A fault prediction method, device and equipment of a wireless repeater and a storage medium

Abstract

The application discloses a fault prediction method and device of a wireless repeater, equipment and a storage medium, and relates to the technical field of communication. The method comprises the following steps: obtaining fusion parameters corresponding to each state data according to state data collected by at least two types of sensors in the wireless repeater; the fusion parameters comprise a high-order fractal dimension and a generalized divergence in a quantum information geometric space; fusing the state data collected by the at least two types of sensors according to the fusion parameters to obtain fusion data; obtaining a time hidden state and a spatial attention weight of the fusion data by using a fault prediction model, and obtaining fusion features according to the time hidden state and the spatial attention weight; and obtaining a fault prediction result of the wireless repeater according to the fault prediction model and the fusion features. The scheme of the application solves the problems of low data fusion accuracy, insufficient data feature mining and insufficient fault prediction capability in the fault prediction of the wireless repeater.

Description

Technical Field

[0001] This application belongs to the field of communication technology, and specifically relates to a fault prediction method, apparatus, device and storage medium for a wireless repeater. Background Technology

[0002] In the field of 5G repeater technology, existing technologies collect equipment operation and environmental data by deploying sensors. For example, temperature and humidity sensors are used to monitor environmental parameters, and power sensors are used to monitor energy consumption. The data collected by these sensors is used to make a preliminary judgment on the equipment's status. In data processing, a weighted average fusion algorithm is used, which suffers from low data fusion accuracy and cannot fully extract data features. In fault prediction, threshold-based fault prediction is used, but this has insufficient predictive capability and makes it difficult to accurately predict potential faults in advance, resulting in long network outage times. Summary of the Invention

[0003] This application provides a method, apparatus, device, and storage medium for fault prediction of wireless repeaters, which solves the problems of low data fusion accuracy, inability to fully mine data features, and insufficient fault prediction capability in the fault prediction of 5G repeaters in the prior art.

[0004] Firstly, a fault prediction method for a wireless repeater is provided, including:

[0005] Based on state data collected by at least two types of sensors within the wireless repeater, fusion parameters corresponding to each state data are obtained; wherein, the fusion parameters include the higher-order fractal dimension and the generalized divergence in the quantum information geometric space;

[0006] According to the fusion parameters, the state data collected by the at least two types of sensors are fused to obtain fused data;

[0007] Using a fault prediction model, the temporal hidden state and spatial attention weights of the fused data are obtained, and fused features are obtained based on the temporal hidden state and the spatial attention weights.

[0008] Based on the fault prediction model and the fusion features, the fault prediction result of the wireless repeater is obtained.

[0009] Optionally, the fault prediction method for the wireless repeater, wherein obtaining fusion parameters corresponding to each of the state data based on state data collected by at least two types of sensors within the wireless repeater includes:

[0010] Based on state data, scale parameters, and the number of cells collected by at least two types of sensors within the wireless repeater, the higher-order fractal dimension corresponding to each of the state data points is obtained; wherein, the number of cells is determined using... The state data is obtained by covering local areas of each state data using a step-by-step coverage method.

[0011] Based on the state data collected by the at least two types of sensors, the data distribution of each state data in the quantum state, the reference distribution, and the quantum state dimension, the generalized divergence in the quantum information geometric space corresponding to each state data is obtained.

[0012] Optionally, the fault prediction method for the wireless repeater, wherein obtaining the temporal hidden state of the fused data using a fault prediction model includes:

[0013] The time-series features of the fused data are obtained by utilizing the first activation function in the spatiotemporal attention recurrent convolutional network of the fault prediction model.

[0014] The mapping value of the fused data is obtained by using the second activation function in the spatiotemporal attention recurrent convolutional network.

[0015] Based on the time series features and the mapping value, the temporal hidden state of the fused data is obtained.

[0016] Optionally, the fault prediction method for the wireless repeater further includes:

[0017] The fault prediction model is trained using a loss function and training samples; wherein the loss function includes a cross-entropy loss function, an adversarial loss function, a knowledge distillation loss function, and a loss function determined based on the training samples; the cross-entropy loss function is used to measure the difference between the fault prediction results of the training samples and the true labels; the adversarial loss function is used to measure the loss of the adversarial generative network; and the knowledge distillation loss function is used to measure the difference between the fault prediction model and the corresponding teacher model.

[0018] Optionally, the fault prediction method for the wireless repeater further includes:

[0019] If the fault prediction result indicates that the wireless repeater has a temperature control fault, then based on the temperature-related state data and reinforcement learning algorithm, the first temperature self-healing strategy of the wireless repeater and the value of the value function corresponding to the first temperature self-healing strategy are obtained; wherein, the value function is obtained by performing fuzzy evaluation on the temperature-related state data and the first temperature self-healing strategy based on fuzzy rules.

[0020] The temperature self-healing strategy of the wireless repeater is obtained based on the first temperature self-healing strategy, the value of the value function corresponding to the first temperature self-healing strategy, and the second temperature self-healing strategy obtained by using optimal control theory.

[0021] Optionally, the fault prediction method for the wireless repeater further includes:

[0022] If the fault prediction result indicates that the wireless repeater has a radio frequency module fault, then the Bayesian deep learning algorithm is used to predict the fault probability of the radio frequency module to obtain the fault probability.

[0023] If the failure probability is greater than a preset threshold, a search is performed based on candidate switching timing and candidate parameter adjustment schemes to obtain an RF self-healing strategy.

[0024] Secondly, a fault prediction device for a wireless repeater is also provided, comprising:

[0025] The processing module is used to obtain fusion parameters corresponding to each of the state data collected by at least two types of sensors in the wireless repeater station; wherein, the fusion parameters include the higher-order fractal dimension and the generalized divergence in the quantum information geometric space;

[0026] The fusion module is used to fuse the state data collected by the at least two types of sensors according to the fusion parameters to obtain fused data;

[0027] The extraction module is used to obtain the temporal hidden state and spatial attention weight of the fused data using a fault prediction model, and to obtain fused features based on the temporal hidden state and the spatial attention weight.

