corresponding receive power are used as a sample dataset. The sample dataset can be obtained through actual measurement, querying past recorded data, etc.

[0054] It should also be noted that, before training the preset neural network model, the embodiments of this application may also normalize the sample dataset, and use the normalized sample dataset to train the preset neural network model to obtain an initial model.

[0055] In this embodiment of the application, the preset neural network model is preferably a BP neural network model (also referred to as a BP neural network).

[0056] In this embodiment of the application, the environmental parameter dataset includes at least one candidate environmental parameter, which may include at least one or more of the following formulas: air density, water vapor partial pressure, air temperature, atmospheric pressure, humidity, transmitter altitude, oxygen content, and rainfall.

[0057] It should be noted that candidate environmental parameters refer to environmental parameters that may affect the modeling results, such as air density, water vapor partial pressure, air temperature, atmospheric pressure, humidity, transmitter altitude, oxygen content, and rainfall. Here, rainfall refers to the rainfall at the current measurement time. If there is no rainfall, then in the BP neural network of this embodiment, the input parameter - rainfall - is 0.

[0058] In addition, millimeter waves may be affected by oxygen, humidity, fog, and rain during atmospheric propagation.

[0059] The impact of oxygen is not uniform; different millimeter-wave bands are affected differently. For example, 60GHz must withstand an oxygen absorption loss of approximately 20dB / km, while the requirements are lower for 28GHz, 38GHz, and 73GHz. This is why some operators currently prioritize 28GHz for their testing.

[0060] Compared to oxygen, humidity has a greater impact on the attenuation of millimeter waves. In high temperature and high humidity environments, the signal can be attenuated by half (3dB / km) within 1 kilometer. Similarly, millimeter waves also experience attenuation when passing through fog and clouds.

[0061] Rain is the biggest enemy of millimeter waves. In extreme cases, during torrential rain (with a rainfall intensity of 50 mm / hour), the propagation loss of millimeter waves can reach 18.4 dB / km.

[0062] In some embodiments, determining the environmental parameter dataset may include:

[0063] Obtain the initial environment parameter dataset;

[0064] Based on a pre-defined correlation analysis algorithm, the influence of each initial candidate environmental parameter on the path loss in the initial environmental parameter dataset is calculated to determine the influence of each initial candidate environmental parameter on the path loss.

[0065] The significance level test results of each initial candidate environmental parameter and path loss in the initial environmental parameter dataset are performed using a pre-defined test algorithm to determine the significance level test results of each initial candidate environmental parameter relative to path loss.

[0066] Based on the results of the impact and significance level tests, initial candidate environmental parameters that are not related to path loss are determined.

[0067] The initial candidate environmental parameters that are not related to path loss are removed from the initial environmental parameter dataset to obtain the environmental parameter dataset.

[0068] It should be noted that, in this embodiment of the application, the correlation between each initial candidate environmental parameter and path loss in the initial environmental parameter dataset is determined by a preset correlation analysis algorithm and a preset test algorithm, and a significance level test is performed. Only the initial candidate environmental parameters that are significantly related to path loss are retained. The retained initial candidate environmental parameters form the environmental parameter dataset. These retained initial candidate environmental parameters are the candidate environmental parameters in the environmental parameter dataset.

[0069] Specifically, based on a pre-defined correlation analysis algorithm, the influence of each initial candidate environmental parameter in the initial environmental parameter dataset on path loss is calculated. This can be determined by calculating the correlation coefficient between each initial candidate environmental parameter and path loss. Thus, the influence of each initial candidate environmental parameter on path loss is determined based on the magnitude of the correlation coefficient. Generally speaking, the larger the correlation coefficient, the more correlated the two are.

[0070] To eliminate the influence of random factors, a pre-defined testing algorithm is needed to perform a significance level test on each initial candidate environmental parameter and path loss in the initial environmental parameter dataset. The significance level test result of each initial candidate environmental parameter relative to path loss in the initial environmental parameter dataset is determined. Combining the aforementioned influence and significance level test results, the initial candidate environmental parameters in the initial environmental parameter dataset that are not related to path loss are identified.

[0071] For example, using a two-sample t-test, initial candidate environmental parameters that are statistically significant at a level of 95% or higher (i.e., a significance level of 0.05) are identified as initial candidate environmental parameters significantly related to path loss. Initial candidate environmental parameters that do not meet this significant correlation requirement are identified as initial candidate environmental parameters not related to path loss, and these unrelated initial candidate environmental parameters are removed from the initial environmental parameter dataset to obtain the final environmental parameter dataset.

[0072] S102. Use a genetic algorithm to optimize the initial model to obtain the target correction model.

[0073] It should be noted that after determining the initial model, a genetic algorithm is used to optimize the initial model to obtain the target correction model.

[0074] The preferred method for optimizing the initial model is to use a genetic algorithm, but other optimization methods, such as Bayesian optimization, can also be used. In this embodiment, a genetic algorithm is used as an example to describe the method for optimizing the initial model.

[0075] It should also be noted that, in the embodiments of this application, preferably, the initial model is obtained by training a preset neural network model, so optimizing the initial model means optimizing the parameters of the neural network model (mainly referring to the weights and thresholds of the neural network).

[0076] By optimizing the model using a genetic algorithm, the optimal weights and thresholds of the current model can be determined. In this step, the initial model is optimized using a genetic algorithm to determine the optimal weights and thresholds of the initial model, and a target correction model is created based on the optimal weights and thresholds.

[0077] In other words, optimizing the initial model refers to optimizing the model's own parameters, rather than the input or output parameters.

[0078] In some embodiments, optimizing the initial model using a genetic algorithm to obtain the target correction model may include:

[0079] Based on the initial model, determine the initial population set; wherein the initial population set includes at least one individual;

[0080] Calculate the fitness of all individuals in the initial population set, sort all individuals in the initial population set according to fitness, and select at least one candidate individual with the highest fitness to form a population set to be mated.

[0081] The population set to be mated is subjected to mutation and crossover processing, and a preset number of target individuals are selected from the mutation and crossover results to form a new population set;

[0082] Based on the new population set, generate the first calibration model;

[0083] If the first calibration model meets the preset termination condition, then the first calibration model is determined as the target calibration model;

[0084] If the first calibration model does not meet the preset termination condition, the new population set is determined as the initial population set, and the process returns to the step of calculating the fitness of all individuals in the initial population set until the first calibration model meets the preset termination condition and is determined as the target calibration model.

[0085] It should be noted that when using genetic algorithms to optimize the initial model, firstly, based on the neural network structure of the initial model, an initial population set is determined, i.e., the initial population is initialized. This initial population set includes at least one individual, which corresponds to a parameter of the neural network.

[0086] Then, the fitness of each individual in the initial population set is calculated, and the individuals in the initial population set are sorted according to their fitness. Based on the sorting results, at least one individual with the highest fitness (i.e., a candidate individual) is selected to form the population set to be mated. For example, two candidate individuals are selected and copied to the population set to be mated. If there is only one individual in the initial population set, that individual is copied twice to the population set to be mated.

[0087] Next, the population set to be bred is subjected to mutation and crossover processing. A predetermined number of target individuals are selected from the mutation and crossover results to form a new population set. For example, the predetermined number can be one, and usually an optimal individual is selected as the target individual from the predetermined crossover and mutation structure.

[0088] Specifically, individuals in the population set to be bred are mutated according to a preset mutation probability to obtain several mutated individuals. Then, these mutated individuals are replicated according to the principle that superior individuals are replicated several times (e.g., 4 times) while inferior individuals are not replicated, to obtain a replication group.

[0089] Next, randomly select several individuals (e.g., 2) from the replication group and perform multiple crossover operations according to the preset crossover probability to obtain the mutated crossover results. The mutated crossover results include multiple individuals after the mutated crossover treatment. Select the best individual from the mutated crossover results as the target individual to form a new population set.

[0090] Based on the new population set, i.e. the neural network parameters corresponding to the target individual, a first calibration model is generated. It is then determined whether the first calibration model meets the preset termination condition. If the preset termination condition is met, the first calibration model is determined as the target calibration model. If not, the new population set is determined as the initial population set, and the process returns to the step of calculating the fitness of all individuals in the initial population set. This process continues until the first calibration model meets the preset termination condition, at which point the first calibration model is determined as the target calibration model.

[0091] In this embodiment of the application, when determining whether the first calibration model meets the preset termination condition, the error value (such as root mean square error) of the first calibration model can be compared with the preset model error threshold. If the error value of the first calibration model is less than or equal to the preset model error threshold, the first calibration model is determined to meet the preset termination condition; otherwise, the first calibration model does not meet the preset termination condition. And / or, a preset iteration number threshold is set, and the process from determining the initial population set to generating the first calibration model is considered as one iteration. When the number of iterations reaches the preset iteration number threshold, the first calibration model is determined to meet the preset termination condition; otherwise, the first calibration model does not meet the preset termination condition.

