Dynamic monitoring and early warning method and system for mobile phone signals

Abstract

The present invention provides a method and system for dynamic monitoring and early warning of mobile phone signals, and relates to the technical field of dynamic monitoring and early warning of mobile phone signals, including: collecting mobile phone signals, segmenting and extracting features, and constructing a signal evolution feature space; analyzing signal stability based on chaos theory, constructing a composite feature space in combination with the feature space, partitioning and mapping the composite space, identifying steady-state and transition areas, establishing normal and abnormal mode libraries respectively, and training a hierarchical prediction model, constructing an early warning decision tree based on the prediction results, including attenuation, interference and interruption early warning branches, triggering compensation, avoidance and emergency response mechanisms respectively, the emergency response mechanism calculates the optimal transmission path, evaluates base station resources, generates a base station switching sequence, and sends a control instruction, and finally dynamically generates an optimal disposal decision, thereby improving the accuracy and effectiveness of signal monitoring and early warning.

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic monitoring and early warning of mobile phone signals, and in particular to a dynamic monitoring and early warning method and system for mobile phone signals. Background Art

[0002] The stability of mobile phone signal quality directly affects the user's call quality, data transmission rate and the normal use of various mobile applications, and is crucial to mobile communications;

[0003] With the rapid development of mobile communication technology and the surge in the number of users, the density of base station deployment continues to increase, and the wireless electromagnetic environment becomes increasingly complex, resulting in mobile phone signals being affected by various factors. In order to ensure user experience and network service quality, real-time monitoring and early warning of mobile phone signals have become particularly important;

[0004] Existing mobile phone signal monitoring and early warning methods mainly rely on the received signal strength indication and adopt a threshold-based early warning mechanism, which cannot accurately reflect the dynamic changes in signal quality and has poor adaptability to complex interference scenarios;

[0005] Therefore, a solution is urgently needed to solve the problems existing in the prior art. Summary of the invention

[0006] The embodiment of the present invention provides a method and system for dynamic monitoring and early warning of mobile phone signals, which can at least solve some of the problems existing in the prior art.

[0007] A first aspect of an embodiment of the present invention provides a dynamic monitoring and early warning method for mobile phone signals, comprising:

[0008] Collect the original signal of the mobile phone terminal, adaptively segment the original signal, divide the signal into a long-period signal segment, a medium-period signal segment and a short-period signal segment, reconstruct the signal trajectory of the long-period signal segment, establish a signal trajectory attractor model, perform cycle analysis on the medium-period signal segment, extract the signal periodic feature vector, extract the signal transient feature of the short-period signal segment, combine the signal trajectory attractor model, the signal periodic feature vector and the signal transient feature to construct an initial feature matrix, perform nonlinear dimensionality reduction to obtain a signal evolution feature space, construct a signal dynamics equation based on chaos theory and solve the corresponding Lyapunov exponent to obtain a signal system stability index, and combine the signal system stability index with the signal evolution feature space to obtain a composite feature space;

[0009] Partition mapping is performed on the composite feature space to establish a feature distribution density map, and the cluster center and divergence area of ​​the signal feature are identified on the feature distribution density map, and the cluster center is used as the signal steady-state area, and the divergence area is used as the signal transition area. In the signal steady-state area, a signal normal mode library based on recursive quantization analysis is established, and in the signal transition area, a signal abnormal mode library based on phase change theory is established. A hierarchical prediction model is trained based on the signal normal mode library and the signal abnormal mode library, and the data in the composite feature space is added to the trained hierarchical prediction model to obtain a prediction output result;

[0010] A signal quality warning decision tree is constructed based on the prediction output result, wherein the warning decision tree includes a signal attenuation warning branch, a signal interference warning branch and a signal interruption warning branch. If the signal attenuation warning branch is triggered, the signal attenuation rate is calculated and the optimal signal compensation strategy is selected based on the signal attenuation rate. If the signal interference warning branch is triggered, the interference source characteristics are analyzed and an interference avoidance plan is generated. If the signal interruption warning branch is triggered, an emergency response mechanism is initiated. According to the emergency response mechanism, an optimal signal transmission path is calculated based on the composite feature space, and available base station resources are evaluated on the optimal signal transmission path. According to the base station resource evaluation result, a base station switching sequence is generated and a control instruction containing a warning level, a base station switching sequence and an avoidance plan is sent to the control terminal, the warning response effect is monitored and the optimal disposal decision is dynamically generated.

[0011] In an optional embodiment,

[0012] The original signal of the mobile phone terminal is collected, and the original signal is adaptively segmented to divide the signal into a long-period signal segment, a medium-period signal segment and a short-period signal segment. The signal trajectory of the long-period signal segment is reconstructed, and a signal trajectory attractor model is established. The cyclic period analysis of the medium-period signal segment is performed to extract the signal periodic feature vector, and the signal transient feature of the short-period signal segment is extracted. The signal trajectory attractor model, the signal periodic feature vector and the signal transient feature are combined to construct an initial feature matrix, and nonlinear dimensionality reduction is performed to obtain a signal evolution feature space. Based on chaos theory, a signal dynamics equation is constructed and the corresponding Lyapunov exponent is solved to obtain a signal system stability index. The signal system stability index is combined with the signal evolution feature space to obtain a composite feature space including:

[0013] Collecting the original signal of the mobile terminal, wherein the original signal includes signal strength, signal quality, signal bandwidth and signal delay, performing adaptive segmentation processing on the original signal, and dividing the original signal into a long-period signal segment, a medium-period signal segment and a short-period signal segment based on the frequency characteristics and energy characteristics of the original signal, wherein the long-period signal segment corresponds to the channel fading trend of the original signal, the medium-period signal segment corresponds to the periodic fluctuation of the original signal, and the short-period signal segment corresponds to the burst interference of the original signal;

[0014] Reconstructing the signal trajectory of the long-period signal segment, mapping the long-period signal segment to a high-dimensional phase space by time delay embedding, performing cluster analysis on the signal trajectory in the high-dimensional phase space to obtain a high-dimensional cluster center, and constructing the high-dimensional cluster center as a signal trajectory attractor model;

[0015] Performing cyclic period analysis on the mesocycle signal segment, calculating the autocorrelation function and power spectrum of the mesocycle signal segment through autocorrelation analysis and Fourier transform, and extracting the period, amplitude and phase of the mesocycle signal segment based on the autocorrelation function and the power spectrum to form a signal period feature vector;

[0016] Performing wavelet transform and empirical mode decomposition on the short-period signal segment, extracting the energy distribution, instantaneous frequency and instantaneous amplitude of the short-period signal segment on different time scales as signal transient features, and combining the signal trajectory attractor model, the signal periodic feature vector and the signal transient features to construct an initial feature matrix;

[0017] The initial feature matrix is ​​subjected to nonlinear dimensionality reduction by the kernel principal component analysis method to obtain a signal evolution feature space, a signal dynamics equation is constructed based on chaos theory, the Lyapunov exponent spectrum of the signal dynamics equation is solved to obtain a signal system stability index, and the signal system stability index is combined with the signal evolution feature space to construct a composite feature space.