[0028] The prediction module is used to obtain the fault prediction result of the wireless repeater based on the fault prediction model and the fusion features.

[0029] Thirdly, a fault prediction device for a wireless repeater is also provided, comprising: a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the processor executes the program or instructions to implement the fault prediction method for the wireless repeater as described in the first aspect.

[0030] Fourthly, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the fault prediction method for a wireless repeater as described in the first aspect.

[0031] Fifthly, a computer program product is also provided, including computer instructions that, when executed by a processor, implement the fault prediction method for a wireless repeater as described in the first aspect.

[0032] Compared with existing technologies, this application provides a method, apparatus, device, and storage medium for fault prediction of a wireless repeater. Based on state data collected by at least two types of sensors within the wireless repeater, fusion parameters are obtained for each type of state data. These fusion parameters include a higher-order fractal dimension and a generalized divergence in the quantum information geometric space. The state data collected by the at least two types of sensors are fused according to the fusion parameters to obtain fused data. A fault prediction model is used to obtain the temporal hidden state and spatial attention weights of the fused data, and fusion features are obtained based on the temporal hidden state and spatial attention weights. Finally, a fault prediction result for the wireless repeater is obtained based on the fault prediction model and the fusion features. Thus, by fusing the state data collected by the at least two types of sensors using the fusion parameters, the problems of low data fusion accuracy and inability to fully mine data features in wireless repeater fault prediction are solved, as is the insufficient fault prediction capability achieved by using a fault prediction model. Attached Figure Description

[0033] Figure 1 This is a flowchart illustrating the fault prediction method for a wireless repeater as described in the embodiments of this application;

[0034] Figure 2 This is a schematic diagram of the module of the fault prediction device for the wireless repeater described in the embodiments of this application;

[0035] Figure 3 This is a hardware block diagram of the fault prediction device for the wireless repeater described in the embodiments of this application. Detailed Implementation

[0036] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0037] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and are not used to describe a specified order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, not limited in number; for example, a first object can be one or more. Furthermore, in the specification and claims, "and" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0038] like Figure 1 As shown in the figure, this application provides a fault prediction method for a wireless repeater, including:

[0039] Step 101: Based on the state data collected by at least two types of sensors in the wireless repeater, obtain the fusion parameters corresponding to each of the state data; wherein, the fusion parameters include the higher-order fractal dimension and the generalized divergence in the quantum information geometric space;

[0040] It should be understood that prior to step 101, at least two types of sensors need to be deployed in the wireless repeater. These sensors each have their own function and work together to provide rich data, including at least two of the following: temperature sensor, humidity sensor, power sensor, radio frequency signal strength sensor, and electromagnetic interference monitoring module.

[0041] For example, the DS18B20 digital temperature sensor is used. With its high precision, this digital temperature sensor can accurately measure the real-time temperature inside the wireless repeater, providing key data for determining whether the wireless repeater may malfunction due to overheating or overcooling.

[0042] The HIH-4000 series capacitive humidity sensor is used to monitor the humidity of the environment in which the wireless repeater is located, because excessively high or low humidity can affect the performance and lifespan of the wireless repeater.

[0043] The AD8361 power sensor is used to measure the power consumption of the wireless repeater, which helps to understand the workload and energy efficiency of the wireless repeater.

[0044] The MAX2267 radio frequency signal strength sensor is used to acquire radio frequency signal strength data, which is crucial for evaluating signal transmission quality and stability.

[0045] An electromagnetic interference monitoring module based on the principle of spectrum analyzer is adopted. By performing spectrum analysis on the surrounding electromagnetic environment, potential electromagnetic interference sources can be detected in a timely manner, thus avoiding the impact of electromagnetic interference sources on the normal operation of the wireless repeater.

[0046] It should be noted that the placement of the above-mentioned types of sensors is based on the specific placement of the relevant components within the wireless repeater, and is not restricted here.

[0047] Understandably, before step 101 above, it is necessary to acquire status data collected by at least two types of sensors in the wireless repeater according to a preset sampling frequency, for example, the preset sampling frequency is 100ms.

[0048] For example, the temperature sensor quickly captures the temperature inside the wireless repeater station every 100ms; the humidity sensor simultaneously records the humidity value at this time.

[0049] The collected status data is rapidly transmitted to a microcontroller (such as an STM32 series microcontroller), which acts as a data aggregation "transportation hub." Simultaneously, the communication module establishes stable communication with the base station via the network interface, acquiring signal quality data such as Reference Signal Receiving Power (RSRP) and Signal-to-Interference Plus Noise Ratio (SINR), as well as network configuration parameters like bandwidth and modulation scheme. The microcontroller then aggregates status data collected from at least two types of sensors within the wireless repeater and data from the network side.

[0050] In one implementation, optionally, based on state data collected by at least two types of sensors within the wireless repeater, fusion parameters corresponding to each of the state data are obtained, including:

[0051] Based on state data, scale parameters, and the number of cells collected by at least two types of sensors within the wireless repeater, the higher-order fractal dimension corresponding to each of the state data points is obtained; wherein, the number of cells is determined using... The state data is obtained by covering local areas of each state data using a step-by-step coverage method.

[0052] Based on the state data collected by the at least two types of sensors, the data distribution of each state data in the quantum state, the reference distribution, and the quantum state dimension, the generalized divergence in the quantum information geometric space corresponding to each state data is obtained.

[0053] In this embodiment, an algorithm based on the geometric fusion of high-order fractal theory and quantum information can be used for each state data. To obtain the corresponding higher-order fractal dimension As shown in the following formula (1):

[0054] (1);

[0055] in, It is the scale parameter, which determines the state data. The scale used when the complexity is considered, when The closer it gets to 0, the more detailed the characterization of the data's complex characteristics becomes, just like using a magnifying glass to observe the data at increasingly higher magnifications;

[0056] Indicates in At scale, using Level coverage method for state data The The number of boxes required to cover a local area; here, "box" can be understood as a tool for quantifying data distribution. The layered coverage method is better able to capture the complex details of the data compared to existing coverage methods;

[0057] It represents the total number of local regions. By analyzing different local regions, we can gain a comprehensive understanding of the data distribution.