[0092] It should also be noted that the error value involved in the embodiments of this application can be used to represent the current optimization effect of the model. The smaller the value, the more accurate the model is.

[0093] S103. Adjust the target calibration model according to each candidate environmental parameter in the environmental parameter dataset to obtain the adjusted error value of the target calibration model corresponding to the candidate environmental parameter, and determine the target environmental parameter based on the error value using a preset greedy algorithm.

[0094] It should be noted that after obtaining the target calibration model, it can be adjusted based on each candidate environmental parameter in the environmental parameter dataset. This yields the adjusted error value for each candidate environmental parameter, and a pre-defined greedy algorithm is used to determine the target environmental parameter based on this error value. Specifically, each candidate environmental parameter in the dataset can be sequentially input into the target calibration model, resulting in the adjusted target calibration model for each candidate environmental parameter, along with the error value of the adjusted model. A greedy algorithm is then used to determine whether each candidate environmental parameter is the target environmental parameter, thus identifying the target environmental parameter. The error value can be the loss function value of the target calibration model, the root mean square error, etc.

[0095] In some embodiments, the number of target environmental parameters is at least one; the step of adjusting the target calibration model according to each candidate environmental parameter in the environmental parameter dataset to obtain the adjusted error value of the target calibration model corresponding to the candidate environmental parameter, and determining the target environmental parameter based on the error value using a preset greedy algorithm, may include:

[0096] The target calibration model is adjusted using each candidate environmental parameter in the environmental parameter dataset to obtain the adjusted calibration model corresponding to each candidate environmental parameter, and the error value of the adjusted calibration model corresponding to each candidate environmental parameter is determined.

[0097] Select at least one target error value from the error values, and determine the target environmental parameters based on at least one target error value.

[0098] It should be noted that, in this embodiment, each candidate environmental parameter in the environmental parameter dataset can be sequentially input into the calibration model. The target calibration model is then adjusted using each candidate environmental parameter to obtain an adjusted calibration model corresponding to each candidate environmental parameter. The error value of each adjusted calibration model is then obtained, which is the error value of the corresponding candidate environmental parameter. At least one target error value is selected from these error values, and the target environmental parameter is determined based on this at least one target error value. The error value represents the error between the received power output by the adjusted calibration model and the actual received power. It can be the model's loss function, root mean square error, etc., and this error value characterizes the accuracy of the model.

[0099] For example, an error value less than a preset error threshold can be defined as the target error value, and the candidate environmental parameters corresponding to the target error value can be defined as the target environmental parameters. If the error value is less than the preset error threshold, it means that the candidate environmental parameter corresponding to the error value has a good optimization effect on the target correction model and has a significant impact on path loss, so it can be used as the target environmental parameter.

[0100] In some embodiments, adjusting the target calibration model according to each candidate environmental parameter in the environmental parameter dataset to obtain the adjusted error value of the target calibration model corresponding to the candidate environmental parameter, and determining the target environmental parameter based on the error value using a preset greedy algorithm, may include:

[0101] Determine the first error value of the target correction model;

[0102] The target calibration model is adjusted using the i-th candidate environmental parameter in the environmental parameter dataset to obtain the second calibration model; where 1≤i≤N, and N is the number of candidate environmental parameters in the environmental parameter dataset;

[0103] Determine the second error value of the second correction model;

[0104] If the difference between the first error value and the second error value is greater than the preset error threshold, then the i-th candidate environmental parameter is determined as one of the target environmental parameters;

[0105] Increment i by one, and then adjust the target calibration model using the i-th candidate environmental parameter in the environmental parameter dataset to obtain the second calibration model. Repeat this process until i equals N or the difference between the first error value and the second error value is not greater than a preset error threshold, so as to determine all target environmental parameters.

[0106] It should be noted that after obtaining the target correction model, the first error value of the target correction model is calculated. This error value represents the error between the received power output by the model and the actual received power. In this embodiment, the error value is preferably the root mean square error.

[0107] In this embodiment, the i-th environmental parameter is input into the target calibration model, and the target calibration model is trained and adjusted using the i-th environmental parameter to obtain a second calibration model. The second error value of the second calibration model is then calculated. In this embodiment, 1 ≤ i ≤ N, where N is the number of candidate environmental parameters in the environmental parameter dataset. Generally, i starts from 1 and continues until i equals N.

[0108] If the difference between the first error value and the second error value is greater than a preset error threshold, then the i-th candidate environmental parameter is determined as one of the target environmental parameters. That is, the difference between the first error value and the second error value is calculated, and it is determined whether the difference is greater than the preset error threshold. If the difference is greater than the preset error threshold, it indicates that the i-th environmental parameter has a good improvement effect on the model.

[0109] For example, the preset error threshold is set to 1 dB, the first error value is 5 dB, and the second error value is 3 dB, with a difference of 2 dB. It can be seen that in this case, after adding the i-th environmental parameter, the error value of the first calibration model is significantly reduced compared to the target calibration model, indicating that the i-th environmental parameter has a good improvement effect on the model. Therefore, the i-th environmental parameter is determined as a target environmental parameter.

[0110] For example, if the preset error threshold is set to 1 dB, the first error value is 5 dB, and the second error value is also 5 dB, with a difference of 0, it can be seen that in this case, after adding the i-th environmental parameter, the error value of the first calibration model remains unchanged compared to the calibration model; the i-th environmental parameter has no effect on improving the model. Therefore, the i-th environmental parameter is not used as the target environmental parameter.

[0111] After determining whether a candidate environmental parameter is a target environmental parameter, we check whether the current i is equal to N. If i is equal to N, it means that every candidate parameter in the environmental parameter dataset has been judged, that is, all target environmental parameters have been determined. If i is not equal to N, it means that there are still candidate environmental parameters in the environmental parameter dataset that have not been judged. We need to continue to use the next environmental parameter to adjust the target calibration model to determine whether the candidate environmental parameter is a target environmental parameter.

[0112] After incrementing i, the target calibration model is adjusted using the i-th candidate environmental parameter in the environmental parameter dataset to obtain the second calibration model. This process is repeated until i equals N or the difference between the first error value and the second error value is not greater than the preset error threshold, at which point all target environmental parameters are determined.

[0113] In other words, the final target environmental parameters can be determined when the following conditions are met: first, i equals N; second, the difference between the first error value and the second error value is not greater than a preset error threshold. In practical applications, these two conditions can be used in combination, or either one can be chosen.

[0114] It should also be noted that, in the embodiments of this application, when the i-th environmental parameter is added to the target correction model, the i-th environmental parameter may be added to the target correction model in a random order, or in an order of high to low or low to high correlation.

[0115] In the embodiments of this application, when determining the target environmental parameters, the target correction model can be updated after each target environmental parameter is determined, that is, the model after adding the target environmental parameter is determined as the initial model for inputting the next candidate environmental parameter.

[0116] Therefore, in some embodiments, the method may further include:

[0117] Determine the first error value of the calibration model.

[0118] The target calibration model is adjusted using the i-th candidate environmental parameter in the environmental parameter dataset to obtain the updated calibration model; where 1≤i≤N, and N is the number of environmental parameters in the environmental parameter dataset;

[0119] Determine the second error value of the updated calibration model;

[0120] If the difference between the first error value and the second error value is greater than the preset error threshold, then the i-th candidate environmental parameter is determined as the target environmental parameter.

[0121] Increment i by one, and determine the updated calibration model as the calibration model. Determine the second error value as the first error value. Return to adjust the target calibration model using the i-th candidate environmental parameter in the environmental parameter dataset to obtain the updated calibration model. Repeat the process until i equals N or the difference between the first error value and the second error value is not greater than the preset error threshold, so as to determine all target environmental parameters.

[0122] In other words, in this embodiment of the application, after determining the first error value of the target correction model, the i-th candidate environmental parameter is added to the target correction model to adjust the target correction model, thereby obtaining the updated target correction model, and the second error value of the updated target correction model is calculated.

[0123] When the difference between the first error value and the second error value is greater than the preset error threshold, the i-th candidate parameter is determined as the target environment parameter.

[0124] Next, after incrementing i, the updated target calibration model is determined as the target calibration model, and the second error value is determined as the first error value. That is, after each target environment parameter is determined, the target calibration model will be updated once. The updated target calibration model includes the target environment parameters that have been determined.

[0125] Then, the target calibration model is adjusted using the i-th candidate environmental parameter in the environmental parameter dataset to obtain the second calibration model. This process is repeated until i equals N or the difference between the first error value and the second error value is not greater than the preset error threshold, at which point all target environmental parameters are determined.