[0018] In an optional embodiment,

[0019] The initial feature matrix is ​​subjected to nonlinear dimension reduction by the kernel principal component analysis method, and the signal evolution feature space obtained includes:

[0020] Performing data center preprocessing on the initial feature matrix, where the number of rows of the initial feature matrix is ​​the number of samples and the number of columns is the feature dimension, calculating the mean vector of each feature, and subtracting the mean of the corresponding feature from each element in the initial feature matrix to obtain a centralized feature matrix;

[0021] Constructing a kernel function matrix, selecting a Gaussian radial basis kernel function as the kernel function, calculating the kernel matrix based on the centralized feature matrix, wherein the number of rows and columns of the kernel matrix are equal to the number of samples, each element of the kernel matrix is ​​composed of a Gaussian radial basis kernel function value between two random sample vectors, and the bandwidth parameter of the Gaussian radial basis kernel function is selected to have an optimal value by a cross-validation method;

[0022] The kernel matrix is ​​centralized, a row mean matrix and a column mean matrix are subtracted from the kernel matrix, and a global mean matrix is ​​added to obtain a centralized kernel matrix, an eigenvalue decomposition is performed on the centralized kernel matrix to obtain an eigenvalue diagonal matrix and an eigenvector matrix, an eigenvector corresponding to the maximum eigenvalue in the eigenvalue diagonal matrix is ​​selected based on the cumulative variance contribution rate, the eigenvector is normalized, and a projection matrix is ​​constructed by multiplying the normalized eigenvector with the inverse of the square root of the corresponding eigenvalue;

[0023] The centered kernel matrix is ​​multiplied by the projection matrix to obtain a signal evolution feature space after dimensionality reduction, wherein the number of rows of the signal evolution feature space is equal to the number of samples, and the number of columns is equal to the feature dimension after dimensionality reduction.

[0024] In an optional embodiment,

[0025] Partition mapping is performed on the composite feature space to establish a feature distribution density map, and the cluster center and divergence area of ​​the signal feature are identified on the feature distribution density map. The cluster center is used as the signal steady-state area, and the divergence area is used as the signal transition area. In the signal steady-state area, a signal normal mode library based on recursive quantization analysis is established, and in the signal transition area, a signal abnormal mode library based on phase change theory is established. A hierarchical prediction model is trained based on the signal normal mode library and the signal abnormal mode library, and the data in the composite feature space is added to the hierarchical prediction model after training, and the prediction output results obtained include:

[0026] Partition mapping is performed on the composite feature space, and the composite feature space is mapped to a two-dimensional plane by a self-organizing feature mapping method to obtain a feature distribution density map, in which a cluster center is determined based on a local maximum point of data density, and a divergent region is determined based on a local minimum point of data density, wherein the cluster center corresponds to a signal steady-state region, and the divergent region corresponds to a signal transition region;

[0027] A signal normal mode library is established in the signal steady-state region, and a recursive quantization analysis method is used to perform vector quantization and recursive clustering on feature samples in the signal steady-state region, and the feature samples are divided into multiple subsets according to the distance metric between the feature samples, and the mean vector of each subset is calculated to construct a quantization codebook, and the mean vector in the quantization codebook is used to replace the original feature sample for iterative clustering until the size of the quantization codebook reaches a preset threshold, and the final quantization codebook is used as the signal normal mode library;

[0028] Establishing a signal anomaly pattern library in the signal transition region, using a hidden Markov model to perform state modeling on feature samples in the signal transition region, training a state transfer matrix and an emission matrix of the hidden Markov model, and constructing the signal anomaly pattern library based on the state transfer matrix and the emission matrix;

[0029] A hierarchical prediction model is trained based on the signal normal mode library and the signal abnormal mode library, wherein the hierarchical prediction model includes a first prediction layer and a second prediction layer, wherein the first prediction layer uses a quantization codebook in the signal normal mode library to perform normal state discrimination on feature data in a composite feature space, and the second prediction layer uses a state transfer matrix and a transmission matrix in the signal abnormal mode library to perform abnormal mode recognition on feature data discriminated as abnormal by the first prediction layer;

[0030] The feature data in the composite feature space is input into the hierarchical prediction model to obtain a signal state discrimination result and an abnormal pattern recognition result and combine them into a prediction output result.

[0031] In an optional embodiment,

[0032] The recursive quantization analysis method is used to perform vector quantization on the characteristic samples in the steady-state region of the signal as shown in the following formula:

[0033] ;

[0034] Where D(X, C) represents the feature samples after vector quantization, X represents the set of all feature samples, C represents the codebook set used in vector quantization, K represents the number of vectors in the codebook, N represents the number of feature samples, i represents the feature index, L represents the number of feature dimensions, k represents the feature dimension index, and w represents the feature dimension index. k represents the weight factor on dimension k, GP k (x i ) represents the sample x i The result after Gaussian process mapping at scale k, c j is the codebook vector, representing the jth codeword in the codebook, λ represents the balancing weight parameter, Represents sample x iThe value after Gaussian process mapping and the codebook vector c j The distance between.

[0035] In an optional embodiment,

[0036] A signal quality warning decision tree is constructed based on the prediction output result, wherein the warning decision tree includes a signal attenuation warning branch, a signal interference warning branch and a signal interruption warning branch. If the signal attenuation warning branch is triggered, the signal attenuation rate is calculated and the optimal signal compensation strategy is selected based on the signal attenuation rate. If the signal interference warning branch is triggered, the interference source characteristics are analyzed and an interference avoidance plan is generated. If the signal interruption warning branch is triggered, an emergency response mechanism is started. According to the emergency response mechanism, the optimal signal transmission path is calculated based on the composite feature space, and the available base station resources are evaluated on the optimal signal transmission path. According to the base station resource evaluation result, a base station switching sequence is generated and a control instruction containing a warning level, a base station switching sequence and an avoidance plan is sent to the control terminal. Monitoring the warning response effect and dynamically generating the optimal disposal decision include:

[0037] Based on the prediction output result, a signal quality warning decision tree including a signal attenuation warning branch, a signal interference warning branch and a signal interruption warning branch is constructed;

[0038] Collect signal strength time series data corresponding to the original signal, perform first-order difference calculation on the signal strength time series data to obtain an attenuation rate, determine whether to trigger the signal attenuation warning branch based on the attenuation rate, and if triggered, compare the attenuation rate with a preset attenuation threshold. If the attenuation rate is greater than the attenuation threshold, switch the signal transmission channel; if the attenuation rate is less than the attenuation threshold, increase the signal transmission power;

[0039] Collect interference signal data, determine whether the interference warning branch is triggered based on the interference signal data, and if triggered, extract spectrum parameters, amplitude parameters and modulation parameters of the interference signal data, determine frequency band switching parameters and modulation scheme parameters based on the spectrum parameters, amplitude parameters and modulation parameters, switch the signal to the target frequency band according to the frequency band switching parameters, and adjust the signal modulation mode according to the modulation scheme parameters;

[0040] Based on the prediction output result, it is queried whether there is a preset external interruption event. If so, the signal interruption warning branch is triggered, the emergency response mechanism is started and the signal data of multiple alternative transmission paths are collected, and the signal strength index, signal-to-noise ratio index and transmission delay index of each alternative transmission path are calculated. The signal strength index, the signal-to-noise ratio index and the transmission delay index are input into a preset scoring model, and the quality score of each alternative transmission path is calculated in combination with the composite feature space, and the alternative transmission path with the highest quality score is selected as the optimal signal transmission path;

[0041] Detecting the number of channel resources, the number of time slot resources and the power resource capacity of each base station on the optimal signal transmission path, generating a base station switching sequence according to the number of channel resources, the number of time slot resources and the power resource capacity, sending the base station switching sequence to each base station, and controlling the signal to switch between the base stations according to the base station switching sequence;

[0042] Collect signal quality data after switching, calculate the quality index of the signal quality data, compare the quality index with the preset signal quality threshold, if the quality index is less than the signal quality threshold, recalculate the quality score of the alternative transmission path and update the base station switching sequence, monitor the early warning response effect based on the updated base station switching sequence and dynamically generate the optimal disposal decision according to the early warning response effect.