[0058] It is the higher-order fractal dimension of the state data. It quantifies the complex structure of the data. The larger the value, the more complex the structure of the data.

[0059] It should be noted that, after obtaining the corresponding higher-order fractal dimension... At that time, continuously decrease The value is calculated multiple times. We can obtain the accurate high-order fractal dimension by using limit operations.

[0060] Next, we obtain the generalized divergence in the corresponding quantum information geometric space. Using the extended form of quantum relative entropy, as shown in the following formula (2):

[0061] (2);

[0062] in, It is state data The first quantum state The data distribution reflects the characteristics of state data at the quantum level;

[0063] It is its reference distribution, which is usually a standard distribution obtained based on historical data or theoretical models, used for comparative analysis;

[0064] It is the quantum state dimension, which determines the complexity of quantum information; different dimensions contain different levels of information.

[0065] The trace of a matrix is ​​used to calculate the sum of the elements on the main diagonal of the matrix, and here it is used to quantify the degree of difference between two distributions.

[0066] Step 102: Based on the fusion parameters, fuse the state data collected by the at least two types of sensors to obtain fused data;

[0067] In this embodiment of the application, firstly, based on the fusion parameters corresponding to each of the stated state data, including the higher-order fractal dimension... and generalized divergence in quantum information geometric space Obtain the fusion weights corresponding to each state data. As shown in the following formula (3):

[0068] (3);

[0069] in, It is state data and The covariance is used to measure the correlation between two state data. The larger the covariance, the more similar the data change trends of the two state data.

[0070] It is a correlation adjustment coefficient used to adjust the influence of covariance in the weight calculation. The impact of data relevance on weights can be emphasized or weakened according to actual needs.

[0071] Then, based on the fusion weights corresponding to each of the aforementioned state data... Status data collected from the at least two types of sensors Perform fusion to obtain fused data This achieves the organic fusion of multi-source status data, making the fused data (i.e., fused data) better reflect the actual operating status of the wireless repeater, as shown in the following formula (4):

[0072] (4).

[0073] In one embodiment, optionally, after step 102 above, the method further includes:

[0074] An algorithm based on variational mode decomposition and compressed sensing sparse representation is used to remove noise from the fused data and each of the state data.

[0075] In this embodiment, variational mode decomposition is first performed on the fused data and each state data. Based on the local features and frequency characteristics of the signal, the complex fused data and each state data are decomposed into multiple intrinsic mode functions (IMF) components. Each IMF component contains the characteristics of the signal in a specific time scale and frequency range, just like decomposing a complex sound into notes of different frequencies.

[0076] Then, each IMF component is sparsely represented, and the data for each IMF is obtained by solving the following formula (5):

[0077] (5);

[0078] in, This refers to IMF data, specifically each IMF data obtained after variational mode decomposition.

[0079] The sparsity coefficient is represented in the dictionary. Below, the sparsity of IMF data is measured by the number of non-zero elements in the sparsity coefficient. The fewer non-zero elements in the sparsity coefficient, the better the sparsity of the data, which is like using as few words as possible to accurately describe something. A dictionary is a pre-constructed set of basis vectors used to linearly represent IMF data. The dictionary can be constructed based on prior knowledge of the signal or trained through machine learning algorithms. It is like a special "dictionary" used to convert IMF data into a sparse representation.

[0080] It is a regularization parameter used to control... The strength of the data smoothing constraint should be adjusted to prevent excessive smoothing from causing the loss of data features. Yes Perform the Laplace operator operation to introduce smoothing constraints on the data, making the signal after sparse representation smoother;

[0081] It is a weighting parameter that balances sparsity and smoothness, and is adjusted by... This can balance data smoothness while ensuring data sparsity.

[0082] By utilizing an improved orthogonal matching pursuit algorithm combined with an adaptive step-size search strategy, from the dictionary... The process involves iteratively selecting the basis vectors that best match the IMF data, minimizing the residuals in each selection. After multiple iterations, the sparse coefficients that most accurately represent the IMF data are found. This process removes noise components and reconstructs the denoised fused data and each state data, providing high-quality data for subsequent fault prediction and analysis.

[0083] Therefore, in steps 101 to 102 of the fault prediction method for a wireless repeater described in this application embodiment, at least two types of sensors are first deployed within the wireless repeater to collect data from the repeater itself and the network side at a preset sampling frequency. This data is then fused using high-order fractal theory and a quantum information geometric fusion algorithm, and further noise is significantly removed using variational mode decomposition and compressed sensing sparse representation algorithms. Thus, the use of high-precision fusion and denoising algorithms in data processing improves data quality and lays the foundation for subsequent fault prediction.

[0084] Step 103: Using a fault prediction model, obtain the temporal hidden state and spatial attention weights of the fused data, and obtain the fused features based on the temporal hidden state and the spatial attention weights;

[0085] In one implementation method, optionally, the temporal hidden state of the fused data is obtained using a fault prediction model, including:

[0086] The time-series features of the fused data are obtained by utilizing the first activation function in the spatiotemporal attention recurrent convolutional network (STAR-Net) in the fault prediction model.

[0087] The mapping value of the fused data is obtained by using the second activation function in the spatiotemporal attention recurrent convolutional network.

[0088] Based on the time series features and the mapping value, the temporal hidden state of the fused data is obtained.

[0089] The embodiments of this application can achieve improvements in the temporal dimension through a spatiotemporal attention recurrent convolutional network:

[0090] The spatiotemporal attention recurrent convolutional network employs innovative recurrent units in the time dimension, and the state update is shown in the following formula (6):

[0091] (6).

[0092] in, This represents the hidden state at the current moment. This hidden state is the key output of the fault prediction model after processing the information at the current time step, integrating information from the past and the current input.

