Cognitive radio spectrum resource management method and system based on neural network

Abstract

The invention discloses a cognitive radio spectrum resource management method and system based on a neural network, and belongs to the technical field of wireless communication. The method comprises the following steps: S100, collecting radio frequency spectrum data; s200, preprocessing the collected radio frequency spectrum data; s300, based on the preprocessed radio frequency spectrum data, constructing a spectrum sensing model and performing training; s400, constructing a spectrum management model and performing training; and S500, realizing spectrum management through the spectrum sensing model and the spectrum management model. According to the method, the application of deep learning in cognitive radio spectrum management can be comprehensively optimized, and a valuable technical scheme is provided for theoretical research and practical application of cognitive radio.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and more particularly to a cognitive radio spectrum resource management method and system based on neural network. Background Art

[0002] The current spectrum utilization rate of spectrum resources is low. Improving spectrum utilization rate is an urgent problem to be solved and a current research hotspot. Cognitive radio technology is one of the effective ways to improve spectrum utilization rate.

[0003] As a powerful data-driven machine learning method, deep learning has achieved remarkable results in image recognition, natural language processing and other fields. However, how to effectively integrate deep learning into cognitive radio to achieve intelligent management and optimization of spectrum resources remains a challenge to be solved. This requires in-depth research on the theoretical properties of deep learning, such as its ability to represent data and generalization performance, and combining the characteristics of cognitive radio to design effective models and algorithms.

[0004] In addition, due to the dynamics and complexity of the spectrum environment, how to make deep learning models adapt to such environmental changes and how to deal with possible non-ideal situations (such as signal interference, equipment failure, etc.) are also important research topics. From a practical point of view, the effective management of spectrum resources is crucial to the performance of wireless communication systems. In a variety of applications such as mobile communications, unmanned driving, and the Internet of Things, the demand for spectrum resources is increasing. However, spectrum resources are limited, and how to improve their utilization and reduce resource waste is an important issue facing the communications industry. In addition, due to the complexity of the wireless communication environment, such as user mobility, device diversity, and signal interference, the problem of spectrum management has become extremely complex. Deep learning can propose effective spectrum allocation strategies, reduce signal interference, and improve system performance by learning and understanding this complex environment.

[0005] However, research in this field is still in its infancy and faces many challenges, such as the design and training of deep learning models, and the application of deep learning in complex wireless communication environments. At the same time, the practical application of deep learning in cognitive radio spectrum management also requires further research and verification to improve the effectiveness of spectrum management.

[0006] Therefore, how to provide a cognitive radio spectrum resource management method and system based on neural network is a problem that technical personnel in this field need to solve urgently. Summary of the invention

[0007] In view of this, the present invention provides a cognitive radio spectrum resource management method and system based on neural network to solve the technical problems existing in the above-mentioned prior art.

[0008] In order to achieve the above object, the present invention provides the following technical solutions:

[0009] A cognitive radio spectrum resource management method based on a neural network, comprising:

[0010] S100: Collect radio spectrum data;

[0011] S200: pre-processing the collected radio spectrum data;

[0012] S300: constructing a spectrum sensing model and performing training based on the preprocessed radio spectrum data;

[0013] S400: Build a spectrum management model and perform training;

[0014] S500: Spectrum management is implemented through a spectrum sensing model and a spectrum management model.

[0015] Preferably, the method further includes S600: verifying and evaluating the spectrum sensing model and the spectrum management model.

[0016] Preferably, the step S100: collecting radio spectrum data includes:

[0017] Collect radio spectrum data at different locations, times and weather conditions.

[0018] Preferably, S200: preprocessing the collected radio spectrum data includes:

[0019] S210: performing denoising and normalization processing on the collected radio spectrum data;

[0020] S220: Divide the processed radio spectrum data into a training set, a validation set, and a test set.