[0043] In an optional embodiment,

[0044] Detecting the number of channel resources, the number of time slot resources, and the power resource capacity of each base station on the optimal signal transmission path, generating a base station switching sequence according to the number of channel resources, the number of time slot resources, and the power resource capacity, sending the base station switching sequence to each base station, and controlling the signal to switch between the base stations according to the base station switching sequence includes:

[0045] Establishing a reinforcement learning model by analyzing historical switching data, the reinforcement learning model is used to learn the characteristic pattern of the optimal switching path, and generating an initial base station switching sequence based on the reinforcement learning model, the initial base station switching sequence is used to store the base stations passed in sequence during the switching process;

[0046] An evolutionary algorithm is used to optimize the weight coefficient of the switching cost function, wherein the switching cost function includes a resource utilization weight, a switching overhead weight, and a signal quality weight, and an optimal weight combination is searched through a crossover operation, a mutation operation, and a selection operation;

[0047] Constructing a fuzzy rule base of resource constraint conditions, using a fuzzy reasoning mechanism to evaluate the switching feasibility of each base station, and selecting feasible base stations that meet the resource constraints from all base stations on the optimal transmission path based on the fuzzy reasoning mechanism to form a subset of feasible base stations;

[0048] A swarm intelligence optimization algorithm is used to generate a base station switching sequence. The swarm intelligence optimization algorithm includes multiple agents. The agents exchange information through a pheromone mechanism and a speed update mechanism, explore the optimal switching path, and establish a deep reinforcement learning model. The deep reinforcement learning model optimizes the time slot configuration strategy through continuous interaction with the network environment and adjusts the time slot length and time slot interval according to the feedback reward signal of the switching performance;

[0049] A machine learning algorithm is used to predict the resource occupancy trend of each base station, and when a risk of resource overlimit is predicted, the base station switching sequence and resource allocation scheme are adjusted;

[0050] Determine whether the feasible base station subset is empty. If the feasible base station subset is empty, match historical switching failure cases based on the case reasoning mechanism to generate exception handling suggestions. If the feasible base station subset is not empty, send the base station switching sequence to each base station, and control the signal to switch between the base stations according to the base station switching sequence.

[0051] A second aspect of an embodiment of the present invention provides a dynamic monitoring and early warning system for mobile phone signals, comprising:

[0052] The first unit is used to collect the original signal of the mobile phone terminal, adaptively segment the original signal, divide the signal into a long-period signal segment, a medium-period signal segment and a short-period signal segment, reconstruct the signal trajectory of the long-period signal segment, establish a signal trajectory attractor model, perform cycle analysis on the medium-period signal segment, extract the signal periodic feature vector, extract the signal transient feature of the short-period signal segment, combine the signal trajectory attractor model, the signal periodic feature vector and the signal transient feature to construct an initial feature matrix, perform nonlinear dimensionality reduction to obtain a signal evolution feature space, construct a signal dynamics equation based on chaos theory and solve the corresponding Lyapunov exponent to obtain a signal system stability index, and combine the signal system stability index with the signal evolution feature space to obtain a composite feature space;

[0053] The second unit is used to partition and map the composite feature space, establish a feature distribution density map, identify the cluster center and divergence area of ​​the signal feature on the feature distribution density map, use the cluster center as the signal steady-state area, and the divergence area as the signal transition area. In the signal steady-state area, a signal normal mode library based on recursive quantization analysis is established, and in the signal transition area, a signal abnormal mode library based on phase change theory is established. A hierarchical prediction model is trained based on the signal normal mode library and the signal abnormal mode library, and the data in the composite feature space is added to the trained hierarchical prediction model to obtain a prediction output result;

[0054] The third unit is used to construct a signal quality warning decision tree based on the prediction output result, wherein the warning decision tree includes a signal attenuation warning branch, a signal interference warning branch and a signal interruption warning branch. If the signal attenuation warning branch is triggered, the signal attenuation rate is calculated and the optimal signal compensation strategy is selected based on the signal attenuation rate. If the signal interference warning branch is triggered, the interference source characteristics are analyzed and an interference avoidance plan is generated. If the signal interruption warning branch is triggered, an emergency response mechanism is initiated. According to the emergency response mechanism, the optimal signal transmission path is calculated based on the composite feature space, and the available base station resources are evaluated on the optimal signal transmission path. According to the base station resource evaluation result, a base station switching sequence is generated and a control instruction containing a warning level, a base station switching sequence and an avoidance plan is sent to the control terminal, the warning response effect is monitored and the optimal disposal decision is dynamically generated.

[0055] According to a third aspect of the embodiments of the present invention,

[0056] An electronic device is provided, comprising:

[0057] processor;

[0058] a memory for storing processor-executable instructions;

[0059] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0060] A fourth aspect of the embodiments of the present invention is:

[0061] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0062] In the present invention, through multi-dimensional feature extraction and composite feature space construction, the signal evolution law can be more comprehensively characterized, and combined with chaos theory and phase change theory, the steady state and transition state of the signal can be effectively identified, thereby improving the accuracy of signal quality warning and avoiding false alarms and missed alarms. By adopting a hierarchical prediction model and a warning decision tree, it is possible to intelligently judge abnormal situations such as signal attenuation, interference and interruption based on the prediction output results, and provide corresponding disposal plans for different abnormal situations, such as signal compensation, interference avoidance and base station switching, etc., to achieve intelligent warning and disposal of signal quality. By calculating the optimal signal transmission path and evaluating the available base station resources, it is possible to dynamically switch base stations to ensure the communication quality of users under abnormal signal conditions, while optimizing network resource utilization and avoiding resource waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a flowchart of a method for dynamic monitoring and early warning of mobile phone signals according to an embodiment of the present invention;

[0064] Figure 2 The figure is a schematic diagram of the structure of a dynamic monitoring and early warning system for mobile phone signals according to an embodiment of the present invention. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings 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.