[0093] As the first activation function, It is the second activation function. Is and Different activation functions, optional. Using the ReLU function effectively solves the gradient vanishing problem, making the fault prediction model more efficient at extracting important features when learning time series data; optionally, The Sigmoid function is used to map the input to a preset range, such as (0,1), and obtain the mapped value, which is used to filter and weight the information.

[0094] , , These correspond to the hidden states of the previous time step. The fused data at the current moment The memory vector of the previous time step The weight matrices at different time steps are continuously optimized through training the fault prediction model to adjust the degree of influence of different information on the hidden state at the current moment. , It is a bias vector that adds additional learnable parameters to the fault prediction model, helping the model to better fit complex time series features;

[0095] It is the memory vector of the previous time step, which saves the key information of the previous time step, so that the fault prediction model can fully take into account the long-term dependencies of history when calculating the hidden state at the current time step.

[0096] , , These are time step weighting coefficients, used to adjust the importance of information from different time steps, for example... The hidden state at different time steps can be adjusted based on its influence on the hidden state at the current time step. Weights in the calculation;

[0097] , , These refer to the historical time steps considered. By setting different time steps, the fault prediction model can capture time series features at different time scales.

[0098] Furthermore, the spatial dimension can be improved through the spatial attention module in the embodiments of this application:

[0099] Introducing a spatial attention module, the spatial attention weights are obtained using the following formula (7). :

[0100] (7);

[0101] in, and Indicates spatial location and The feature vectors at that location contain information such as the status of the wireless repeater and the environment at that spatial location;

[0102] It represents the total number of spatial locations, used to normalize attention weights;

[0103] ,in, This indicates a splicing operation, which combines spatial positions. and The feature vectors at each location are concatenated along their dimensions to fuse feature information from different locations;

[0104] and It is a weight matrix. These are bias vectors, used to perform a linear transformation on the concatenated feature vectors to obtain a score. This score reflects spatial location. and The degree of correlation between feature vectors. yes and In the Correlation measures on a feature dimension, such as calculating the Pearson correlation coefficient of two location feature vectors in the feature dimension of wireless repeater status, to measure their correlation in that dimension.

[0105] It is the relevance feature weight coefficient, used to adjust the importance of the relevance of different feature dimensions in the attention weight calculation;

[0106] It is the number of feature dimensions to consider. By comprehensively considering the correlation of multiple feature dimensions, the spatial attention weight is calculated more comprehensively.

[0107] Next, this embodiment of the application implements dynamic adjustment of the weights of the convolution operation through the feature extraction module:

[0108] The weights of the convolution operation are obtained as shown in formula (8) below:

[0109] (8);

[0110] in, Based on time-hidden state Spatial attention weights The attention weights combine the effects of temporal and spatial information on the current convolutional kernel weights. For example, when the information at a certain moment in the time series is closely related to the information at the current spatial location, the attention weights will increase.

[0111] The mean is Standard deviation is The Gaussian function is used to adjust the spatial location of the convolution kernel weights. By adjusting the mean and standard deviation, the convolution kernel can focus on different spatial locations at different time steps. For example, when the device state changes drastically, the convolution kernel can focus more on the changing region.

[0112] It is the time step adjustment coefficient, used to adjust the degree of influence of different time steps on the convolution kernel weights;

[0113] It is the number of time steps, used to accumulate adjustments at different time steps.

[0114] Step 104: Obtain the fault prediction result of the wireless repeater based on the fault prediction model and the fusion features.

[0115] In one embodiment, optionally, before step 103, the method further includes:

[0116] The fault prediction model is trained using a loss function and training samples; wherein the loss function includes a cross-entropy loss function, an adversarial loss function, a knowledge distillation loss function, and a loss function determined based on the training samples; the cross-entropy loss function is used to measure the difference between the fault prediction results of the training samples and the true labels; the adversarial loss function is used to measure the loss of the adversarial generative network; and the knowledge distillation loss function is used to measure the difference between the fault prediction model and the corresponding teacher model.

[0117] In this embodiment, the preprocessed fused data is divided into a training set and a test set according to a time series and a preset ratio (e.g., 7:3). The training set is used to train the fault prediction model, allowing it to learn the features and patterns in the data, such as the relationship between the time series features and spatial location features of the wireless repeater under different operating states. The test set is used to evaluate the performance of the fault prediction model and test its generalization ability on unseen data, such as the accuracy of the fault prediction model in predicting new fault modes or the state of the wireless repeater under different environments.

[0118] The loss function based on adversarial learning, knowledge distillation, and adversarial example generation is adopted as shown in the following formula (9):

[0119] (9);

[0120] in, It is the cross-entropy loss function, used to measure the difference between the fault prediction results predicted by the fault prediction model and the true labels. Specifically, , This refers to the number of categories, such as the number of fault types. It is the first in the real tag The probability of a class (usually 0 or 1); The fault prediction model predicts the first The probability of a class;

[0121] It is an adversarial loss function, which comes from the adversarial process between the generator and the discriminator in the adversarial generative network in the fault prediction model. The generator tries to generate fault prediction results that make it difficult for the discriminator to distinguish between true and false, while the discriminator tries to distinguish the fault prediction results of the generator from the real labels. Through this adversarial process, the generator is prompted to generate predictions that are closer to the real distribution, thereby improving the generalization ability of the fault prediction model.

[0122] It is knowledge distillation loss, which passes the knowledge of the teacher model (e.g., the output probability distribution of the teacher model) to the fault prediction model (as the student model), allowing the fault prediction model to learn richer feature representations and improving model performance. Specifically, ,in, The teacher model predicts the first The probability of a class The fault prediction model predicts the first The probability of a class;

[0123] It is a loss based on adversarial example generation, used to enhance the robustness of the fault prediction model to adversarial examples. By generating adversarial examples, the fault prediction model learns the difference features between adversarial examples and normal examples in the training samples, thereby improving the model's generalization ability.

[0124] , , It is a balancing coefficient used to adjust the relative importance of the cross-entropy loss function, adversarial loss function, knowledge distillation loss function, and loss function determined based on adversarial examples in the training samples in the total loss function. Their values ​​are adjusted experimentally to achieve the best training effect.