[0021] Preferably, S300: constructing a spectrum sensing model based on the preprocessed radio spectrum data and performing training, including:

[0022] S310: Selecting a type of spectrum sensing model, where the type of spectrum sensing model includes a convolutional neural network model and a recurrent neural network model, where the convolutional neural network model is used to process radio spectrum data having grid data characteristics in a training data set, and the recurrent neural network model is used to process radio spectrum data having time dependence in a training data set;

[0023] S320: Setting the structure of the spectrum sensing model;

[0024] S330: Selecting an optimizer and a loss function of a spectrum sensing model;

[0025] S340: Determine a training strategy for the spectrum sensing model.

[0026] Preferably, S400: constructing a spectrum management model and performing training, including:

[0027] S410: Using reinforcement learning RL model framework as the basis of spectrum management model;

[0028] S420: Determine the state space, action space and reward function of the spectrum management model;

[0029] S430: Determine an optimizer, a loss function, and training parameters of a spectrum management model.

[0030] Preferably, in the S420: determining the state space, action space and reward function of the spectrum management model:

[0031] The state space includes information on spectrum usage and user needs;

[0032] The action space includes decisions on allocating spectrum and adjusting power;

[0033] The reward function includes objectives that reflect spectrum management.

[0034] A cognitive radio spectrum resource management system based on a neural network, comprising:

[0035] Data collection module: collects radio spectrum data;

[0036] Data processing module: pre-processing the collected radio spectrum data;

[0037] Spectrum sensing model building module: builds and trains spectrum sensing models based on preprocessed radio spectrum data;

[0038] Spectrum management model building module: builds spectrum management models and performs training;

[0039] Spectrum management module: implements spectrum management through spectrum perception model and spectrum management model.

[0040] Preferably, it also includes:

[0041] Verification and evaluation module: Verify and evaluate the spectrum perception model and spectrum management model.

[0042] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method and system for cognitive radio spectrum resource management based on neural network, which has certain advantages and effectiveness in spectrum perception and management issues. The specific beneficial effects are:

[0043] (1) Data-driven: The present invention relies entirely on the processing and analysis of big data. By applying big data analysis to the decision-making process, the present invention can provide deeper and more accurate insights to help teams make better decisions. This data-driven approach is the unique feature of the present invention.

[0044] (2) Advanced machine learning techniques: The present invention uses some of the latest machine learning techniques, such as deep learning, reinforcement learning, and natural language processing, etc. This means that the present invention can handle complex problems and produce results that traditional methods cannot achieve.

[0045] (3) Expandable and modular design: The design of the present invention allows it to be easily expanded and adjusted to meet future needs. The present invention is modular in design, and each module can be independently upgraded and replaced. This flexible design allows the present invention to continue to evolve over time and maintain its leading position. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0047] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0048] Figure 2 A schematic diagram of the spectrum data collection and preprocessing process provided by an embodiment of the present invention;

[0049] Figure 3 A schematic diagram of the spectrum sensing model design and training process provided by an embodiment of the present invention;

[0050] Figure 4 A schematic diagram of the spectrum management model design and training process provided by an embodiment of the present invention;

[0051] Figure 5 A schematic diagram of a comparative experiment and performance evaluation process provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] See also Figure 1 The embodiment of the present invention discloses a cognitive radio spectrum resource management method based on a neural network, comprising:

[0054] S100: Collect radio spectrum data;

[0055] S200: pre-processing the collected radio spectrum data;

[0056] S300: constructing a spectrum sensing model and performing training based on the preprocessed radio spectrum data;

[0057] S400: Build a spectrum management model and perform training;

[0058] S500: Spectrum management is achieved through spectrum sensing model and spectrum management model;

[0059] S600: Verify and evaluate the spectrum perception model and spectrum management model.

[0060] Specifically, this embodiment includes deepening the theory of deep learning and cognitive radio, and providing theoretical support for its practical application. The application of deep learning in spectrum sensing will be explored. The present invention provides a model that can self-learn and identify different radio signals to improve the accuracy and efficiency of spectrum sensing. Spectrum management is optimized using deep reinforcement learning. The created spectrum management model can self-learn and adjust the spectrum allocation strategy in an uncertain and dynamically changing radio environment, improve spectrum utilization efficiency and ensure communication quality. In addition, this embodiment also includes the application of deep learning in processing signal interference in radio communications. The present invention provides a model that can effectively identify and eliminate signal interference to improve the quality of wireless communications. The performance of the provided deep learning model will be evaluated and analyzed in detail to ensure its practical application value in cognitive radio spectrum management. By achieving these goals, the present invention will comprehensively optimize the application of deep learning in cognitive radio spectrum management and provide valuable technical solutions for theoretical research and practical application of cognitive radio.