[0126] It should also be noted that the values ​​of each preset error threshold involved in the embodiments of this application can be the same or different, and can be set according to the actual needs of the scenario.

[0127] In some embodiments, adjusting the target calibration model according to each candidate environmental parameter in the environmental parameter dataset to obtain the adjusted error value of the target calibration model corresponding to the candidate environmental parameter, and determining the target environmental parameter based on the error value using a preset greedy algorithm, may include:

[0128] The environmental parameter dataset is randomly grouped to obtain a first subset of environmental parameter data and a second subset of environmental parameter data;

[0129] The target calibration model is adjusted using each candidate environmental parameter from the first subset of environmental parameter data to obtain a third calibration model corresponding to each candidate environmental parameter. A third error value for each candidate environmental parameter's third calibration model is determined. The first target environmental parameters are then determined based on the correlation between the third error value and the prediction error.

[0130] The target calibration model is adjusted using each candidate environmental parameter in the second set of environmental parameter data subset to obtain the fourth calibration model corresponding to each candidate environmental parameter. The fourth error value of the fourth calibration model corresponding to each candidate environmental parameter is determined. The second target environmental parameter is determined based on the fourth error value using a traditional greedy algorithm.

[0131] Determine the first optimization result of the first target environmental parameters on the calibration model, and determine the second optimization result of the second target environmental parameters on the target calibration model;

[0132] Select the target optimization result from the first optimization result and the second optimization result, and determine the environmental parameters input to the target correction model under the target optimization result as the target environmental parameters.

[0133] It should be noted that, in this embodiment of the application, the target environmental parameters can also be determined by randomly grouping the environmental parameter dataset.

[0134] In this embodiment, the environmental parameter dataset can be randomly divided into a first set of environmental parameter data subsets and a second set of environmental parameter data subsets. Then, each candidate environmental parameter in the first set of environmental parameter data subsets is sequentially input into the target calibration model. The target calibration model is adjusted using each candidate environmental parameter in the first set of environmental parameter data subsets to obtain a third calibration model corresponding to each candidate environmental parameter in the first set of environmental parameter data subsets. A third error value is obtained for each third calibration model. The first target environmental parameter is determined by ranking the parameters based on the correlation between the third error value and the prediction error. In other words, the parameters in the first set of environmental parameter data subsets that are most correlated with the model's prediction error are determined as the first target environmental parameter.

[0135] Simultaneously, each candidate environmental parameter in the second subset of environmental parameter data is sequentially input into the target calibration model. The target calibration model is then adjusted using each candidate environmental parameter to obtain a fourth calibration model corresponding to each candidate environmental parameter. A fourth error value is determined for each fourth calibration model, and a traditional greedy algorithm is used to determine the second target environmental parameter based on this fourth error value. Since the traditional greedy algorithm can typically find the optimal solution, the second target environmental parameter can be determined by solving for the optimal solution using the traditional greedy algorithm after sequentially inputting the candidate environmental parameters into the target calibration model and obtaining the fourth error value.

[0136] Then, based on the first objective environmental parameters, the first optimization result of the calibration model can be obtained, and based on the second objective parameters, the second optimization result of the calibration model can be obtained.

[0137] Finally, the target optimization result is selected from the first optimization result and the second optimization result, and the environmental parameters input to the target correction model under the target optimization result are used to determine the target environmental parameters.

[0138] Furthermore, since this step uses a greedy algorithm to filter parameters, the greedy algorithm can quickly determine the local optimum. However, the local optimum is not necessarily equal to the global optimum. The greedy algorithm is prone to getting trapped in a local optimum too early. Therefore, this embodiment of the application introduces a bidirectional priority randomization strategy to expand the search space of the greedy algorithm and reduce the possibility of the greedy algorithm getting trapped in a local optimum.

[0139] Therefore, in some embodiments, the preset greedy algorithm is a bidirectional priority-based random greedy algorithm. The step of adjusting the target correction model based on each candidate environmental parameter in the environmental parameter dataset to obtain the adjusted error value of the target correction model corresponding to the candidate environmental parameter, and then using the preset greedy algorithm to determine the target environmental parameter based on the error value, may include:

[0140] The environmental parameter dataset is randomly grouped using a first-priority strategy, resulting in a first subset and a second subset of environmental parameter data; and

[0141] The environmental parameter dataset is randomly grouped using the second priority strategy to obtain a third and a fourth subset of environmental parameter data; wherein the first and second priority strategies are opposite priorities to each other.

[0142] Based on the first set of environmental parameter data subsets and the second set of environmental parameter data subsets, determine the target environmental parameters and corresponding optimization results under the first priority strategy;

[0143] Based on the third set of environmental parameter data subsets and the fourth set of environmental parameter data subsets, determine the target environmental parameters and corresponding optimization results under the second priority strategy;

[0144] From the target environment parameters and corresponding optimization results under the first priority strategy and the target environment parameters and corresponding optimization results under the second priority strategy, select the target optimization result and determine the target environment parameters by taking the environmental parameters input to the target correction model under the target optimization result.

[0145] It should be noted that when using the bidirectional priority-based random greedy algorithm to determine the target environmental parameters, it can be divided into two parts. First, the environmental parameter dataset is randomly grouped using the first priority strategy (i.e., the forward priority strategy) to obtain the first set of environmental parameter data subsets and the second set of environmental parameter data subsets. Then, the environmental parameter dataset is randomly grouped using the second priority strategy (i.e., the reverse priority strategy) to obtain the third set of environmental parameter data subsets and the fourth set of environmental parameter data subsets.

[0146] In other words, the first priority strategy and the second priority strategy have opposite priorities.

[0147] Then, based on the first set of environmental parameter data subsets and the second set of environmental parameter data subsets, the target environmental parameters and corresponding optimization results under the first priority strategy are determined; and based on the third set of environmental parameter data subsets and the fourth set of environmental parameter data subsets, the target environmental parameters and corresponding optimization results under the second priority strategy are determined.

[0148] Finally, from the target environment parameters and corresponding optimization results under the first priority strategy and the target environment parameters and corresponding optimization results under the second priority strategy, the target optimization result is selected, and the environment parameters input to the calibration model under the target optimization result are determined as the target environment parameters.

[0149] For example, randomly grouping the environmental parameter dataset using the first priority strategy can be as follows: assign priority 2 to the first environmental parameter, and assign priority 1 to all other environmental parameters in the dataset except the first environmental parameter; multiply the priority of each environmental parameter by a random number to obtain the first update priority of each environmental parameter; sort the first update priorities in descending order according to their priority values, and determine the sorted value as the first index value of each environmental parameter; obtain randomly generated environmental parameters based on the first index value of each environmental parameter; group the parameters found based on their ranking according to the first index value, for example, determine the top-ranked environmental parameters as the first subset of environmental parameter data, and the rest as the second subset of environmental parameter data.

[0150] The second priority strategy can be used to randomly group the environmental parameter dataset as follows: assign priority 1 to the first environmental parameter and priority 2 to the other environmental parameters in the dataset besides the first environmental parameter; multiply the priority of each environmental parameter by a random number to obtain the second update priority of each environmental parameter; sort the second update priorities in descending order according to the priority values, and use the sorted value as the second index value of each environmental parameter; obtain randomly generated environmental parameters based on the second index value of each environmental parameter; group the parameters found by the first index value according to their ranking, for example, determine the top-ranked environmental parameters as the third subset of environmental parameter data, and the rest as the fourth subset of environmental parameter data.

[0151] The optimization results under different priority strategies are obtained. Based on the actual application requirements, the better or more suitable target optimization result is selected from the optimization results to obtain the target environment parameters.

[0152] For example, the root mean square error of the models trained based on the first and second optimal parameter sets is calculated respectively, and the optimal parameter set corresponding to the model with the smaller root mean square error is determined as the optimal parameter set.

[0153] In some embodiments, the method may further include:

[0154] Determine the first optimized sample dataset and the second optimized sample dataset;

[0155] The target correction model is trained using the first optimized sample dataset and the second optimized sample dataset, respectively, to obtain the first model and the second model.

[0156] Determine the error values ​​for the first model and the second model respectively;

[0157] Compare the error values ​​of the first model and the second model;

[0158] The model with the smaller error value is identified as the path loss model;

[0159] The first optimized sample dataset is determined from the initial environmental parameter dataset using a pre-defined correlation analysis algorithm and a pre-defined verification algorithm, while the second optimized sample dataset is determined from the environmental parameter sample dataset using a pre-defined greedy algorithm.