[0066] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0067] Figure 1 FIG. 1 is a flow chart of a method for dynamic monitoring and early warning of mobile phone signals according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0068] S1. Collect the original signal of the mobile phone terminal, adaptively segment the original signal, divide the signal into a long-period signal segment, a medium-period signal segment and a short-period signal segment, reconstruct the signal trajectory of the long-period signal segment, establish a signal trajectory attractor model, perform cycle analysis on the medium-period signal segment, extract the signal periodic feature vector, extract the signal transient feature of the short-period signal segment, combine the signal trajectory attractor model, the signal periodic feature vector and the signal transient feature to construct an initial feature matrix, perform nonlinear dimensionality reduction to obtain a signal evolution feature space, construct a signal dynamics equation based on chaos theory and solve the corresponding Lyapunov exponent to obtain a signal system stability index, and combine the signal system stability index with the signal evolution feature space to obtain a composite feature space;

[0069] The signal trajectory reconstruction is a method of re-embedding one-dimensional signal data into a high-dimensional space by analyzing time series data to reveal its inherent dynamic characteristics. The signal trajectory attractor model is a mathematical model that describes the evolution trajectory of the signal in the high-dimensional phase space and is used to reveal the stable state and chaotic characteristics of the system. The cycle analysis is a technical method for analyzing the periodic changes of the signal and mining the periodic laws and influencing factors hidden in the signal. The signal evolution feature space is a multidimensional feature space constructed based on the dynamic characteristics of the signal, which is used to describe the evolution law of the signal in the time dimension. The chaos theory is a theory for studying the complex behavior of nonlinear dynamic systems, and is widely used to analyze the chaotic characteristics and dynamic evolution laws existing in signals. The signal dynamics equation is a mathematical model that describes the signal change process and is used to reveal the dynamic mechanism of the signal changing over time. The Lyapunov exponent is an important indicator for measuring the sensitivity to initial values ​​in a dynamic system and is used to determine whether the system behavior tends to be stable or exhibits chaotic characteristics.

[0070] In an optional embodiment,

[0071] The original signal of the mobile phone terminal is collected, and the original signal is adaptively segmented to divide the signal into a long-period signal segment, a medium-period signal segment and a short-period signal segment. The signal trajectory of the long-period signal segment is reconstructed, and a signal trajectory attractor model is established. The cyclic period analysis of the medium-period signal segment is performed to extract the signal periodic feature vector, and the signal transient feature of the short-period signal segment is extracted. The signal trajectory attractor model, the signal periodic feature vector and the signal transient feature are combined to construct an initial feature matrix, and nonlinear dimensionality reduction is performed to obtain a signal evolution feature space. Based on chaos theory, a signal dynamics equation is constructed and the corresponding Lyapunov exponent is solved to obtain a signal system stability index. The signal system stability index is combined with the signal evolution feature space to obtain a composite feature space including:

[0072] Collecting the original signal of the mobile terminal, wherein the original signal includes signal strength, signal quality, signal bandwidth and signal delay, performing adaptive segmentation processing on the original signal, and dividing the original signal into a long-period signal segment, a medium-period signal segment and a short-period signal segment based on the frequency characteristics and energy characteristics of the original signal, wherein the long-period signal segment corresponds to the channel fading trend of the original signal, the medium-period signal segment corresponds to the periodic fluctuation of the original signal, and the short-period signal segment corresponds to the burst interference of the original signal;

[0073] Reconstructing the signal trajectory of the long-period signal segment, mapping the long-period signal segment to a high-dimensional phase space by time delay embedding, performing cluster analysis on the signal trajectory in the high-dimensional phase space to obtain a high-dimensional cluster center, and constructing the high-dimensional cluster center as a signal trajectory attractor model;

[0074] Performing cyclic period analysis on the mesocycle signal segment, calculating the autocorrelation function and power spectrum of the mesocycle signal segment through autocorrelation analysis and Fourier transform, and extracting the period, amplitude and phase of the mesocycle signal segment based on the autocorrelation function and the power spectrum to form a signal period feature vector;

[0075] Performing wavelet transform and empirical mode decomposition on the short-period signal segment, extracting the energy distribution, instantaneous frequency and instantaneous amplitude of the short-period signal segment on different time scales as signal transient features, and combining the signal trajectory attractor model, the signal periodic feature vector and the signal transient features to construct an initial feature matrix;

[0076] The initial feature matrix is ​​subjected to nonlinear dimensionality reduction by the kernel principal component analysis method to obtain a signal evolution feature space, a signal dynamics equation is constructed based on chaos theory, the Lyapunov exponent spectrum of the signal dynamics equation is solved to obtain a signal system stability index, and the signal system stability index is combined with the signal evolution feature space to construct a composite feature space.

[0077] The high-dimensional phase space is a mathematical space constructed by embedding a one-dimensional time series into a multi-dimensional space, and is used to analyze the dynamic characteristics of complex signals. The kernel principal component analysis method is a nonlinear dimensionality reduction method that combines kernel functions with principal component analysis, and is used to extract the main features in high-dimensional data.

[0078] Collect the original signal of the mobile terminal, including signal strength, signal quality, signal bandwidth and signal delay. For example, collect the signal strength value every 1 millisecond and record it as -80dBm, the signal quality value is 30, the signal bandwidth is 5MHz, and the signal delay is 20ms. Continue to collect for a period of time, such as 1 minute, to obtain an original signal sequence containing multiple data points.

[0079] Adaptively segment the original signal. According to the changes in signal frequency and energy, the original signal is divided into long-period, medium-period and short-period signal segments. For example, by observing the changes in the signal strength value, it is found that the signal strength shows a slow decay trend in a longer time range (such as 10 seconds), then the signal data of these 10 seconds is divided into a long-period signal segment. Similarly, if it is found that the signal strength shows periodic fluctuations within a few seconds, it is divided into a medium-period signal segment. For sudden changes at the millisecond level, such as a sudden drop or rise in signal strength, it is divided into a short-period signal segment. The specific division method can adjust the size and threshold of the time window according to the actual situation. Assume that in a 1-minute original signal, 3 long-period signal segments are identified, each lasting about 10 seconds; 10 medium-period signal segments are identified, each lasting about 2 seconds; and 50 short-period signal segments are identified, each lasting about 50 milliseconds.

[0080] The signal trajectory of the long-period signal segment is reconstructed and the signal trajectory attractor model is established. Taking the first long-period signal segment (10 seconds of data) as an example, the data of the four dimensions of signal strength, signal quality, signal bandwidth and signal delay are constructed into a high-dimensional phase space. The time delay embedding technology is used to map this 10-second signal sequence to a higher-dimensional space. For example, taking the delay time as 10 milliseconds, the signal strength, signal quality, signal bandwidth and signal delay at each moment and their values ​​at the next 10 milliseconds, 20 milliseconds, etc. are combined into a high-dimensional vector. The signal data at all moments are processed in this way to obtain a trajectory in a high-dimensional phase space. This trajectory is clustered and analyzed to find the cluster center. The cluster center represents the attractor of the signal trajectory. For example, through the K-means clustering algorithm, the trajectory points in the high-dimensional phase space are clustered into 3 categories, and 3 cluster centers are obtained. These 3 cluster centers constitute the signal trajectory attractor model.

[0081] Perform cyclic period analysis on the mesocycle signal segment and extract the signal period feature vector. Taking the first mesocycle signal segment (2 seconds of data) as an example, use autocorrelation analysis to calculate the autocorrelation function of the signal, and use Fourier transform to calculate the power spectrum of the signal. The period, amplitude and phase information of the signal can be extracted from the autocorrelation function and the power spectrum. For example, by analyzing the power spectrum, it is found that the main frequency component of the signal is 1Hz, the corresponding amplitude is 10, and the phase is 0.5π. Combine the period, amplitude and phase into a vector, such as [1, 10, 0.5π], as the signal period feature vector of the mesocycle signal segment. Perform similar processing on all mesocycle signal segments.