[0061] In a specific embodiment, the key scientific problems to be solved mainly include:

[0062] (1) The feasibility and effectiveness of deep learning in spectrum sensing, specifically:

[0063] Radio spectrum resources are limited, and demand continues to grow, making effective management and utilization of these resources an important issue. Spectrum sensing is a key technology for cognitive radio, which can monitor and identify spectrum usage in a radio environment, helping to better utilize unused spectrum resources. Deep learning, as an advanced machine learning method, has demonstrated extremely high performance in many fields. Therefore, leveraging the advantages of deep learning to improve the accuracy and efficiency of spectrum sensing is a key scientific issue.

[0064] (2) Use deep reinforcement learning for effective spectrum management, specifically:

[0065] Spectrum management is another key technology of cognitive radio, which involves how to allocate limited radio spectrum resources to various radio systems to achieve the best overall communication effect. This problem is highly complex due to the uncertainty and dynamics of the radio environment. Deep reinforcement learning is a method that can learn and make decisions in uncertain and dynamic environments, so it can be an effective tool to solve this problem. Designing and implementing deep reinforcement learning models suitable for spectrum management, and using such models to improve spectrum usage efficiency and ensure communication quality, is another key scientific issue.

[0066] In a specific embodiment, a large amount of spectrum data collection and preprocessing is required in the data collection and preprocessing stage to provide available training data for the subsequent deep learning model. First, spectrum data is collected in a variety of radio environments. This involves using a spectrum analyzer or radio receiving equipment to collect data in different locations, at different times, and under different weather conditions. Data collection can be performed in different environments and conditions such as city centers, suburbs, indoors, outdoors, during the day, at night, on sunny days, and rainy days to obtain spectrum data that is as comprehensive and diverse as possible. The collected spectrum data is multidimensional, including information such as frequency, power, and time. The power spectrum density of the entire frequency band is measured once at a certain time interval (such as 10 seconds) to obtain a two-dimensional spectrum data. Each data point represents the power spectrum density at a specific frequency and time point. After the data is collected, the data needs to be preprocessed. The goal of preprocessing is to convert the raw data into a format suitable for deep learning model input and minimize the impact of noise and outliers.

[0067] In a specific embodiment, the pre-processing step comprises:

[0068] 1) Denoising: Since radio signals are affected by various factors, the collected data may contain noise. In this process, filters, smoothers and other methods can be used to remove noise.

[0069] 2) Normalization: Since the power spectral density at different frequencies and time points may vary greatly, the data needs to be normalized so that all data are in the same range (such as 0-1). This can be achieved by subtracting the minimum value and then dividing by the difference between the maximum and minimum values.

[0070] 3) Divide the data into training set, validation set and test set: Divide the data into training set, validation set and test set. The training set is used to train the model, the validation set is used to adjust the parameters of the model, and the test set is used to evaluate the performance of the model.

[0071] Through the above technical solution, a clean and well-formed data set can be obtained, which provides a basis for subsequent deep learning model training. The specific flow chart of the above technical solution is as follows Figure 2 shown.

[0072] In a specific embodiment, in the spectrum sensing model design and training phase, the key steps include selecting a suitable deep learning model, performing model design, and performing model training. This process involves a large number of decisions, including selecting the model type, designing the model structure, selecting the optimizer and loss function, and determining the training strategy. When selecting a deep learning model, the decision is made based on the characteristics of the problem. If the spectrum data has the characteristics of grid data, such as a two-dimensional image or a multi-dimensional tensor, a convolutional neural network (CNN) can be selected. Because CNN has excellent local perception characteristics, it can effectively extract local features of grid data. If the spectrum data has the characteristics of a time series, a recurrent neural network (RNN) can be selected. RNN can process data with time dependence because it has the ability to remember past information.