[0160] It should be noted that, firstly, according to the aforementioned method, a first optimized sample dataset is determined from the initial environmental parameter dataset by using a preset correlation analysis algorithm and a preset verification algorithm, and then the target calibration model is further optimized using the first optimized sample dataset to obtain the first model.

[0161] Meanwhile, in accordance with the aforementioned method, a second optimized sample dataset is determined from the environmental parameter sample dataset using a pre-defined greedy algorithm, and the target correction model is further optimized using the second optimized sample dataset to obtain the second model.

[0162] The error values ​​of the first model and the second model are calculated separately. The error values ​​of the first model and the second model are compared, and the model with the smaller error value is determined as the path loss model.

[0163] S104. Optimize the target correction model based on the target environment parameters to obtain the path loss model.

[0164] It should be noted that after selecting the target environmental parameters from the environmental parameter dataset, the target calibration model can be optimized based on these parameters to obtain the path loss model. Specifically, the target calibration model can be trained using the target environmental parameter sample dataset corresponding to the target environmental parameters to obtain the path loss model.

[0165] It should also be noted that in some cases, the target correction model is in a state of continuous iterative updates. In this case, the target correction model obtained from the most recent update can be directly determined as the path loss model.

[0166] Thus, the path loss model created by the path loss modeling method provided in this application is a millimeter-wave channel path loss model under different environmental parameters (such as air density, water vapor partial pressure, air temperature, atmospheric pressure, humidity, transmitter altitude, oxygen content, and rainfall). This not only improves the versatility of the path loss model but also reduces the time complexity of the modeling algorithm. Specifically, this paper quantifies the impact of different types of environmental parameters on path loss from the perspective of correlation analysis, removes environmental parameters that are not related to path loss by combining significance level detection, further optimizes the model parameters through a genetic algorithm, and selects environmental parameters through an improved greedy algorithm. While fully considering the impact of environmental parameters on path loss (such as millimeter-wave channel path loss), environmental parameters are also screened according to their correlation with path loss and their improvement effect on the model. Finally, only environmental parameters that have a good improvement effect on the path loss model are added to the path loss model. The resulting path loss model has strong applicability and can adapt to a variety of different scenarios. The improved greedy algorithm adopts a two-way priority random strategy to determine the type of environmental parameter, which reduces the time complexity of the traditional greedy algorithm and the possibility of the greedy algorithm getting trapped in local optima. In addition, the path loss modeling method provided in this application has relatively low complexity and does not rely on empirical parameters, so it has high versatility.

[0167] In summary, this embodiment provides a path loss modeling method. It involves determining an initial model and an environmental parameter dataset, where the environmental parameter dataset includes at least one candidate environmental parameter. A genetic algorithm is used to optimize the initial model, resulting in a target calibration model. The target calibration model is then adjusted based on each candidate environmental parameter in the dataset, yielding the adjusted error value for each candidate environmental parameter. A pre-defined greedy algorithm is used to determine the target environmental parameter based on this error value. Finally, the target calibration model is optimized based on the target environmental parameter to obtain the path loss model. Thus, when establishing the path loss model, both genetic and greedy algorithms are used for model optimization and environmental parameter selection, fully considering the impact of each environmental parameter on the path loss model. The environmental parameter with the greater impact on the model is selected for calibration, thereby reducing the modeling complexity of the path loss model and improving its modeling accuracy and applicability.

[0168] In another embodiment of this application, participants Figure 2This illustrates a detailed flowchart of a path loss modeling method provided in an embodiment of this application. Figure 2 As shown, the method may include:

[0169] S201. Correlation analysis to screen environmental parameters.

[0170] The embodiments of this application first quantify the impact of environmental parameters (such as air density, water vapor partial pressure, air temperature, atmospheric pressure, humidity, transmitter altitude, oxygen content, and rainfall) on path loss from the perspective of correlation coefficients. Then, a two-sample t-test is used to test the significance level between each environmental parameter and path loss. Environmental parameters that are not related to path loss are removed, and only environmental parameters that are significant at the 95% statistical level are added to the calibration model.

[0171] In this embodiment of the application, when quantitatively analyzing the sensitivity of millimeter-wave channel propagation path loss to various environmental parameters, a correlation coefficient is used to measure the degree of influence of different environmental parameters on path loss. The formula for calculating the correlation coefficient R is:

[0172]

[0173] Where, x i For environmental parameters, y i For path loss, and For x i and y i The mean.

[0174] The two-sample t-test is a parametric test, also known as the independent samples t-test. It is commonly used to compare the means of two independent groups to determine whether there is statistical evidence that there is a significant difference between the means of different groups. Accepting or rejecting the hypothesis cannot be 100% accurate; therefore, a fixed significance level is needed as the standard for judging whether the hypothesis is valid. In this embodiment, preferably, the significance level is set to α = 0.05.

[0175] S202. Import the sample dataset into the BP neural network and normalize it.

[0176] Import the sample data required for training and prediction of the BP neural network, and normalize the sample data.

[0177] S203. Determine the number of hidden layer neurons in the BP neural network.

[0178] In this step, the network structure of the BP neural network is constructed, and the structural characteristics of the neural network are determined. See also Figure 3 This illustrates a schematic diagram of a BP neural network structure provided in an embodiment of this application. Figure 3As shown, the neural network consists of three layers: an input layer, a hidden layer, and an output layer. The number of neurons in the hidden layer is determined by the following formula:

[0179]

[0180] Where M is the number of neurons in the hidden layer, m and n are the number of neurons in the input layer and the output layer, and a is selected between 0 and 10. In the embodiments of this application, a is preferably 5.

[0181] The input parameters of the BP neural network are: carrier frequency, straight-line distance between transmitter and receiver, transmit power, and other environmental parameters. The output parameter of the BP neural network is: receive power.

[0182] In this embodiment, the activation function used from the input layer to the hidden layer is the tansig function, and the activation function from the hidden layer to the output layer is y = x. The analytical expression of the tansig function is as follows:

[0183]

[0184] S204. Initialize the weights and thresholds of the BP neural network.

[0185] S205. Determine the initial model.

[0186] It should be noted that, in the embodiments of this application, the method for initializing the weights and thresholds of the BP neural network can be constant initialization, Gaussian distribution initialization, uniform distribution initialization, bilinear initialization, random initialization, etc. This application does not specifically limit the method in this regard.

[0187] It should also be noted that when determining the initial model, only the carrier frequency, the straight-line distance between the transmitter and receiver, and the transmit power are used as input parameters to train the BP neural network to obtain the initial model (also known as the initial calibration model). The step of determining the initial model can also be called initializing the input parameters.

[0188] S206. Calculate the optimal weights and thresholds for the current neural network structure.

[0189] It should be noted that in this embodiment, an improved genetic algorithm is used to optimize and obtain the optimal solution under the current BP neural network structure, thereby obtaining the optimal weights and thresholds. For example, by optimizing using a genetic algorithm, the optimal weights and thresholds of the initial model can be obtained. Specifically, see [link to relevant documentation]. Figure 4It illustrates a flowchart of an embodiment of this application for optimizing the weights and thresholds of a BP neural network, as shown below. Figure 4 As shown, the process of optimizing the weights and thresholds of the BP neural network may include:

[0190] S401, Initialize the population.

[0191] S402. Determine the first generation population.

[0192] It should be noted that after initializing the population, we obtain the first generation population, also known as the initial population. At this point, we denote the population generation G = 0.

[0193] S403. Calculate individual fitness.

[0194] S404. Sort individuals according to their fitness.

[0195] All individuals are sorted according to their fitness level.

[0196] S405, First operation selection.

[0197] Replication is carried out in a certain proportion, that is, selecting a number of individuals with the highest fitness from the current population for replication. For example, the two individuals with the highest fitness in the current population are structurally replicated into the population to be bred.

[0198] S406, Mutation operation.

[0199] Individuals in the mating population are subjected to mutation operations according to a preset mutation probability. In this embodiment of the application, the mutation probability is preferably 0.1.

[0200] The mutated individuals are replicated according to the principle of replicating four copies of superior individuals and not replicating inferior individuals, resulting in a replication group. In the embodiments of this application, superior individuals may include individuals that result in higher accuracy or lower error values ​​for the BP neural network, while inferior individuals may include individuals that do not contribute to the accuracy of the BP neural network, or result in lower accuracy or higher error values ​​for the BP neural network.

[0201] S407, Second selection operation.

[0202] S408, cross operation.

[0203] Two individuals (or any number of individuals) are randomly selected from the replication group, and then these two individuals are crossbred multiple times according to a preset crossbring probability. In this embodiment, the preset crossbring probability is preferably 0.3.

[0204] S409, Third selection operation.

[0205] S4010, obtain the next generation population.

[0206] From the results of the crossover, select the best individual and store it in the new population to obtain the next generation population, i.e., the new generation population.