[0125] It should be noted that an optimizer is used in the fault prediction model training process of this application embodiment. Here, the selection of the optimizer is explained:

[0126] An adaptive high-order moment estimator optimizer (AHO-MO) is used, where the first-order moment estimate is shown in the following formula (10):

[0127] (10);

[0128] The second moment estimate is shown in the following formula (11):

[0129] (11);

[0130] in, It is the first moment estimate of the gradient at time step 1. It approximates the mean of the gradient and is used to adjust the direction of parameter updates, for example, when 1 / 2 t = 1 / 2 t. When it is positive, it means that the mean of the gradient in that direction is positive, and the parameter update direction may be adjusted in that direction;

[0131] yes The second moment estimate of the gradient at each time step approximates the variance of the gradient and is used to adjust the step size of parameter updates. The larger the variance, the greater the gradient fluctuation, and it may be necessary to adjust the step size to avoid excessively drastic parameter updates.

[0132] yes The gradient at time step is obtained by differentiating the loss function with respect to the model parameters. and These are parameters that change dynamically based on the data distribution. For example, they can be adjusted using an adaptive function based on the variance, mean, and other statistical measures of the current data to adapt to different data distributions.

[0133] It is a loss function Regarding parameters The Hessian matrix contains information about the second derivative of the loss function with respect to the parameters, which is used to adjust the gradient direction more precisely.

[0134] It is a loss function Regarding parameters The Jacobian matrix, which contains the matrix form of the first derivative of the loss function with respect to the parameters, is used to adjust the magnitude of the gradient. , These are higher-order moment adjustment coefficients, used to adjust the degree of influence of the Hessian matrix and Jacobian matrix on the estimation of first-order and second-order moments;

[0135] , It considers the number of historical gradient steps. By taking historical gradient information into account, the optimizer can better adapt to the changing trend of gradients.

[0136] Therefore, in steps 103 and 104 of the fault prediction method for wireless repeaters described in this application embodiment, a spatiotemporal attention recurrent convolutional network is constructed. Feature extraction is improved and dynamically adjusted from the temporal and spatial dimensions, and the data is divided into training and testing sets according to a preset ratio. The model is trained using a loss function based on adversarial learning, knowledge distillation, and adversarial sample generation, along with an adaptive high-order moment estimation optimizer. In terms of fault prediction, the innovative network architecture and training method make the model prediction more accurate and enable early detection of potential faults.

[0137] In one embodiment, optionally, after step 104, the method further includes:

[0138] If the fault prediction result indicates that the wireless repeater has a temperature control fault, then based on the temperature-related state data and reinforcement learning algorithm, the first temperature self-healing strategy of the wireless repeater and the value of the value function corresponding to the first temperature self-healing strategy are obtained; wherein, the value function is obtained by performing fuzzy evaluation on the temperature-related state data and the first temperature self-healing strategy based on fuzzy rules.

[0139] The temperature self-healing strategy of the wireless repeater is obtained based on the first temperature self-healing strategy, the value of the value function corresponding to the first temperature self-healing strategy, and the second temperature self-healing strategy obtained by using optimal control theory.

[0140] In this embodiment of the application, if the fault prediction result indicates that the wireless repeater has a temperature control fault, specifically, if the fault prediction result indicates that the wireless repeater has an overheating fault, then a strategy based on the fusion of reinforcement learning, optimal control theory and fuzzy logic is adopted.

[0141] First, define the state space of the wireless repeater. ,in, This indicates the temperature, which is a key indicator for judging whether a wireless repeater is overheating. Excessive temperature may lead to a decrease in the performance of the wireless repeater or even damage. This indicates power. Abnormal power changes may also be related to equipment overheating. For example, excessive power may be caused by some components inside the wireless repeater malfunctioning and generating too much heat.

[0142] Action space ,in, This indicates that the cooling fan speed is adjusted. By changing the fan speed, the cooling efficiency of the wireless repeater can be adjusted. This indicates whether to activate an additional cooling device. When conventional cooling methods are insufficient, activating an additional cooling device can further reduce the temperature of the wireless repeater.

[0143] Next, the value function is updated using the following formula (12):

[0144] (12);

[0145] in, yes The real-time status of the wireless repeater, which includes various status information such as temperature and power, is the basis for decision-making.

[0146] It is a reward value, used to measure the immediate benefit obtained after taking a certain action at a certain time. For example, when the device temperature drops after adjusting the speed of the cooling fan, the reward value can be set to a positive number. Conversely, if the temperature continues to rise, the reward value may be a negative number.

[0147] This is a discount factor, typically ranging from (0,1), which represents the degree of importance placed on future rewards. The closer it is to 1, the more emphasis is placed on future rewards, that is, on long-term optimization of the wireless repeater status;

[0148] It is a state-action value function, which evaluates the state-action value in the context of action. Always in a state of readiness Take action at the time The long-term value lies in the fact that by continuously learning and updating this function, the agent (which can be understood as the system's decision-making module) can find the optimal sequence of actions.

[0149] Based on the Fuzzy rules are used for fuzzy evaluation of states and actions. These fuzzy rules are formulated based on expert experience and equipment operating characteristics. For example, if the temperature is high and the power is large, then an additional heat dissipation device will be turned on. Fuzzy logic can handle some uncertainties and information that is difficult to quantify precisely.

[0150] These are fuzzy rule weights, used to adjust the first... The importance of fuzzy rules in comprehensive evaluation: different fuzzy rules may be applicable to different scenarios, and adjusting the weights can make the system more flexible in responding to various situations;

[0151] It is the total number of fuzzy rules, which covers all possible combinations of device states and actions.