[0082] Extract the transient features of the short-period signal segment. Take the first short-period signal segment (50 milliseconds of data) as an example, use wavelet transform and empirical mode decomposition to extract the energy distribution, instantaneous frequency and instantaneous amplitude of the signal at different time scales. For example, through wavelet transform, it is found that the energy of the signal is higher at the 10 millisecond scale, and the corresponding instantaneous frequency is 5Hz and the instantaneous amplitude is 2. Combine this information into a vector, such as [10, 5, 2], as the transient feature of the signal of the short-period signal segment. Perform similar processing on all short-period signal segments.

[0083] Construct an initial feature matrix. Combine the signal trajectory attractor models of all long-period signal segments, the signal periodic feature vectors of all medium-period signal segments, and the signal transient features of all short-period signal segments into an initial feature matrix. For example, combine the attractor models of 3 long-period signal segments, the periodic feature vectors of 10 medium-period signal segments, and the transient features of 50 short-period signal segments into one matrix.

[0084] The kernel principal component analysis method is used to reduce the dimension of the initial feature matrix to obtain a low-dimensional signal evolution feature space. Based on the signal evolution feature space, the signal dynamics equation is constructed. Then, the Lyapunov index spectrum of the equation is solved to obtain the stability index of the signal system. For example, the maximum Lyapunov index calculated is 0.5, indicating that the signal system is in a chaotic state.

[0085] The signal system stability index is combined with the signal evolution feature space to construct a composite feature space.

[0086] In this embodiment, through multi-scale analysis and nonlinear dimensionality reduction, the dynamic characteristics of the signal can be captured more comprehensively, thereby improving the accuracy of signal analysis. Through nonlinear dimensionality reduction, the feature dimension can be effectively reduced, thereby reducing the computational complexity and improving the efficiency of signal processing. The essential characteristics of the signal can be effectively extracted, and it has strong robustness against noise and interference, thereby enhancing the reliability of signal recognition.

[0087] In an optional embodiment,

[0088] The initial feature matrix is ​​subjected to nonlinear dimension reduction by the kernel principal component analysis method, and the signal evolution feature space obtained includes:

[0089] Performing data center preprocessing on the initial feature matrix, where the number of rows of the initial feature matrix is ​​the number of samples and the number of columns is the feature dimension, calculating the mean vector of each feature, and subtracting the mean of the corresponding feature from each element in the initial feature matrix to obtain a centralized feature matrix;

[0090] Constructing a kernel function matrix, selecting a Gaussian radial basis kernel function as the kernel function, calculating the kernel matrix based on the centralized feature matrix, wherein the number of rows and columns of the kernel matrix are equal to the number of samples, each element of the kernel matrix is ​​composed of a Gaussian radial basis kernel function value between two random sample vectors, and the bandwidth parameter of the Gaussian radial basis kernel function is selected to have an optimal value by a cross-validation method;

[0091] The kernel matrix is ​​centralized, a row mean matrix and a column mean matrix are subtracted from the kernel matrix, and a global mean matrix is ​​added to obtain a centralized kernel matrix, an eigenvalue decomposition is performed on the centralized kernel matrix to obtain an eigenvalue diagonal matrix and an eigenvector matrix, an eigenvector corresponding to the maximum eigenvalue in the eigenvalue diagonal matrix is ​​selected based on the cumulative variance contribution rate, the eigenvector is normalized, and a projection matrix is ​​constructed by multiplying the normalized eigenvector with the inverse of the square root of the corresponding eigenvalue;

[0092] The centered kernel matrix is ​​multiplied by the projection matrix to obtain a signal evolution feature space after dimensionality reduction, wherein the number of rows of the signal evolution feature space is equal to the number of samples, and the number of columns is equal to the feature dimension after dimensionality reduction.

[0093] The Gaussian radial basis kernel function is a commonly used kernel function based on the Euclidean distance between sample points, and is commonly used in algorithms such as support vector machines and kernel principal component analysis. The bandwidth parameter is an important hyperparameter of the Gaussian radial basis kernel function, which determines the perception range of the kernel function, thereby affecting the similarity calculation results between data points.

[0094] Preprocess the raw signal data. Assume that the collected signal data forms a matrix, where each row represents a time sample and each column represents a signal feature. In order to eliminate the influence of dimensions between different features, the data needs to be centered. Specifically, calculate the average value of each feature over all samples, and then subtract the average value of the feature from the corresponding feature value of each sample.

[0095] Construct a kernel function matrix and select the Gaussian radial basis kernel function to calculate the similarity between samples. For each pair of samples, calculate the Gaussian radial basis kernel function value, which reflects the distance between the two samples in the high-dimensional space. The kernel function values ​​of all sample pairs are combined into a square matrix, namely the kernel matrix. A key parameter in the kernel function is the bandwidth, which determines the width of the Gaussian function. The choice of bandwidth has a great influence on the dimensionality reduction effect, and the cross-validation method is usually used to select the best bandwidth value. For example, divide the data set into a training set and a validation set, try different bandwidth values, and select the bandwidth value with the best performance on the validation set.

[0096] Center the kernel matrix. Calculate the average value of each row, each column, and all elements of the kernel matrix. Subtract the average value of each row and column from each element of the kernel matrix, and add the average value of all elements. The resulting matrix is ​​the centralized kernel matrix.

[0097] Perform eigenvalue decomposition on the centralized kernel matrix. Eigenvalue decomposition can decompose the matrix into eigenvalues ​​and eigenvectors. Arrange the eigenvalues ​​in descending order and calculate the variance contribution rate corresponding to each eigenvalue. The variance contribution rate reflects the contribution of each eigenvalue in explaining the data variance. Select the first several largest eigenvalues ​​and their corresponding eigenvectors whose cumulative variance contribution rate reaches a preset threshold (for example, 95%). Normalize these eigenvectors, that is, scale the length of each eigenvector to 1. Then, multiply each normalized eigenvector by the inverse of the square root of its corresponding eigenvalue to obtain the projection matrix.

[0098] Multiply the central kernel matrix by the projection matrix to obtain the reduced-dimensional signal evolution feature space. The number of rows in this space is the same as the number of original samples, and the number of columns is equal to the number of selected feature vectors, that is, the feature dimension after dimensionality reduction.

[0099] For example, suppose there are 100 time samples, each with 10 features. After the above steps, the 10-dimensional feature space can be reduced to 3 dimensions. Suppose that after cross-validation, the bandwidth is selected as 1. Suppose that after eigenvalue decomposition, the cumulative variance contribution rate of the first 3 eigenvalues ​​reaches 95%. The final signal evolution feature space is a matrix with 100 rows and 3 columns, and each row represents the low-dimensional feature representation of a sample.

[0100] In this embodiment, by mapping high-dimensional data to low-dimensional space, the subsequent analysis and processing process can be simplified, which can effectively reduce the data dimension, reduce the computational complexity and storage space. The kernel principal component analysis can effectively capture the nonlinear structure in the data through the introduction of kernel functions. By selecting the eigenvectors corresponding to the eigenvalues ​​with higher variance contribution rates, the features that contribute most to the signal changes can be extracted, thereby better understanding the evolution law of the signal.