[0207] Increment the population generation by one, and denote the population generation G = G + 1.

[0208] S4011. Determine whether the optimal individual meets the preset optimal conditions.

[0209] S4012. Determine the optimal weights and thresholds for the new generation population.

[0210] It should be noted that if the optimal individual meets the termination condition, then the optimal weight and threshold are determined, and the optimal individual included in the new generation population is the optimal weight and threshold; otherwise, return to step S403 and continue to execute this method process on the current population until the optimal weight and threshold are found.

[0211] It should also be noted that, in the embodiments of this application, the determination of whether the optimal individual meets the preset optimal conditions can be made in the following two ways:

[0212] Method 1: Calculate the accuracy of the BP neural network created by the optimal individual, and compare this value with a preset threshold to determine whether the optimal individual meets the preset conditions. For example, calculate the loss function value of the BP neural network; if the loss function value is less than the preset loss function threshold, then the preset conditions are met. Alternatively, calculate the root mean square error (RMSE) of the BP neural network; if the RMSE is less than the preset RMSE threshold, then the preset conditions are met. Or, calculate the reduction in RMSE of the BP neural network; if the reduction in RMSE is less than the absolute value of the preset error threshold, then the preset optimal conditions are met, etc. Otherwise, return to step S503 and continue this process for the current population until the preset conditions are met, and the optimal weights and thresholds are obtained.

[0213] It should also be noted that, in the embodiments of this application, the root mean square error (RMSE) of the BP neural network refers to the root mean square error between the received power output by the current model and the actual received power. The reduction in the RMSE of the BP neural network refers to the difference between the RMSE of the current model and the RMSE of the previous model.

[0214] Method 2: Determine whether the population generation G has reached the preset population generation threshold GEN.

[0215] If G>GEN, then the preset conditions are met, and the best individual in the current population is determined as the optimal weight and threshold; otherwise, return to step S503 and continue to execute this process on the current population until the preset conditions are met and the optimal weight and threshold are obtained.

[0216] The final target correction model (also known as the neural network correction model) is obtained.

[0217] S207. Calculate the reduction in root mean square error.

[0218] It should be noted that calculating the root mean square error reduction means calculating the root mean square error of the existing model and the root mean square error of the model obtained after inputting environmental parameters into the target calibration model (i.e., the second calibration model in the aforementioned embodiment), and then determining the difference between the root mean square error of the existing model and the root mean square error of the second calibration model as the root mean square error reduction (which can be represented by ΔMSE). Here, the existing model refers to the model without inputting environmental parameters (i.e., the target calibration model obtained after step S206).

[0219] In a specific example, the root mean square error of the computational model can refer to the root mean square error between the received power output by the computational model and the actual receiver rate.

[0220] S208. Determine whether △MSE is greater than 1db.

[0221] For S208, the question is whether △MSE > 1db. If the result is yes, then step S209 is executed to determine whether the next environmental parameter is the optimal parameter; otherwise, the process ends and the target environmental parameter is obtained.

[0222] S209. Determine the next optimal environmental parameters using an improved greedy algorithm.

[0223] In this embodiment, the improved greedy algorithm is the same as the preset greedy algorithm in the foregoing embodiments.

[0224] It should be noted that the remaining environmental parameters are added to the target correction model in sequence, the root mean square error is calculated, and several environmental parameters with the best improvement effect (for example, the environmental parameter with the best improvement effect) are selected and added to the correction model. A genetic algorithm is used to obtain the optimal weights and thresholds of the BP neural network under the current input parameters.

[0225] Furthermore, for S209, an improved greedy algorithm (such as a bidirectional priority-based randomized greedy algorithm) can be used to filter environmental parameters. Greedy algorithms can quickly determine local optima; however, the sum of local optima does not necessarily equal the global optimum. Greedy algorithms are prone to prematurely getting trapped in local optima. In the bidirectional priority-based randomized greedy algorithm, a bidirectional priority-based randomized grouping strategy can be introduced to expand the search space of the greedy search, reducing the possibility of the greedy algorithm getting trapped in local optima.

[0226] Based on a bidirectional priority-based randomization grouping strategy, environmental parameters are divided into a correlation analysis group and a greedy algorithm group. The correlation analysis group selects the optimal environmental parameters based on their correlation with the prediction error. The greedy algorithm group uses a traditional greedy strategy to iterate through and add environmental parameters to the neural network, determining the environmental parameters that best optimize the BP neural network. Finally, the optimization effects of the two groups of optimal environmental parameters are compared, and the environmental parameters with better optimization effects are added to the neural network.

[0227] In a more specific example, such as Figure 5 As shown, it illustrates a detailed flowchart of another path loss modeling method provided in an embodiment of this application, as follows: Figure 5 As shown, the method may include:

[0228] S501. Initialize the BP neural network.

[0229] S502. Construct an initial model using carrier frequency, propagation distance, and transmission power as inputs.

[0230] The propagation distance is the straight-line distance between the transmitter and the receiver.

[0231] It should be noted that steps S501 to S502 are the same as... Figure 2 Steps S201 to S205 correspond to each other and mainly involve creating the initial model.

[0232] S503. Determine the optimal weights and thresholds.

[0233] The description of determining the optimal weights and thresholds is consistent with that described above, and will not be repeated here.

[0234] S504. Calculate the reduction in root mean square error.

[0235] Here, the reduction in root mean square error is represented by ΔMSE.

[0236] S505. Determine whether △MSE is greater than 1db.

[0237] For step S505, i.e., determining whether △MSE > 1db? If the determination result is yes, then proceed to step S506; otherwise, end the method flow and obtain the target environment parameters.

[0238] S506. A priority-based randomization grouping strategy is used to divide environmental parameters into a correlation analysis group and a greedy algorithm group.

[0239] Given that different environmental parameters have varying impacts on modeling, the random grouping strategy employs a priority-based random algorithm, followed by a repetition with the opposite priority to expand the search space. The optimal result is selected from these two different priority-based random strategies (a bidirectional priority-based random greedy algorithm), thus increasing the search space of the greedy algorithm. For example:

[0240] The rainfall was assigned a priority of 2 in the first round, and the remaining environmental parameters were assigned a priority of 1. A priority-based random algorithm was used to group the parameters.

[0241] The second time, other environmental parameters were assigned a priority of 2, and rainfall was assigned a priority of 1. A priority-based random algorithm was used to group the data.

[0242] The priority-based random algorithm is calculated as follows:

[0243] 1. Multiply the parameter priority by a random number (note that all random numbers have the same start and end range) to get the new priority.

[0244] 2. Sort the new priorities in descending order to obtain the index of the corresponding parameter.

[0245] 3. Obtain randomly generated environment parameters based on the index, and group the parameters found by the index according to their ranking.

[0246] We obtained the greedy algorithm group and the related analysis group.

[0247] S507. Determine the optimal parameter A within the group of the greedy algorithm group.

[0248] S508. Determine the optimal parameter B within the correlation analysis group.

[0249] S509. Determine the overall optimal parameters.

[0250] The correlation analysis group sorts the environmental parameters according to their correlation with the prediction error and selects the optimal parameters within the group, thus obtaining the optimal parameters A within the group. The greedy algorithm group uses the traditional greedy strategy to traverse and add environmental parameters to the neural network, and determines the environmental parameters that have the best optimization effect on the BP neural network, thus obtaining the optimal parameters B within the group.

[0251] By comparing the optimization effects of the two sets of optimal environmental parameters, the overall optimal parameters are obtained, and the environmental parameters with better optimization effects are selected and added to the neural network.

[0252] Then continue to optimize the current model to determine the optimal weights and thresholds, and calculate the reduction in root mean square error (repeated steps) until the result of the new calibration model is not significantly optimized (judgment criterion: the reduction in root mean square error is not greater than 1dB), and obtain the target parameter set.

[0253] Based on the target parameter set, a path loss model is obtained. Since the path loss model in this embodiment is trained based on multiple environmental parameters, it is preferably applied to millimeter-wave path loss modeling. Therefore, it can also be called a multi-environment parameter millimeter-wave channel path loss model.

[0254] It should also be noted that, in the embodiments of this application, the step numbers of S507 and S508 do not constitute a restriction on the order of implementation. S507 and S508 can be performed simultaneously or sequentially.

[0255] Furthermore, the method also includes the following steps:

[0256] Calculate and store the weight matrices W1 and W2 of the BP neural network, as well as the relevant thresholds B1 and B2;

[0257] Based on W1, W2, B1, and B2, assign values ​​to the weights of the BP neural network and the net structure, and save the relevant matrices.