[0073] Specifically, for the design of the spectrum sensing model, the design of the input layer is determined according to the characteristics of the spectrum data. For the convolutional neural network CNN, the input layer is a two-dimensional spectrum image or a three-dimensional spectrum data tensor. For the recurrent neural network RNN, the input layer is a time series of spectrum data. The number and size of hidden layers, as well as the choice of activation function, need to be determined according to the complexity of the problem and the performance requirements of the model. The design of the output layer depends on the type of task. For the spectrum classification task, the output layer is a softmax layer used to output the probability of each category. In the model training stage, the optimizer, loss function, and training parameters such as the number of training rounds and batch size are determined. The optimizer is an algorithm used to update the model weights to minimize the loss function. The optimizer used in this embodiment is stochastic gradient descent (SGD) and its variants, such as SGD with momentum and Adam. The loss function is a function used to evaluate the gap between the model prediction and the true value. For the classification problem, the loss function used in this embodiment is the cross entropy loss, and its calculation formula is shown in formula (1):

[0074] L=-(1 / N)Σ[i=1toN]y i *log(p(y i )) (1)

[0075] Where N is the total number of samples, y i is the true label of the i-th sample, p(y i ) is the probability of the ith sample predicted by the spectrum sensing model.

[0076] Specifically, the training parameters include the number of training rounds and batch size, which need to be determined based on the size of the data and the complexity of the model. The number of training rounds determines the number of times the model learns on all the training data. The batch size determines the number of samples used for each weight update.

[0077] Specifically, to prevent overfitting, some strategies are used, such as early stopping, weight decay, and dropout. Early stopping stops training when the performance of the validation set no longer improves. Weight decay is a regularization method that encourages the model to learn smaller weights by adding a term proportional to the square of the weight to the loss function. Dropout randomly turns off a portion of neurons during training to increase the robustness of the model.

[0078] The above technical solutions are the key to achieving effective spectrum sensing. Model selection and design are carried out according to specific problems and data characteristics, and appropriate training strategies are used. The specific flow chart of this stage is as follows: Figure 3 shown.

[0079] In a specific embodiment, the spectrum management model design and training phase includes the steps of selecting a suitable model, designing a model structure, selecting an optimizer and a loss function, and determining a training strategy.

[0080] Specifically, the selection of the spectrum management model is usually based on the characteristics of the problem. The spectrum management problem can be formalized as a sequential decision problem. Therefore, the embodiment of the present invention uses a reinforcement learning (RL) model. Reinforcement learning maximizes the cumulative reward from environmental feedback by learning a strategy.

[0081] Specifically, during the spectrum management model design phase, the state space, action space, and reward function must be determined. The state space includes information such as spectrum usage and user needs. The action space includes decisions such as allocating spectrum and adjusting power. The reward function needs to reflect the goals of spectrum management, such as maximizing system throughput or minimizing interference. During the model training phase, the optimizer, loss function, and training parameters are determined. The optimizer is an algorithm used to update model weights to maximize rewards. Common optimizers include stochastic gradient descent (SGD) and its variants, such as SGD with momentum and Adam. The loss function is used to evaluate the difference between the model prediction and the actual reward. For reinforcement learning problems, the loss function used in this embodiment is the mean square error loss of Q-learning, and its calculation formula is shown in formula (2):

[0082] L=(1 / N)Σ[i=1toN](Q(s i ,a i )-(r i +γmax a Q(s′ i ,a))) 2 (2)

[0083] Where N is the total number of samples, s i and a i are the state and action of the i-th sample, r i is the actual reward, γ is the discount factor, s' i is the next state, max a Q(s′ i ,a) is the maximum Q value in the next state.