[0152] Next, optimal control theory is used, which aims to find a control strategy that achieves optimal performance under certain constraints. For a wireless repeater experiencing overheating, the theoretically optimal control strategy is found by solving the Hamilton-Jacobi-Bellman (HJB) equations. The HJB equations are partial differential equations that describe the relationship between the value function and the state and control variables under optimal control. In practical applications, numerical methods, such as the finite difference method and dynamic programming, are typically used to solve the HJB equations. The strategy obtained from reinforcement learning and the fuzzy logic evaluation results are combined with the strategy obtained from optimal control theory, taking into account the actual operation of the wireless repeater, expert experience, and the theoretical optimal solution, to obtain the final heat dissipation control strategy. For example, if both the reinforcement learning strategy and the optimal control strategy suggest increasing the cooling fan speed, and the fuzzy logic evaluation also supports this decision, then this action will be executed; if a conflict occurs, a decision will be made based on pre-set weights or priorities.

[0153] In one embodiment, optionally, after step 104, the method further includes:

[0154] If the fault prediction result indicates that the wireless repeater has a radio frequency module fault, then the Bayesian deep learning algorithm is used to predict the fault probability of the radio frequency module to obtain the fault probability.

[0155] If the failure probability is greater than a preset threshold, a search is performed based on candidate switching timing and candidate parameter adjustment schemes to obtain an RF self-healing strategy.

[0156] In this embodiment of the application, if the fault prediction result indicates that the wireless repeater has a radio frequency module fault, a backup module switching strategy based on Bayesian deep learning, swarm intelligence and quantum annealing optimization is adopted.

[0157] First, Bayesian deep learning is used to predict the failure probability of the radio frequency module in the wireless repeater station, and the failure probability distribution is obtained. ,in, This indicates a fault, which can be of different types, such as abnormal radio frequency signal strength or excessive signal interference. This refers to monitoring data, including status data such as radio frequency signal strength and interference levels collected by sensors, as well as signal quality data obtained from base stations. Bayesian deep learning, by incorporating prior knowledge and uncertainty estimation, can more accurately predict fault probabilities and can provide the range of uncertainty in the predictions, which is crucial for decision-making.

[0158] Next, a swarm intelligence algorithm is introduced, treating each candidate switching opportunity and candidate parameter adjustment scheme as an agent. Agents cooperate and compete through pheromones and local search mechanisms. Pheromones are virtual substances used to record the agents' experience and information during the search process. For example, when an agent discovers a switching scheme that effectively resolves an RF module fault, it leaves pheromones along its path. Other agents choose their search direction based on pheromone concentration, with higher concentration paths being more likely to be selected. The local search mechanism involves each agent searching near its current location, attempting to find a better switching scheme. For example, an agent can search before and after the current switching opportunity, or fine-tune the current parameter adjustment scheme to find a superior solution.

[0159] Next, we introduce the quantum annealing optimization algorithm to optimize the search process of the swarm intelligence algorithm. The quantum annealing algorithm utilizes the superposition state and quantum tunneling effect of qubits to escape local optima and is more likely to find the global optimum. In this application, qubits can represent the agent's state, such as whether to choose a candidate switching timing or a candidate parameter adjustment scheme. The quantum tunneling effect gives the agent a certain probability of jumping from a local optimum to a better solution, even if this solution appears unreachable in the traditional search space. In this way, the probability of finding the optimal switching timing and parameter adjustment scheme is increased. For example, when the swarm intelligence algorithm gets stuck in a local optimum and cannot find a better switching scheme, the quantum annealing algorithm can help the agent escape this local optimum and continue searching for a better solution.

[0160] It should be noted that after the fault prediction model outputs a fault prediction result (including the fault type), it automatically selects and executes the corresponding self-healing strategy from the self-healing strategy library. For example, if the fault prediction model outputs a fault prediction result indicating that the wireless repeater has an overheating fault, then a corresponding heat dissipation control action is selected based on a strategy based on reinforcement learning, optimal control theory, and fuzzy logic fusion, such as adjusting the cooling fan speed or activating additional cooling devices. If the fault prediction model outputs a fault prediction result indicating that the wireless repeater has an RF module fault, then a backup module switching timing and parameter adjustment scheme are selected based on a strategy based on Bayesian deep learning, swarm intelligence, and quantum annealing optimization.

[0161] It should also be noted that during the self-healing process, the status changes of the wireless repeater are continuously monitored. Sensors collect real-time status data such as temperature, power, and RF signal strength of the wireless repeater, and compare these data with preset normal ranges. If the self-healing effect is found to be poor, for example, if the wireless repeater temperature remains too high after implementing heat dissipation measures, or if the RF signal quality does not significantly improve after switching to a backup module, the system will take action.

[0162] Based on a strategy fused from reinforcement learning, optimal control theory, and fuzzy logic, reinforcement learning is re-performed, updating the value function and policy according to the new state data and reward values ​​of the wireless repeaters; optimal control is re-calculated, solving the HJB equations based on the latest state data and constraints of the wireless repeaters to obtain a new optimal control policy; and the fuzzy rules are adjusted, with their weights adjusted according to the actual situation. Adjustments can be made to the conditions and conclusions of fuzzy rules to improve the effectiveness of the strategy.

[0163] Based on a strategy of Bayesian deep learning, swarm intelligence, and quantum annealing optimization, the system recalculates the failure probability, updates the Bayesian deep learning model using new wireless repeater state data, and obtains a more accurate failure probability distribution. It then iterates using swarm intelligence algorithms, with the agent continuously searching for better switching timing and parameter adjustment schemes based on new pheromone levels and local search results. Finally, it performs quantum annealing optimization, utilizing the characteristics of qubits to attempt to escape the current local optimum and find better switching timing and parameter adjustment schemes. By continuously executing and optimizing the self-healing strategy, the system achieves intelligent fault prediction and self-healing of the wireless repeater, improving the reliability and stability of the wireless network.

[0164] Finally, for overheating and RF module failures, a self-healing strategy library was established using a fusion strategy based on reinforcement learning, optimal control theory, and fuzzy logic, and a optimization strategy based on Bayesian deep learning, swarm intelligence, and quantum annealing, respectively. After the fault prediction model outputs the fault prediction result, the self-healing strategy is automatically executed, with continuous monitoring during execution. If the effect is unsatisfactory, the self-healing strategy is relearned, calculated, and adjusted. This self-healing strategy integrates multiple advanced theories and algorithms to achieve intelligent decision-making and dynamic optimization for different faults, quickly responding to and repairing faults. This greatly improves the reliability and stability of the wireless repeater, ensures stable operation of the wireless network, reduces maintenance costs and network downtime, and enhances user experience.