[0258] It should also be noted that, in this embodiment, when using a greedy algorithm to filter environmental parameters, further filtering is performed on the environmental parameters already filtered by the preset correlation analysis algorithm and the preset verification algorithm. In this embodiment, a greedy algorithm can also be used to filter all environmental parameters (i.e., the initial environmental parameter dataset), and the filtered environmental parameters can be compared with those filtered by the preset correlation analysis algorithm and the preset verification algorithm to determine the target environmental parameters.

[0259] In summary, the path loss modeling method provided in this application can be applied to millimeter-wave channel path loss modeling under different environmental parameters (such as air density, water vapor partial pressure, air temperature, atmospheric pressure, humidity, transmitter altitude, oxygen content, and rainfall), improving versatility and reducing the time complexity of the modeling algorithm.

[0260] The path loss modeling method in this application introduces an improved greedy algorithm based on a BP neural network. This algorithm sequentially determines the input environment parameters of the BP neural network, reducing the time complexity of the traditional greedy algorithm. Furthermore, in this application embodiment, the improved genetic algorithm ensures that the offspring retain the best individuals from the parent generation during each evolutionary process. This allows for the search for better samples in the "family direction with high fitness patterns as ancestors," guaranteeing that the global optimum can be found and the optimal BP neural network structure under the current input parameters can be obtained. This addresses the problem that BP neural networks cannot guarantee convergence to the global minimum. Therefore, by introducing the improved greedy algorithm and genetic algorithm to optimize the time complexity of the algorithm for determining the optimal input environment parameters of the BP neural network, the overall time complexity is improved.

[0261] In establishing a multi-environmental-parameter millimeter-wave channel path loss model, this application first employs a correlation analysis algorithm to select parameters related to millimeter-wave channel path loss as candidate environmental parameters from water vapor partial pressure, air temperature, atmospheric pressure, humidity, transmitter altitude, and rainfall. Then, an improved greedy algorithm is used to sequentially find influencing factors that improve the accuracy of path loss prediction, ultimately establishing an effective millimeter-wave channel path loss model. In short, this modeling method may include the following steps:

[0262] Step 1: First, quantify the impact of environmental parameters (air density, water vapor partial pressure, air temperature, atmospheric pressure, humidity, transmitter altitude, oxygen content, and rainfall) on path loss from the perspective of correlation coefficients. Then, use a two-sample t-test to test the significance level between the two sets of data, remove parameters that are not related to path loss, and only add environmental parameters that are significant at the 95% statistical level to the calibration model.

[0263] In this application embodiment, when quantitatively analyzing the sensitivity of millimeter-wave channel propagation path loss to various environmental parameters, the correlation coefficient is used to measure the degree of influence of different environmental parameters on path loss. The formula for calculating the correlation coefficient R is as shown in equation (1).

[0264] The two-sample t-test is a parametric test, also known as the independent samples t-test. It is commonly used to compare the means of two independent groups to determine whether there is statistical evidence of a significant difference between the means of different groups. Accepting or rejecting the hypothesis cannot be 100% accurate; therefore, a fixed significance level is needed as the criterion for judging whether the hypothesis is valid. In this embodiment, the preferred significance level is α = 0.05.

[0265] Step 2: Import the sample data required for training and prediction of the BP neural network (i.e., the first sample dataset and the environmental parameter sample dataset), and normalize the sample data.

[0266] Step 3: Construct the network structure of the BP neural network and determine its structural characteristics. The neural network has a three-layer structure, with one input layer, one hidden layer, and one output layer. The number of neurons in the hidden layer is determined by the following formula (2). The input parameters of the BP neural network are: carrier frequency, straight-line distance between the transmitter and receiver, average transmit power, and other environmental parameters. The output parameter of the BP neural network is: receiver power.

[0267] In this embodiment, the activation function used from the input layer to the hidden layer is the tansig function, and the activation function from the hidden layer to the output layer is y = x. The analytical expression of the tansig function is shown in equation (3).

[0268] Step 4: Initialize the weights and thresholds of the BP neural network.

[0269] Step 5: Using only the carrier frequency, the straight-line distance between the transmitter and receiver, and the transmitter's transmit power as inputs to the BP neural network, perform BP neural network correction to obtain the initial model.

[0270] Step 6: In this embodiment, an improved genetic algorithm is used to optimize and obtain the optimal solution under the current BP neural network structure, thereby obtaining the optimal weights and thresholds to obtain the optimized model. The steps for obtaining the optimal weights and thresholds are briefly described below:

[0271] First, in the initial population, all individuals are sorted according to their fitness.

[0272] Secondly, the two individuals with the highest fitness in the current population are replicated in a certain proportion (that is, the two individuals with the highest fitness in the current population are completely replicated into the population to be mated).

[0273] Next, the individuals are mutated with a mutation probability of 0.1. Then, the individuals are replicated according to the principle of replicating 4 copies of superior individuals and not replicating inferior individuals, thus obtaining a replication group.

[0274] Next, two individuals are randomly selected from the replication group, and these two individuals are cross-crossed multiple times. From the results, the best individual is selected and stored in the new population.

[0275] Finally, if the termination condition is met, stop; otherwise, return to the step of sorting individuals by fitness until the optimal weights and thresholds are found.

[0276] Step 7: Calculate the root mean square error of the existing model (the root mean square error between the model's output received power and the actual received power).

[0277] Step 8: Add the remaining environmental parameters to the optimization model in sequence, calculate the root mean square error, select the parameter with the best improvement effect to add to the optimization model, and use a genetic algorithm to obtain the optimal weights and thresholds of the BP neural network under the current input parameters.

[0278] This step uses an improved greedy algorithm (bidirectional priority-based randomized greedy algorithm) to filter parameters. Greedy algorithms can quickly determine local optima; however, a local optimum is not necessarily the global optimum. Greedy algorithms are prone to getting trapped in local optima too early. Introducing a bidirectional priority-based randomization strategy expands the search space of the greedy search, reducing the likelihood of it getting trapped in local optima.

[0279] The sample data from the correlation analysis group are sorted by correlation and the optimal parameters are selected. The sample data from the greedy algorithm group are then processed using a greedy strategy to iterate and add parameters to the BP neural network, determining the environmental parameters that best optimize the BP neural network. Finally, the optimization effects of the two optimal environmental parameter groups are compared, and the environmental parameter with the better optimization effect is added to the neural network.

[0280] Given that different environmental parameters have different effects on modeling, the bidirectional priority randomization strategy adopts a priority-based random algorithm and then loops again with the opposite priority to expand the search space. The optimal result is selected from the two different priority random strategies (bidirectional priority random greedy algorithm), thus increasing the search space of the greedy algorithm.

[0281] For example, using a priority-based randomization strategy, the rainfall is assigned a priority of 2 in the first instance, and the remaining environmental parameters are assigned a priority of 1.

[0282] A priority-based randomization strategy is adopted. In the second iteration, other environmental parameters are assigned a priority of 2, while rainfall is assigned a priority of 1. The priority-based randomization algorithm is as follows:

[0283] (a) Multiply the priority of the environmental parameter by a random number (note that all random numbers have the same start and end range) to obtain a new priority.

[0284] (b) Sort the new priorities in descending order to obtain the index of the corresponding environment parameters.

[0285] (c) Obtain randomly generated environment parameters based on the index.

[0286] Step 9: Repeat the above optimization until the result of the latest optimized model is no longer significantly improved (judgment criterion: the reduction in root mean square error is less than 1dB).

[0287] Step 8 may include the following steps:

[0288] Step 8.1: Calculate and save the weight matrices W1 and W2 of the BP neural network, as well as the relevant thresholds B1 and B2;

[0289] Step 8.2: Assign values ​​to the weights of the BP neural network and the net structure, and save the relevant matrices.

[0290] The above embodiments provide a detailed explanation of the specific implementation of the aforementioned embodiments. It can be seen that an improved greedy algorithm combined with a genetic algorithm optimizes the path loss modeling based on a BP neural network. Furthermore, the improved greedy algorithm employs a bidirectional priority-based random strategy to determine the selection parameter type, which reduces both the time complexity of the traditional greedy algorithm and the possibility of it getting trapped in local optima. Compared to traditional empirical models, the embodiments of this application consider millimeter-wave channel propagation under various scenarios, fully taking into account the impact of environmental parameters on millimeter-wave channel path loss, thus exhibiting strong applicability and effectively improving the accuracy and applicability of millimeter-wave channel path loss modeling. Compared to deterministic models, the embodiments of this application have relatively low complexity and do not rely on empirical parameters, exhibiting high versatility.