[0101] S2. Partition mapping is performed on the composite feature space to establish a feature distribution density map, and the cluster center and divergence area of ​​the signal feature are identified on the feature distribution density map, and the cluster center is used as the signal steady-state area, and the divergence area is used as the signal transition area. In the signal steady-state area, a signal normal mode library based on recursive quantization analysis is established, and in the signal transition area, a signal abnormal mode library based on phase change theory is established. A hierarchical prediction model is trained based on the signal normal mode library and the signal abnormal mode library, and the data in the composite feature space is added to the trained hierarchical prediction model to obtain a prediction output result;

[0102] The partition mapping is a process of dividing signals or data into different areas in a specific space according to certain criteria for classification, processing or analysis. The feature distribution density map is a graph drawn by statistically analyzing the distribution of signal or data features, which is used to intuitively display the concentration trend and variation range of the features. The signal steady-state region refers to the region where the signal exhibits stable characteristics and no obvious changes over a period of time, which is usually used to identify the normal working state of the system. The signal transition region is the process region where the signal changes from one stable state to another, and has significant dynamic characteristics and pattern changes. The signal normal mode library based on recursive quantization analysis is a signal normal mode database established by a recursive quantization method, which is used to store and identify common signal patterns and serve as a benchmark for normal behavior. The signal abnormal mode library based on phase change theory is a signal abnormal mode library constructed based on phase change theory, which is used to identify and store abnormal behavior patterns of signals under specific conditions.

[0103] In an optional embodiment,

[0104] Partition mapping is performed on the composite feature space to establish a feature distribution density map, and the cluster center and divergence area of ​​the signal feature are identified on the feature distribution density map. The cluster center is used as the signal steady-state area, and the divergence area is used as the signal transition area. In the signal steady-state area, a signal normal mode library based on recursive quantization analysis is established, and in the signal transition area, a signal abnormal mode library based on phase change theory is established. A hierarchical prediction model is trained based on the signal normal mode library and the signal abnormal mode library, and the data in the composite feature space is added to the hierarchical prediction model after training, and the prediction output results obtained include:

[0105] Partition mapping is performed on the composite feature space, and the composite feature space is mapped to a two-dimensional plane by a self-organizing feature mapping method to obtain a feature distribution density map, in which a cluster center is determined based on a local maximum point of data density, and a divergent region is determined based on a local minimum point of data density, wherein the cluster center corresponds to a signal steady-state region, and the divergent region corresponds to a signal transition region;

[0106] A signal normal mode library is established in the signal steady-state region, and a recursive quantization analysis method is used to perform vector quantization and recursive clustering on feature samples in the signal steady-state region, and the feature samples are divided into multiple subsets according to the distance metric between the feature samples, and the mean vector of each subset is calculated to construct a quantization codebook, and the mean vector in the quantization codebook is used to replace the original feature sample for iterative clustering until the size of the quantization codebook reaches a preset threshold, and the final quantization codebook is used as the signal normal mode library;

[0107] Establishing a signal anomaly pattern library in the signal transition region, using a hidden Markov model to perform state modeling on feature samples in the signal transition region, training a state transfer matrix and an emission matrix of the hidden Markov model, and constructing the signal anomaly pattern library based on the state transfer matrix and the emission matrix;

[0108] A hierarchical prediction model is trained based on the signal normal mode library and the signal abnormal mode library, wherein the hierarchical prediction model includes a first prediction layer and a second prediction layer, wherein the first prediction layer uses a quantization codebook in the signal normal mode library to perform normal state discrimination on feature data in a composite feature space, and the second prediction layer uses a state transfer matrix and a transmission matrix in the signal abnormal mode library to perform abnormal mode recognition on feature data discriminated as abnormal by the first prediction layer;

[0109] The feature data in the composite feature space is input into the hierarchical prediction model to obtain a signal state discrimination result and an abnormal pattern recognition result and combine them into a prediction output result.

[0110] The quantization codebook is a code table used to quantize signals or data, containing a set of discrete quantization values ​​for mapping the original signal to a finite digital representation. The hidden Markov model is a statistical model that assumes that the state of the system is implicit and infers the probabilities of these states and their transition laws through the observed output sequence. The state modeling is the process of using mathematical models to describe the possible states of the system and their transition laws, which is usually used for the analysis and prediction of dynamic systems. The emission matrix is ​​an important component of the hidden Markov model, which represents the probability distribution of observing a certain output under a given hidden state.

[0111] Partition mapping of the composite feature space. The self-organizing feature mapping method is used to map the high-dimensional composite feature space to a two-dimensional plane. This process can be imagined as unfolding a crumpled paper ball and flattening it to make the feature distribution clearer. On this two-dimensional plane, the density distribution of the feature data presents a peak-to-valley pattern. High-density areas represent feature aggregation, and low-density areas represent feature dispersion. By finding the local highest point of data density, the cluster centers are determined. These cluster centers represent the steady-state areas of the signal, like the peaks on the map. Conversely, by finding the local lowest point of data density, the divergent areas are determined. These divergent areas represent the transition areas of the signal, like the valleys on the map. For example, assuming that the density distribution of feature data on a two-dimensional plane presents three obvious peaks and two obvious valleys, three cluster centers and two divergent areas can be determined.

[0112] A signal normal mode library is established in the signal steady-state region. Recursive quantization analysis is performed on the feature samples in each steady-state region. It can be understood as grouping and classifying the data points on each peak, dividing the feature samples into multiple subsets according to the distance between the feature samples, and calculating the average feature vector of each subset to form a quantization codebook, just like representing the data points on each peak with several representative points, replacing the original data points with these representative points, performing iterative clustering, and continuously streamlining the number of representative points until the size of the quantization codebook reaches a preset threshold, for example, the codebook size is 100. The final quantization codebook constitutes the signal normal mode library, which can be regarded as a standard model for each peak. Assuming that a certain steady-state region has obtained 100 representative feature vectors after recursive quantization analysis, the normal mode library of the region contains these 100 vectors.

[0113] A signal anomaly pattern library is established in the signal transition region. A hidden Markov model is used to perform state modeling on the feature samples in each transition region. This can be understood as a path analysis of the data points in the valley. By training the hidden Markov model, the probability of the signal transitioning between different states (state transfer matrix) and the probability of emitting specific features in each state (emission matrix) are learned. These state transfer matrices and emission matrices constitute the signal anomaly pattern library, which can be regarded as a probability model of different paths in the valley. For example, assuming that the hidden Markov model of a transition region contains three states, the state transfer matrix is ​​a 3x3 matrix, which represents the probability of transitioning between states; the emission matrix determines the size according to the feature dimension, which represents the probability of emitting specific features in each state.

[0114] The hierarchical prediction model is trained based on the signal normal mode library and the signal abnormal mode library. The model consists of two layers: the first prediction layer and the second prediction layer. The first prediction layer uses the quantization codebook in the signal normal mode library to judge the normal state of the input feature data, just like comparing the new data point with the standard model of the peak to determine whether it belongs to the normal state. If the first prediction layer determines that it is abnormal, the data is passed to the second prediction layer. The second prediction layer uses the state transfer matrix and emission matrix in the signal abnormal mode library to identify the abnormal mode of the abnormal data, just like comparing the new data point with the probability model of different paths in the valley to identify the abnormal mode to which it belongs. For example, suppose a new feature data is input, and the first prediction layer determines that it does not belong to any steady-state area, then it is passed to the second prediction layer. The second prediction layer determines which abnormal mode the data is most likely to belong to based on the hidden Markov model.

[0115] The feature data in the composite feature space is input into the trained hierarchical prediction model to obtain the signal state discrimination result and the abnormal pattern recognition result, and the two are combined as the final prediction output result.