[0165] In summary, this application provides a fault prediction method for wireless repeaters. First, at least two types of sensors are deployed within the wireless repeater to collect data from the repeater itself and the network at a preset sampling frequency. This data is then fused using high-order fractal theory and a quantum information geometric fusion algorithm, followed by noise removal using variational mode decomposition and compressed sensing sparse representation algorithms. Next, a spatiotemporal attention recurrent convolutional network is constructed, and feature extraction is performed by improving and dynamically adjusting the convolutional kernel weights from temporal and spatial dimensions. The data is divided into training and testing sets according to a preset ratio. A fault prediction model is trained using a loss function based on adversarial learning, knowledge distillation, and adversarial sample generation, along with an adaptive high-order moment estimation optimizer. Finally, for overheating and RF module faults, self-healing strategy libraries are established using a fusion strategy based on reinforcement learning, optimal control theory, and fuzzy logic, and a optimization strategy based on Bayesian deep learning, swarm intelligence, and quantum annealing, respectively. After obtaining the fault prediction result, the fault prediction model automatically executes the strategy, continuously monitoring the process. If the effect is unsatisfactory, the self-healing strategy is relearned, calculated, and adjusted.

[0166] Therefore, the data fusion technology based on higher-order fractal theory and quantum information geometric fusion algorithm in this application can fully consider the complex structure and distribution differences of data by calculating the higher-order fractal dimension and the generalized divergence in the quantum information geometric space. This can more accurately fuse multi-source state data, improve the accuracy of data fusion, and provide a more reliable data foundation for subsequent analysis.

[0167] The novel spatiotemporal attention recurrent convolutional network integrates time series processing, spatial information utilization, and attention mechanisms. In the time dimension, it comprehensively captures long-term dependencies of time series through improved recurrent units, while in the spatial dimension, it introduces an attention module to weightedly fuse spatial location information. During feature extraction, the convolutional weights are dynamically adjusted to achieve accurate fault prediction, effectively reduce network outage time, and improve the reliability of 5G networks.

[0168] Based on strategies such as reinforcement learning, optimal control theory and fuzzy logic fusion, the value function is updated in real time according to the state data of the wireless repeater. By combining fuzzy rules and optimal control theory, dynamic adaptive self-healing is achieved, resulting in faster response speed. The self-healing strategy can be optimized according to the actual situation, improving the self-healing effect, reducing maintenance costs and shortening network recovery time.

[0169] like Figure 2 As shown in the figure, this application embodiment also provides a fault prediction device for a wireless repeater, including:

[0170] The processing module 201 is used to obtain fusion parameters corresponding to each of the state data collected by at least two types of sensors in the wireless repeater; wherein the fusion parameters include at least one of the following: high-order fractal dimension, generalized divergence in quantum information geometric space, and fusion weight.

[0171] The fusion module 202 is used to fuse the state data collected by the at least two types of sensors according to the fusion parameters to obtain fused data;

[0172] Extraction module 203 is used to obtain the temporal hidden state and spatial attention weight of the fused data using a fault prediction model, and to obtain fused features based on the temporal hidden state and the spatial attention weight;

[0173] The prediction module 204 is used to obtain the fault prediction result of the wireless repeater based on the fault prediction model and the fusion features.

[0174] Optionally, in the aforementioned fault prediction device for the wireless repeater, the processing module 201 is specifically used for at least one of the following:

[0175] Based on state data, scale parameters, and the number of cells collected by at least two types of sensors within the wireless repeater, the higher-order fractal dimension corresponding to each of the state data points is obtained; wherein, the number of cells is determined using... The state data is obtained by covering local areas of each state data using a step-by-step coverage method.

[0176] Based on the state data collected by the at least two types of sensors, the data distribution of each state data in the quantum state, the reference distribution, and the quantum state dimension, the generalized divergence in the quantum information geometric space corresponding to each state data is obtained.

[0177] Optionally, in the aforementioned fault prediction device for the wireless repeater, the extraction module 203 is specifically used for:

[0178] The time-series features of the fused data are obtained by utilizing the first activation function in the spatiotemporal attention recurrent convolutional network of the fault prediction model.

[0179] The mapping value of the fused data is obtained by using the second activation function in the spatiotemporal attention recurrent convolutional network.

[0180] Based on the time series features and the mapping value, the temporal hidden state of the fused data is obtained.

[0181] Optionally, the fault prediction device for the wireless repeater further includes:

[0182] The training module is used to train the fault prediction model using a loss function and training samples; wherein, the loss function includes a cross-entropy loss function, an adversarial loss function, a knowledge distillation loss function, and a loss function determined based on the training samples; the cross-entropy loss function is used to measure the difference between the fault prediction results of the training samples and the true labels; the adversarial loss function is used to measure the loss of the adversarial generative network; and the knowledge distillation loss function is used to measure the difference between the fault prediction model and the corresponding teacher model.

[0183] Optionally, the fault prediction device for the wireless repeater further includes:

[0184] The first acquisition module is configured to, if the fault prediction result indicates that the wireless repeater has a temperature control fault, obtain a first temperature self-healing strategy of the wireless repeater and the value of the value function corresponding to the first temperature self-healing strategy based on temperature-related state data and reinforcement learning algorithm; wherein, the value function is obtained by performing fuzzy evaluation on the temperature-related state data and the first temperature self-healing strategy based on fuzzy rules.

[0185] The second obtaining module is used to obtain the temperature self-healing strategy of the wireless repeater based on the first temperature self-healing strategy, the value of the value function corresponding to the first temperature self-healing strategy, and the second temperature self-healing strategy obtained by using optimal control theory.