[0291] In another embodiment of this application, see [reference needed]. Figure 6 This illustrates a schematic diagram of the composition of a path loss modeling device 40 provided in an embodiment of this application. Figure 6 As shown, the path loss modeling device 40 may include a first determining unit 401, a first optimizing unit 402, a second determining unit 403, and a second optimizing unit 404, wherein,

[0292] The first determining unit 401 is configured to determine an initial model and an environmental parameter dataset; wherein the environmental parameter dataset includes at least one candidate environmental parameter;

[0293] The first optimization unit 402 is configured to use a genetic algorithm to optimize the initial model to obtain the target correction model;

[0294] The second determining unit 403 is configured to adjust the target correction model according to each candidate environmental parameter in the environmental parameter dataset, obtain the adjusted error value of the target correction model corresponding to the candidate environmental parameter, and determine the target environmental parameter according to the error value using a preset greedy algorithm;

[0295] The second optimization unit 404 is configured to optimize the target correction model based on the target environment parameters to obtain a path loss model.

[0296] In some embodiments, the first determining unit 401 is further configured to: acquire a sample dataset; normalize the sample dataset to obtain a normalized sample dataset; and train a preset neural network model using the normalized sample dataset to obtain the initial model.

[0297] In some embodiments, the first determining unit 401 is further configured to: acquire an initial environmental parameter dataset; calculate the influence of each initial candidate environmental parameter in the initial environmental parameter dataset relative to path loss based on a preset correlation analysis algorithm, and determine the influence of each initial candidate environmental parameter relative to path loss; perform a significance level test on each initial candidate environmental parameter and path loss in the initial environmental parameter dataset using a preset test algorithm, and determine the significance level test result of each initial candidate environmental parameter relative to path loss; determine initial candidate environmental parameters that are not related to path loss based on the influence and the significance level test result; and remove initial candidate environmental parameters that are not related to path loss from the initial environmental parameter dataset to obtain the environmental parameter dataset.

[0298] In some embodiments, the first optimization unit 402 is further configured to: determine an initial population set based on the initial model; wherein the initial population set includes at least one individual; calculate the fitness of all individuals in the initial population set, sort all individuals in the initial population set according to fitness, and select at least one individual with the highest fitness to form a population set to be bred; perform mutation and crossover processing on the population set to be bred, and select a preset number of target individuals from the mutation and crossover results to form a new population set; generate a first calibration model based on the new population set; if the first calibration model meets a preset termination condition, determine the first calibration model as the target calibration model; and if the first calibration model does not meet the preset termination condition, determine the new population set as the initial population set, and return to calculating the fitness of all individuals in the initial population set and repeat the process until the first calibration model meets the preset termination condition and the first calibration model is determined as the target calibration model.

[0299] In some embodiments, the number of target environmental parameters is at least one, and the second determining unit 403 is further configured to adjust the target correction model using each candidate environmental parameter in the environmental parameter dataset to obtain an adjusted correction model corresponding to each candidate environmental parameter, and determine the error value of the adjusted correction model corresponding to each candidate environmental parameter; and select at least one target error value from the error values, and determine the target environmental parameter based on the at least one target error value.

[0300] In some embodiments, the second determining unit 403 is further configured to: determine a first error value of the target calibration model; adjust the target calibration model using the i-th candidate environmental parameter in the environmental parameter dataset to obtain a second calibration model; wherein 1≤i≤N, and N is the number of candidate environmental parameters in the environmental parameter dataset; determine a second error value of the second calibration model; and if the difference between the first error value and the second error value is greater than a preset error threshold, determine the i-th candidate environmental parameter as the target environmental parameter; increment i by one, return to adjusting the target calibration model using the i-th candidate environmental parameter in the environmental parameter dataset to obtain a second calibration model, and repeat the process until i equals N or the difference between the first error value and the second error value is not greater than the preset error threshold, thereby determining all target environmental parameters.

[0301] In some embodiments, the second determining unit 403 is further configured to: randomly group the environmental parameter dataset to obtain a first set of environmental parameter data subsets and a second set of environmental parameter data subsets; adjust the target calibration model using each candidate environmental parameter in the first set of environmental parameter data subsets to obtain a third calibration model corresponding to each candidate environmental parameter, determine a third error value of the third calibration model corresponding to each candidate environmental parameter, determine a first target environmental parameter based on the correlation between the third error value and the prediction error, adjust the target calibration model using each candidate environmental parameter in the second set of environmental parameter data subsets to obtain a fourth calibration model corresponding to each candidate environmental parameter, determine a fourth error value of the fourth calibration model corresponding to each candidate environmental parameter, determine a second target environmental parameter based on the fourth error value using a conventional greedy algorithm; determine a first optimization result of the first target environmental parameter on the target calibration model, and determine a second optimization result of the second target environmental parameter on the target calibration model; select a target optimization result from the first optimization result and the second optimization result, and determine the environmental parameter input to the target calibration model under the target optimization result as the target environmental parameter.

[0302] In some embodiments, the preset greedy algorithm is a bidirectional priority-based random greedy algorithm. The second determining unit 403 is further configured to: randomly group the environmental parameter dataset using a first priority strategy to obtain a first set of environmental parameter data subsets and a second set of environmental parameter data subsets; and randomly group the environmental parameter dataset using a second priority strategy to obtain a third set of environmental parameter data subsets and a fourth set of environmental parameter data subsets; wherein the first priority strategy and the second priority strategy are opposite priorities to each other; and determine the target environmental parameters and corresponding optimization results under the first priority strategy based on the first set of environmental parameter data subsets and the second set of environmental parameter data subsets; determine the target environmental parameters and corresponding optimization results under the second priority strategy based on the third set of environmental parameter data subsets and the fourth set of environmental parameter data subsets; and select a target optimization result from the target environmental parameters and corresponding optimization results under the first priority strategy and the target environmental parameters and corresponding optimization results under the second priority strategy, and determine the environmental parameters input to the target correction model under the target optimization result as the target environmental parameters.

[0303] Understandably, in this embodiment, a "unit" can be a portion of a circuit, a portion of a processor, a portion of a program or software, etc., and can also be a module or a non-modular component. Furthermore, the components in this embodiment can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional module.

[0304] If the integrated unit is implemented as a software functional module and not sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this embodiment, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the method described in this embodiment. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0305] Therefore, this embodiment provides a computer storage medium storing a computer program that, when executed by at least one processor, implements any of the path loss modeling methods described in the foregoing embodiments.

[0306] Based on the above-described composition of a path loss modeling device 40 and its computer storage medium, see [link to relevant documentation]. Figure 7 This illustrates a schematic diagram of the structural composition of an electronic device 50 provided in an embodiment of this application. For example... Figure 7 As shown, it may include: a communication interface 501, a memory 502, and a processor 503; the various components are coupled together through a bus system 504. It is understood that the bus system 504 is used to implement communication between these components. In addition to a data bus, the bus system 504 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 7 All buses are labeled as bus system 504. Among them, communication interface 501 is used for receiving and sending signals during information exchange with other external network elements;

[0307] Memory 502 is used to store computer programs that can run on processor 503;

[0308] Processor 503, when running the computer program, performs the following:

[0309] Determine the initial model and environmental parameter dataset; wherein the environmental parameter dataset includes at least one candidate environmental parameter;

[0310] The initial model is optimized using a genetic algorithm to obtain the target correction model;

[0311] The target calibration model is adjusted according to each candidate environmental parameter in the environmental parameter dataset to obtain the adjusted error value of the target calibration model corresponding to the candidate environmental parameter. The target environmental parameter is then determined based on the error value using a preset greedy algorithm.

[0312] The calibration model is optimized based on the target environmental parameters to obtain the path loss model.

[0313] It is understood that the memory 502 in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), and Direct Rambus RAM (DRRAM). The memory 502 of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0314] The processor 503 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 503 or by instructions in software form. The processor 503 can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory 502, and the processor 503 reads the information in memory 502 and, in conjunction with its hardware, completes the steps of the above method.

[0315] It is understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described herein, or combinations thereof.

[0316] For software implementation, the techniques described herein can be achieved through modules (e.g., procedures, functions, etc.) that perform the functions described herein. The software code can be stored in memory and executed by a processor. The memory can be implemented within the processor or externally.

[0317] Alternatively, as another embodiment, the processor 503 is also configured to perform the method described in any of the foregoing embodiments when running the computer program.

[0318] Based on the composition of the path loss modeling device 40 described above, see [link to documentation]. Figure 8 This illustrates a schematic diagram of the structural composition of another electronic device 50 provided in an embodiment of this application. For example... Figure 8 As shown, the electronic device 50 includes at least the path loss modeling device 40 described in any of the foregoing embodiments.

[0319] For electronic device 50, when establishing the path loss model, genetic algorithm and greedy algorithm are used to optimize the model and select environmental parameters respectively. The influence of each environmental parameter on the path loss model is fully considered, and the environmental parameter with a greater influence on the model is selected to correct the model. This not only reduces the modeling complexity of the path loss model, but also improves the modeling accuracy and applicability of the model.