[0116] In this embodiment, by constructing a library of normal and abnormal signal patterns and adopting a hierarchical prediction model, the abnormal state of the signal can be identified more accurately, the false alarm rate and the missed alarm rate can be reduced, and the signal transition region can be modeled using a hidden Markov model, which can identify different abnormal patterns and provide more detailed abnormal information. By adopting methods such as recursive quantitative analysis and self-organizing feature mapping, high-dimensional, nonlinear complex signals can be effectively processed, and the robustness and generalization ability of the model can be improved.

[0117] In an optional embodiment,

[0118] The recursive quantization analysis method is used to perform vector quantization on the characteristic samples in the steady-state region of the signal as shown in the following formula:

[0119] ;

[0120] Where D(X, C) represents the feature samples after vector quantization, X represents the set of all feature samples, C represents the codebook set used in vector quantization, K represents the number of vectors in the codebook, N represents the number of feature samples, i represents the feature index, L represents the number of feature dimensions, k represents the feature dimension index, and w represents the feature dimension index. k represents the weight factor on dimension k, GP k (x i ) represents the sample x i The result after Gaussian process mapping at scale k, c j is the codebook vector, representing the jth codeword in the codebook, λ represents the balancing weight parameter, Represents sample xi The value after Gaussian process mapping and the codebook vector c j The distance between.

[0121] In this embodiment, by introducing Gaussian process mapping, the nonlinear relationship between features can be better captured, thereby improving the accuracy of quantization. By assigning different weight factors to each feature dimension, the role of important features can be highlighted, thereby enhancing the feature differentiation ability. Gaussian process mapping can effectively handle noise and outliers, thereby improving the robustness of quantization.

[0122] S3. Construct a signal quality warning decision tree based on the prediction output result, wherein the warning decision tree includes a signal attenuation warning branch, a signal interference warning branch and a signal interruption warning branch. If the signal attenuation warning branch is triggered, the signal attenuation rate is calculated and the optimal signal compensation strategy is selected based on the signal attenuation rate. If the signal interference warning branch is triggered, the interference source characteristics are analyzed and an interference avoidance plan is generated. If the signal interruption warning branch is triggered, an emergency response mechanism is initiated. According to the emergency response mechanism, the optimal signal transmission path is calculated based on the composite feature space, and the available base station resources are evaluated on the optimal signal transmission path. According to the base station resource evaluation result, a base station switching sequence is generated and a control instruction containing a warning level, a base station switching sequence and an avoidance plan is sent to the control terminal. The warning response effect is monitored and the optimal disposal decision is dynamically generated.

[0123] The signal quality warning decision tree is an algorithm model based on a decision tree, which is used to classify the signal quality according to multi-dimensional signal characteristics and timely identify the abnormality or potential problems of the signal. The signal attenuation warning branch is a branch in the decision tree, which is responsible for judging whether the signal has a warning of quality degradation according to the attenuation of the signal strength, and is usually used to detect the early stage of signal attenuation. The signal interference warning branch is another branch in the decision tree, which is specifically used to detect the situation where the signal is interfered with by external interference, such as noise or interference from other signal sources, and send out a warning signal. The signal interruption warning branch refers to a branch of the decision tree, which is used to judge whether the signal has a fault condition such as interruption or disconnection, and send out an interruption warning signal in time. The optimal signal compensation strategy is a strategy formulated to restore or improve the signal quality, which usually includes measures such as adjusting the signal transmission power, adjusting the frequency, and using compensation technology to ensure the stability and quality of the signal. The interference source feature is a descriptive feature of the interference source, which usually includes information such as the frequency, strength, duration, and change mode of the interference source, so as to facilitate the identification and suppression of the interference source. The base station switching sequence refers to the sequence and process of a series of base station switching events formed when the user equipment switches between different base stations in the mobile communication network, which affects the network quality and service stability.

[0124] In an optional embodiment,

[0125] A signal quality warning decision tree is constructed based on the prediction output result, wherein the warning decision tree includes a signal attenuation warning branch, a signal interference warning branch and a signal interruption warning branch. If the signal attenuation warning branch is triggered, the signal attenuation rate is calculated and the optimal signal compensation strategy is selected based on the signal attenuation rate. If the signal interference warning branch is triggered, the interference source characteristics are analyzed and an interference avoidance plan is generated. If the signal interruption warning branch is triggered, an emergency response mechanism is started. According to the emergency response mechanism, the optimal signal transmission path is calculated based on the composite feature space, and the available base station resources are evaluated on the optimal signal transmission path. According to the base station resource evaluation result, a base station switching sequence is generated and a control instruction containing a warning level, a base station switching sequence and an avoidance plan is sent to the control terminal. Monitoring the warning response effect and dynamically generating the optimal disposal decision include:

[0126] Based on the prediction output result, a signal quality warning decision tree including a signal attenuation warning branch, a signal interference warning branch and a signal interruption warning branch is constructed;

[0127] Collect signal strength time series data corresponding to the original signal, perform first-order difference calculation on the signal strength time series data to obtain an attenuation rate, determine whether to trigger the signal attenuation warning branch based on the attenuation rate, and if triggered, compare the attenuation rate with a preset attenuation threshold. If the attenuation rate is greater than the attenuation threshold, switch the signal transmission channel; if the attenuation rate is less than the attenuation threshold, increase the signal transmission power;

[0128] Collect interference signal data, determine whether the interference warning branch is triggered based on the interference signal data, and if triggered, extract spectrum parameters, amplitude parameters and modulation parameters of the interference signal data, determine frequency band switching parameters and modulation scheme parameters based on the spectrum parameters, amplitude parameters and modulation parameters, switch the signal to the target frequency band according to the frequency band switching parameters, and adjust the signal modulation mode according to the modulation scheme parameters;

[0129] Based on the prediction output result, it is queried whether there is a preset external interruption event. If so, the signal interruption warning branch is triggered, the emergency response mechanism is started and the signal data of multiple alternative transmission paths are collected, and the signal strength index, signal-to-noise ratio index and transmission delay index of each alternative transmission path are calculated. The signal strength index, the signal-to-noise ratio index and the transmission delay index are input into a preset scoring model, and the quality score of each alternative transmission path is calculated in combination with the composite feature space, and the alternative transmission path with the highest quality score is selected as the optimal signal transmission path;

[0130] Detecting the number of channel resources, the number of time slot resources and the power resource capacity of each base station on the optimal signal transmission path, generating a base station switching sequence according to the number of channel resources, the number of time slot resources and the power resource capacity, sending the base station switching sequence to each base station, and controlling the signal to switch between the base stations according to the base station switching sequence;

[0131] Collect signal quality data after switching, calculate the quality index of the signal quality data, compare the quality index with the preset signal quality threshold, if the quality index is less than the signal quality threshold, recalculate the quality score of the alternative transmission path and update the base station switching sequence, monitor the early warning response effect based on the updated base station switching sequence and dynamically generate the optimal disposal decision according to the early warning response effect.

[0132] The first-order difference is a common method in signal processing. It describes the changing trend of the signal by calculating the difference between the signal at adjacent time points. It is usually used to remove high-frequency noise in the signal or detect changes in the signal. The time slot resource is a communication system in which time is divided into discrete time slots. Each time slot can be allocated to different users or tasks for data transmission, which is used to improve the resource utilization of the network.