[0084] Specifically, the training parameters of the spectrum management model include the number of training rounds and batch size, which need to be determined based on the size of the data and the complexity of the model. In order to prevent overfitting, early stopping, weight decay and dropout methods are also applied. The spectrum management model design and training phase is the key to achieving effective spectrum management. It is also necessary to select and design a reasonable model based on specific problems and data characteristics, and use appropriate training strategies. The specific flow chart of this phase is as follows: Figure 4 shown.

[0085] In a specific embodiment, the comparative experiment and performance evaluation phase is a key component of a machine learning project. The main goal of this phase is to verify whether the proposed model is effective and to conduct a specific quantitative evaluation of its performance. The comparative experiment usually includes the following points:

[0086] 1) Design comparative experiments: The purpose of comparative experiments is to verify whether the new model is better than the existing methods. This requires the selection of some benchmark models for comparison. These models should be classic machine learning models, such as decision trees, support vector machines, etc., or existing methods in related fields. In addition, different parameter settings will be tested for the proposed new model to study the impact of parameter changes on model performance.

[0087] 2) Select evaluation indicators: Evaluation indicators are tools used to quantify model performance. Common evaluation indicators include accuracy, precision, recall, F1 score, and AUC-ROC.

[0088] The following are the calculation formulas for these evaluation indicators:

[0089] Accuracy:

[0090] Accuracy=(TP+TN) / (TP+TN+FP+FN) (3)

[0091] Precision:

[0092] Precision=TP / (TP+FP) (4)

[0093] Recall:

[0094] Recall=TP / (TP+FN) (5)

[0095] F1 score:

[0096] F1_score=2*(Precision*Recall) / (Precision+Recall) (6)

[0097] Among them, TP, TN, FP and FN represent true positives, true negatives, false positives and false negatives, respectively.

[0098] Specifically, AUC-ROC is the area under the receiver operating characteristic curve (ROC), which is used to evaluate the performance of the model at different thresholds.

[0099] 3) Perform performance evaluation: During the performance evaluation phase, cross-validation or holdout method is used to divide the training set and test set to obtain the performance of the model on unknown data.

[0100] 4) Analyze and interpret the results: This is an important step in the comparative experiment and performance evaluation phase.

[0101] In this step, the performance of the model needs to be analyzed in depth to find out which aspects of the model perform well and which aspects need improvement so that continuous optimization can be carried out.

[0102] Specifically, the comparative experiment and performance evaluation phase is the core of the machine learning project. Through this phase, the effectiveness of the model can be proved, the performance of the model can be evaluated, and the decision-making process of the model can be understood. The specific flow chart of this phase is as follows Figure 5 shown.

[0103] Specifically, the present invention adopts a deep learning method to systematically solve the spectrum resource management task from data to model, from training to evaluation, and provides valuable ideas for related research and engineering applications, which is one of the outstanding advantages of the present invention.

[0104] In a specific embodiment, the experimental method used is as follows:

[0105] (1) Data preprocessing: The experiment starts with data cleaning, including steps such as removing duplicate values, filling missing values, normalizing data, encoding categorical variables, etc. These steps can be performed using Python libraries such as pandas and scikit-learn.

[0106] (2) Feature Engineering: The experiment includes the steps of feature selection and feature generation. Feature selection uses filtering methods, packaging methods, or embedding methods. Feature generation requires feature crossover, polynomial features, etc.

[0107] (3) Model training: The experiment includes training models using various algorithms, such as logistic regression, random forest, support vector machine, and neural network. These steps can be performed using libraries such as scikit-learn, TensorFlow, and PyTorch.

[0108] The experimental methods used in model validation and testing are as follows:

[0109] (1) Verification and adjustment: An important experimental method is to use cross-validation to adjust model parameters. Grid search or random search can be used to find the optimal parameter combination in a given parameter space.

[0110] (2) Performance evaluation: The experiment uses different evaluation indicators to measure the performance of the model, including accuracy, precision, recall, F1 score, AUC-ROC, etc. These evaluations can be performed using the metrics module of scikit-learn.

[0111] Specifically, during all these steps, the experimental design and implementation will record all the details and results to facilitate subsequent analysis and interpretation.