[0186] Optionally, the fault prediction device for the wireless repeater further includes:

[0187] The third acquisition module is used to predict the failure probability of the radio frequency module using a Bayesian deep learning algorithm if the fault prediction result indicates that there is a radio frequency module fault in the wireless repeater.

[0188] The fourth acquisition module is used to search for a radio frequency self-healing strategy based on candidate switching timing and candidate parameter adjustment scheme if the fault probability is greater than a preset threshold.

[0189] It should be noted that the apparatus provided in this application embodiment can implement all the method steps implemented in the above-mentioned wireless repeater fault prediction method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0190] This application also provides a fault prediction device for a wireless repeater, such as... Figure 3 As shown, it includes:

[0191] The processor 301, memory 302, transceiver 303, and programs or instructions stored in the memory 302 and executable on the processor 301; when the processor 301 executes the programs or instructions, it implements the various processes of the above-described wireless repeater fault prediction method embodiment and achieves the same technical effect. To avoid repetition, these will not be described again here.

[0192] The transceiver 303 is used to receive and send data under the control of the processor 301.

[0193] Among them, Figure 3 In this context, the bus architecture can include any number of interconnected buses and bridges, specifically connecting various circuits of one or more processors represented by processor 301 and memory represented by memory 302. The bus architecture can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. Transceiver 303 can be multiple elements, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium. For different user equipment, the user interface 304 can also be an interface capable of connecting external or internal devices, including but not limited to keypads, displays, speakers, microphones, joysticks, etc.

[0194] The processor 301 is responsible for managing the bus architecture and general processing, while the memory 302 can store the data used by the processor 301 when performing operations.

[0195] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the various processes of the above-described wireless repeater fault prediction method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0196] This application also provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, they implement the various processes of the above-described wireless repeater fault prediction method embodiment and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0197] It should be noted that, in this document, 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. Unless otherwise specified, 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.

[0198] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0199] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A fault prediction method for a wireless repeater, characterized in that, include: Based on state data collected by at least two types of sensors within the wireless repeater, fusion parameters corresponding to each state data are obtained; wherein, the fusion parameters include the higher-order fractal dimension and the generalized divergence in the quantum information geometric space; According to the fusion parameters, the state data collected by the at least two types of sensors are fused to obtain fused data; Using a fault prediction model, the temporal hidden state and spatial attention weights of the fused data are obtained, and fused features are obtained based on the temporal hidden state and the spatial attention weights. Based on the fault prediction model and the fusion features, the fault prediction result of the wireless repeater is obtained.

2. The method according to claim 1, characterized in that, Based on status data collected by at least two types of sensors within the wireless repeater, fusion parameters corresponding to each of the status data are obtained, including: Based on state data, scale parameters, and the number of cells collected by at least two types of sensors within the wireless repeater, the higher-order fractal dimension corresponding to each of the state data points is obtained; wherein, the number of cells is determined using... The state data is obtained by covering local areas of each state data using a step-by-step coverage method. Based on the state data collected by the at least two types of sensors, the data distribution of each state data in the quantum state, the reference distribution, and the quantum state dimension, the generalized divergence in the quantum information geometric space corresponding to each state data is obtained.

3. The method according to claim 1, characterized in that, Using a fault prediction model, the temporal hidden state of the fused data is obtained, including: The time-series features of the fused data are obtained by utilizing the first activation function in the spatiotemporal attention recurrent convolutional network of the fault prediction model. The mapping value of the fused data is obtained by using the second activation function in the spatiotemporal attention recurrent convolutional network. Based on the time series features and the mapping value, the temporal hidden state of the fused data is obtained.

4. The method according to claim 1, characterized in that, The method further includes: The fault prediction model is trained using a loss function and training samples; wherein the loss function includes a cross-entropy loss function, an adversarial loss function, a knowledge distillation loss function, and a loss function determined based on the training samples; the cross-entropy loss function is used to measure the difference between the fault prediction results of the training samples and the true labels; the adversarial loss function is used to measure the loss of the adversarial generative network; and the knowledge distillation loss function is used to measure the difference between the fault prediction model and the corresponding teacher model.

5. The method according to claim 1, characterized in that, The method further includes: If the fault prediction result indicates that the wireless repeater has a temperature control fault, then based on the temperature-related state data and reinforcement learning algorithm, the first temperature self-healing strategy of the wireless repeater and the value of the value function corresponding to the first temperature self-healing strategy are obtained; wherein, the value function is obtained by performing fuzzy evaluation on the temperature-related state data and the first temperature self-healing strategy based on fuzzy rules. The temperature self-healing strategy of the wireless repeater is obtained based on the first temperature self-healing strategy, the value of the value function corresponding to the first temperature self-healing strategy, and the second temperature self-healing strategy obtained by using optimal control theory.

6. The method according to claim 1, characterized in that, The method further includes: If the fault prediction result indicates that the wireless repeater has a radio frequency module fault, then the Bayesian deep learning algorithm is used to predict the fault probability of the radio frequency module to obtain the fault probability. If the failure probability is greater than a preset threshold, a search is performed based on candidate switching timing and candidate parameter adjustment schemes to obtain an RF self-healing strategy.

7. A fault prediction device for a wireless repeater, characterized in that, include: The processing module is used to obtain fusion parameters corresponding to each of the state data collected by at least two types of sensors in the wireless repeater station; wherein, the fusion parameters include the higher-order fractal dimension and the generalized divergence in the quantum information geometric space; The fusion module is used to fuse the state data collected by the at least two types of sensors according to the fusion parameters to obtain fused data; The extraction module is used to obtain the temporal hidden state and spatial attention weight of the fused data using a fault prediction model, and to obtain fused features based on the temporal hidden state and the spatial attention weight. The prediction module is used to obtain the fault prediction result of the wireless repeater based on the fault prediction model and the fusion features.

8. A fault prediction device for a wireless repeater, characterized in that, include: A processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the processor, when executing the program or instructions, implements the fault prediction method for a wireless repeater as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the fault prediction method for a wireless repeater as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the fault prediction method for a wireless repeater as described in any one of claims 1 to 6.

Images