[0320] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application.

[0321] It should be noted that, in this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0322] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0323] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.

[0324] The features disclosed in the several product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.

[0325] The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method or device embodiments.

[0326] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A path loss modeling method, characterized in that, The method includes: Determine the initial model and environmental parameter dataset; wherein the environmental parameter dataset includes at least one candidate environmental parameter; The initial model is optimized using a genetic algorithm to obtain the target correction model; The target calibration model is adjusted according to each candidate environmental parameter in the environmental parameter dataset to obtain the adjusted error value of the target calibration model corresponding to the candidate environmental parameter. The target environmental parameter is then determined based on the error value using a preset greedy algorithm. The target calibration model is optimized based on the target environmental parameters to obtain a path loss model; wherein the preset greedy algorithm is a bidirectional priority-based random greedy algorithm; the step of adjusting the target calibration model according to each candidate environmental parameter in the environmental parameter dataset to obtain the adjusted error value of the target calibration model corresponding to the candidate environmental parameter; and determining the target environmental parameters based on the error value using the preset greedy algorithm, including: The environmental parameter dataset is randomly grouped using a first priority strategy to obtain a first subset of environmental parameter data and a second subset of environmental parameter data; and The environmental parameter dataset is randomly grouped using a second priority strategy to obtain a third set of environmental parameter data subsets and a fourth set of environmental parameter data subsets; wherein the first priority strategy and the second priority strategy are opposite priorities to each other; Based on the first set of environmental parameter data subsets and the second set of environmental parameter data subsets, determine the target environmental parameters and corresponding optimization results under the first priority strategy; Based on the third set of environmental parameter data subsets and the fourth set of environmental parameter data subsets, determine the target environmental parameters and corresponding optimization results under the second priority strategy; From the target environment parameters and corresponding optimization results under the first priority strategy and the target environment parameters and corresponding optimization results under the second priority strategy, select the target optimization result, and determine the environment parameters input to the target correction model under the target optimization result as the target environment parameters.

2. The method according to claim 1, characterized in that, Determine the environmental parameter dataset, including: Obtain the initial environment parameter dataset; Based on a preset correlation analysis algorithm, the influence of each initial candidate environmental parameter in the initial environmental parameter dataset on the path loss is calculated to determine the influence of each initial candidate environmental parameter on the path loss. A preset testing algorithm is used to perform a significance level test on each initial candidate environmental parameter and path loss in the initial environmental parameter dataset, and the significance level test result of each initial candidate environmental parameter relative to the path loss is determined. Based on the impact degree and the significance level test results, initial candidate environmental parameters that are not related to the path loss are determined; The initial candidate environmental parameters that are not related to the path loss are removed from the initial environmental parameter dataset to obtain the environmental parameter dataset.

3. The method according to claim 1, characterized in that, The optimization of the initial model using a genetic algorithm to obtain the target correction model includes: Based on the initial model, an initial population set is determined; wherein the initial population set includes at least one individual; Calculate the fitness of all individuals in the initial population set, sort all individuals in the initial population set according to fitness, and select at least one candidate individual with the highest fitness to form a population set to be mated. The population set to be mated is subjected to mutation and crossover processing, and a preset number of target individuals are selected from the mutation and crossover results to form a new population set; Based on the new population set, a first calibration model is generated; If the first correction model meets the preset termination condition, then the first correction model is determined as the target correction model; If the first calibration model does not meet the preset termination condition, the new population set is determined as the initial population set, and the calculation of the fitness of all individuals in the initial population set is returned and repeated until the first calibration model meets the preset termination condition, and the first calibration model is determined as the target calibration model.

4. The method according to claim 1, characterized in that, The number of target environmental parameters is at least one; the step of adjusting the target correction model according to each candidate environmental parameter in the environmental parameter dataset to obtain the adjusted error value of the target correction model corresponding to the candidate environmental parameter, and determining the target environmental parameter based on the error value using a preset greedy algorithm, includes: The target calibration model is adjusted using each candidate environmental parameter in the environmental parameter dataset to obtain the adjusted calibration model corresponding to each candidate environmental parameter, and the error value of the adjusted calibration model corresponding to each candidate environmental parameter is determined. At least one target error value is selected from the error values, and the target environmental parameter is determined based on the at least one target error value.

5. The method according to claim 1, characterized in that, The step of adjusting the target calibration model based on each candidate environmental parameter in the environmental parameter dataset to obtain the adjusted error value of the target calibration model corresponding to the candidate environmental parameter, and determining the target environmental parameter based on the error value using a preset greedy algorithm, includes: Determine the first error value of the target correction model; The target calibration model is adjusted using the i-th candidate environmental parameter in the environmental parameter dataset to obtain a second calibration model; where 1≤i≤N, and N is the number of candidate environmental parameters in the environmental parameter dataset; Determine the second error value of the second correction model; If the difference between the first error value and the second error value is greater than a preset error threshold, then the i-th candidate environmental parameter is determined as the target environmental parameter; Increment i by one, and then adjust the target calibration model using the i-th candidate environmental parameter in the environmental parameter dataset to obtain a second calibration model. Repeat this process until i equals N or the difference between the first error value and the second error value is not greater than a preset error threshold, so as to determine all target environmental parameters.

6. The method according to any one of claims 1 to 5, characterized in that, The step of adjusting the target calibration model based on each candidate environmental parameter in the environmental parameter dataset to obtain the adjusted error value of the target calibration model corresponding to the candidate environmental parameter, and determining the target environmental parameter based on the error value using a preset greedy algorithm, includes: The environmental parameter dataset is randomly grouped to obtain a first subset of environmental parameter data and a second subset of environmental parameter data; The target calibration model is adjusted using each candidate environmental parameter in the first subset of environmental parameter data to obtain a third calibration model corresponding to each candidate environmental parameter. A third error value for each candidate environmental parameter's third calibration model is determined. The first target environmental parameter is then determined based on the correlation between the third error value and the prediction error. The target calibration model is adjusted using each candidate environmental parameter in the second set of environmental parameter data subset to obtain a fourth calibration model corresponding to each candidate environmental parameter, and a fourth error value of the fourth calibration model corresponding to each candidate environmental parameter is determined. The second target environmental parameter is determined based on the fourth error value using a traditional greedy algorithm. Determine a first optimization result of the first target environment parameters on the target correction model, and determine a second optimization result of the second target environment parameters on the target correction model; Select the target optimization result from the first optimization result and the second optimization result, and determine the environmental parameters input to the target correction model under the target optimization result as the target environmental parameters.

7. A path loss modeling device, characterized in that, The path loss modeling device includes a first determining unit, a first optimizing unit, a second determining unit, and a second optimizing unit, wherein, The first determining unit is configured to determine an initial model and an environmental parameter dataset; wherein the environmental parameter dataset includes at least one candidate environmental parameter; The first optimization unit is configured to use a genetic algorithm to optimize the initial model to obtain a target correction model; The second determining unit is configured to adjust the target correction model according to each candidate environmental parameter in the environmental parameter dataset, obtain the adjusted error value of the target correction model corresponding to the candidate environmental parameter, and determine the target environmental parameter according to the error value using a preset greedy algorithm; The second optimization unit is configured to optimize the target correction model based on the target environment parameters to obtain a path loss model; The preset greedy algorithm is a bidirectional priority-based random greedy algorithm, and the second determining unit is specifically configured as follows: The environmental parameter dataset is randomly grouped using a first priority strategy to obtain a first subset of environmental parameter data and a second subset of environmental parameter data; and The environmental parameter dataset is randomly grouped using a second priority strategy to obtain a third set of environmental parameter data subsets and a fourth set of environmental parameter data subsets; wherein the first priority strategy and the second priority strategy are opposite priorities to each other; Based on the first set of environmental parameter data subsets and the second set of environmental parameter data subsets, determine the target environmental parameters and corresponding optimization results under the first priority strategy; Based on the third set of environmental parameter data subsets and the fourth set of environmental parameter data subsets, determine the target environmental parameters and corresponding optimization results under the second priority strategy; From the target environment parameters and corresponding optimization results under the first priority strategy and the target environment parameters and corresponding optimization results under the second priority strategy, select the target optimization result, and determine the environment parameters input to the target correction model under the target optimization result as the target environment parameters.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor, wherein, The memory is used to store computer programs that can run on the processor; The processor is configured to execute the path loss modeling method as described in any one of claims 1 to 6 when running the computer program.

9. A computer storage medium, characterized in that, The computer storage medium stores a computer program that, when executed by at least one processor, implements the path loss modeling method as described in any one of claims 1 to 6.

Images