[0112] In a specific embodiment, the following aspects are specifically included:

[0113] (1) Data preprocessing techniques: Data preprocessing is one of the key steps in a machine learning project because the performance of the model depends largely on the quality of the input data. Data preprocessing techniques include data cleaning (such as removing duplicate values ​​and handling missing values), data transformation (such as normalizing or standardizing numerical data and encoding categorical data), feature selection, and feature engineering.

[0114] (2) Machine learning techniques: The present invention uses various machine learning algorithms, including supervised learning algorithms (such as logistic regression, decision tree, random forest, support vector machine, neural network, etc.) and unsupervised learning algorithms (such as clustering, principal component analysis, etc.).

[0115] (3) Model training and optimization techniques: This includes using appropriate optimization algorithms (such as stochastic gradient descent, Adam, etc.) for model training, and using regularization techniques (such as L1 and L2 regularization) to prevent overfitting. In addition, in order to find the optimal model parameters, hyperparameter adjustment techniques such as grid search or random search are required.

[0116] (4) Model validation and evaluation techniques: Use cross-validation, leave-one-out validation and other techniques to validate the model. These techniques can help understand the performance of the model on unseen data. The model needs to be evaluated using various evaluation metrics (such as accuracy, precision, recall, F1 score, AUC-ROC, etc.).

[0117] (5) Model interpretation technology: As machine learning models become more and more complex, model interpretation technology becomes more and more important. These technologies include local interpretability model-sensitivity analysis, hierarchical interpretation of machine learning models, and help understand the decision-making process of the model.

[0118] Through the above technical solutions, this embodiment provides a complete machine learning workflow from data preprocessing to model training, and then to model verification and interpretation.

[0119] On the other hand, this embodiment also discloses a cognitive radio spectrum resource management system based on a neural network, including:

[0120] Data collection module: collects radio spectrum data;

[0121] Data processing module: pre-processing the collected radio spectrum data;

[0122] Spectrum sensing model building module: builds and trains spectrum sensing models based on preprocessed radio spectrum data;

[0123] Spectrum management model building module: builds spectrum management models and performs training;

[0124] Spectrum management module: realizes spectrum management through spectrum perception model and spectrum management model;

[0125] Verification and evaluation module: Verify and evaluate the spectrum perception model and spectrum management model.

[0126] Specifically, the present invention can also provide a user-friendly interface, which allows non-technical users to easily use the tools and functions provided by the present invention without having to understand the complex technical details behind it. This user-friendly design is not common in many technical projects.

[0127] Specifically, the present invention also includes the protection of data privacy and algorithmic fairness. When processing sensitive data and making decisions, the present invention also specifically designs a series of measures to protect data privacy and ensure algorithmic fairness. These measures include the use of anonymization and de-identification technologies to protect data privacy, and the use of fairness measurement and adjustment technologies to ensure algorithmic fairness. These measures are the present invention's innovative response to current data ethics issues.

[0128] This embodiment first collects a large amount of spectrum data in different radio environments and performs preprocessing. The specific preprocessing techniques include denoising, normalization and partitioning of training sets, validation sets and test sets. For spectrum perception problems, a suitable deep learning model, such as a convolutional neural network or a recurrent neural network, is selected and trained using optimizer Adam or SGD. During the training process, strategies such as early stopping and weight decay are implemented to prevent overfitting. Model performance is evaluated by indicators such as accuracy and recall. For spectrum management problems, a deep reinforcement learning model is designed, which includes a neural network for estimating the state-action value function. The state of the environment consists of factors such as the current spectrum occupancy and the demand for radio equipment, and the action is the spectrum resource allocated to each device. The model is trained using reward information from historical experience, and the reward can be determined based on factors such as spectrum utilization and communication quality. The model training will use reinforcement learning algorithms such as Q-learning and SARSA, and the model performance will be evaluated by indicators such as spectrum utilization and communication quality. The advantages and effectiveness of the proposed deep learning method in spectrum perception and management problems will be verified through comparative experiments with existing technologies.

[0129] In a specific embodiment, the feasibility analysis of a neural network-based cognitive radio spectrum resource management method and system provided by the present invention is as follows:

[0130] (1) Technical feasibility: At present, the technology in the field of machine learning and data science is becoming more and more mature, providing a solid technical foundation for the present invention. Key steps such as data preprocessing, model training, model verification and model interpretation have widely verified methods and strategies for reference. In addition, many open source libraries such as Python's pandas, scikit-learn, TensorFlow and PyTorch provide rich tools and functions, making the technical implementation of the present invention more convenient.

[0131] (2) Economic feasibility: Economic benefits are an important factor in evaluating the feasibility of the present invention. The neural network-based cognitive radio spectrum resource management method and system provided by the present invention will bring a series of economic benefits, such as improving decision-making efficiency, reducing errors, improving work processes, etc., which will generate significant economic benefits. The expected economic benefits will exceed the cost of the present invention, including the cost of data collection and cleaning, the cost of model development and maintenance, and the cost of required hardware and software. Therefore, from an economic point of view, the present invention is completely feasible.

[0132] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0133] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A cognitive radio spectrum resource management method based on neural network, characterized in that: include: S100: Collect radio spectrum data; S200: pre-processing the collected radio spectrum data; S300: constructing a spectrum sensing model and performing training based on the preprocessed radio spectrum data; S400: Build a spectrum management model and perform training; S500: Spectrum management is implemented through a spectrum sensing model and a spectrum management model.

2. The method for cognitive radio spectrum resource management based on neural network according to claim 1, characterized in that: Also included is S600: verifying and evaluating a spectrum sensing model and a spectrum management model.

3. The method for cognitive radio spectrum resource management based on neural network according to claim 1, characterized in that: The step S100: collecting radio spectrum data includes: Collect radio spectrum data at different locations, times and weather conditions.

4. The method for cognitive radio spectrum resource management based on neural network according to claim 1, characterized in that: S200: pre-processing the collected radio spectrum data, including: S210: performing denoising and normalization processing on the collected radio spectrum data; S220: Divide the processed radio spectrum data into a training set, a validation set, and a test set.

5. The method for cognitive radio spectrum resource management based on neural network according to claim 1, characterized in that: S300: Based on the preprocessed radio spectrum data, a spectrum sensing model is constructed and trained, including: S310: Selecting a type of spectrum sensing model, where the type of spectrum sensing model includes a convolutional neural network model and a recurrent neural network model, where the convolutional neural network model is used to process radio spectrum data having grid data characteristics in a training data set, and the recurrent neural network model is used to process radio spectrum data having time dependence in a training data set; S320: Setting the structure of the spectrum sensing model; S330: Selecting an optimizer and a loss function of a spectrum sensing model; S340: Determine a training strategy for the spectrum sensing model.

6. The method for cognitive radio spectrum resource management based on neural network according to claim 1, characterized in that: S400: Build and train spectrum management models, including: S410: Using the reinforcement learning RL model framework as the basis for spectrum management models; S420: Determine the state space, action space and reward function of the spectrum management model; S430: Determine an optimizer, a loss function, and training parameters of a spectrum management model.

7. The method for cognitive radio spectrum resource management based on neural network according to claim 6, characterized in that: In the step S420: determining the state space, action space and reward function of the spectrum management model: The state space includes information on spectrum usage and user needs; The action space includes decisions on allocating spectrum and adjusting power; The reward function includes objectives that reflect spectrum management.

8. A neural network-based cognitive radio spectrum resource management system using the neural network-based cognitive radio spectrum resource management method according to any one of claims 1 to 7, characterized in that: include: Data collection module: collects radio spectrum data; Data processing module: pre-processing the collected radio spectrum data; Spectrum sensing model building module: builds and trains spectrum sensing models based on preprocessed radio spectrum data; Spectrum management model building module: builds spectrum management models and performs training; Spectrum management module: implements spectrum management through spectrum perception model and spectrum management model.

9. The neural network-based cognitive radio spectrum resource management system according to claim 8, further comprising: Verification and evaluation module: Verify and evaluate the spectrum perception model and spectrum management model.

Images