[0133] Construct a signal quality warning decision tree. The decision tree contains three warning branches: signal attenuation warning branch, signal interference warning branch, and signal interruption warning branch. These three branches correspond to different signal abnormalities, so that targeted countermeasures can be taken.

[0134] Collect the signal strength time series data of the original signal, for example, record the signal strength value every 100 milliseconds. Perform first-order difference calculation on the collected time series data to obtain the signal attenuation rate. For example, if the current signal strength is -60dBm and the signal strength 100 milliseconds ago is -55dBm, then the attenuation rate is -5dBm / 100ms. Compare the calculated attenuation rate with the preset attenuation threshold (for example, -3dBm / 100ms). If the attenuation rate is greater than the threshold, indicating that the signal attenuation is serious, switch the signal transmission channel, for example, switch from channel 1 to channel 2. If the attenuation rate is less than the threshold, increase the signal transmission power, for example, increase the transmission power from 10mW to 15mW, to compensate for the signal attenuation.

[0135] Collect interference signal data. Analyze the collected interference signal data to determine whether the interference warning branch is triggered. The basis for judgment can be whether the intensity of the interference signal exceeds a preset threshold, such as -70dBm. If the interference warning branch is triggered, extract the spectrum parameters (such as the center frequency of the interference signal), amplitude parameters (such as the peak amplitude of the interference signal) and modulation parameters (such as the modulation method of the interference signal) of the interference signal data. Assume that the extracted spectrum parameters show that the interference signal is concentrated in the 900MHz frequency band, the amplitude parameters show that the peak amplitude of the interference signal is -50dBm, and the modulation parameters show that the interference signal uses FSK modulation. Based on these parameters, determine the frequency band switching parameters, such as switching the signal to the 950MHz frequency band, and determine the modulation scheme parameters, such as adjusting the signal modulation method to PSK modulation. According to the determined parameters, switch the signal to the target frequency band and adjust the signal modulation method.

[0136] Based on the prediction output results, check whether there is a pre-set external interruption event, such as base station failure or fiber interruption. If so, the signal interruption warning branch is triggered and the emergency response mechanism is started. Collect signal data of multiple alternative transmission paths. For example, collect signal data of path 1, path 2, and path 3. Calculate the signal strength index (e.g. -65dBm), signal-to-noise ratio index (e.g. 15dB), and transmission delay index (e.g. 20ms) of each alternative transmission path. Input these indicators into the preset scoring model, and combine the composite feature space (e.g. geographic location, base station load, etc.) to calculate the quality score of each alternative transmission path. Assume that path 1 scores 80 points, path 2 scores 90 points, and path 3 scores 70 points. Select path 2 with the highest score as the optimal signal transmission path.

[0137] Detect the number of channel resources (e.g., the number of available channels), the number of time slot resources (e.g., the number of available time slots), and the power resource capacity (e.g., the remaining power) of each base station on the optimal signal transmission path. Assume that the number of channel resources of base stations A, B, and C on path 2 are 10, 8, and 12, respectively, the number of time slot resources are 50, 40, and 60, respectively, and the power resource capacity is 100W, 80W, and 120W, respectively. Generate a base station switching sequence based on this resource information, such as A->B->C. Send the base station switching sequence to each base station, and the control signal switches between the base stations according to the sequence.

[0138] Collect signal quality data after switching and calculate its quality indicators, such as signal strength, signal-to-noise ratio, and transmission delay. Compare the calculated quality indicators with the preset signal quality thresholds. Assume that the signal strength threshold is -75dBm, the signal-to-noise ratio threshold is 10dB, and the transmission delay threshold is 30ms. If the quality indicator is lower than the threshold, recalculate the quality score of the alternative transmission path and update the base station switching sequence. Monitor the early warning response effect based on the updated base station switching sequence, and dynamically generate the optimal disposal decision based on the early warning response effect, such as adjusting the base station switching sequence, adjusting the signal transmission power, or adjusting the signal modulation method.

[0139] In this embodiment, by real-time monitoring of signal quality and issuing early warnings, problems such as signal attenuation, interference and interruption can be discovered and handled in a timely manner, thereby avoiding communication interruptions caused by signal quality problems, thereby improving the stability of signal transmission. By analyzing various indicators of alternative transmission paths, the path with the highest quality score is selected for transmission, and the base station switching sequence is dynamically adjusted according to actual conditions, which can effectively avoid sections with poor signal quality, thereby optimizing the signal transmission path and improving signal transmission quality. By evaluating base station resources and generating base station switching sequences, base station resources can be effectively allocated and utilized, avoiding resource waste and improving resource utilization efficiency.

[0140] In an optional embodiment,

[0141] Detecting the number of channel resources, the number of time slot resources, and the power resource capacity of each base station on the optimal signal transmission path, generating a base station switching sequence according to the number of channel resources, the number of time slot resources, and the power resource capacity, sending the base station switching sequence to each base station, and controlling the signal to switch between the base stations according to the base station switching sequence includes:

[0142] Establishing a reinforcement learning model by analyzing historical switching data, the reinforcement learning model is used to learn the characteristic pattern of the optimal switching path, and generating an initial base station switching sequence based on the reinforcement learning model, the initial base station switching sequence is used to store the base stations passed in sequence during the switching process;

[0143] An evolutionary algorithm is used to optimize the weight coefficient of the switching cost function, wherein the switching cost function includes a resource utilization weight, a switching overhead weight, and a signal quality weight, and an optimal weight combination is searched through a crossover operation, a mutation operation, and a selection operation;

[0144] Constructing a fuzzy rule base of resource constraint conditions, using a fuzzy reasoning mechanism to evaluate the switching feasibility of each base station, and selecting feasible base stations that meet the resource constraints from all base stations on the optimal transmission path based on the fuzzy reasoning mechanism to form a subset of feasible base stations;

[0145] A swarm intelligence optimization algorithm is used to generate a base station switching sequence. The swarm intelligence optimization algorithm includes multiple agents. The agents exchange information through a pheromone mechanism and a speed update mechanism, explore the optimal switching path, and establish a deep reinforcement learning model. The deep reinforcement learning model optimizes the time slot configuration strategy through continuous interaction with the network environment and adjusts the time slot length and time slot interval according to the feedback reward signal of the switching performance;

[0146] A machine learning algorithm is used to predict the resource occupancy trend of each base station, and when a risk of resource overlimit is predicted, the base station switching sequence and resource allocation scheme are adjusted;

[0147] Determine whether the feasible base station subset is empty. If the feasible base station subset is empty, match historical switching failure cases based on the case reasoning mechanism to generate exception handling suggestions. If the feasible base station subset is not empty, send the base station switching sequence to each base station, and control the signal to switch between the base stations according to the base station switching sequence.

[0148] The evolutionary algorithm is a type of optimization algorithm based on natural selection and genetic mechanisms. It continuously optimizes the solution space by simulating biological evolution processes, such as selection, crossover, mutation and other operations, and is widely used in optimization problems. The fuzzy rule base of the resource constraint condition is a rule base based on fuzzy logic, which is used to process and express how to make resource allocation decisions based on fuzzy rules when resources are limited. The fuzzy reasoning mechanism is the core part of the fuzzy system, which is used to input fuzzy data according to the rules of the fuzzy rule base and derive fuzzy conclusions through reasoning. It is usually used to process uncertain or imprecise information. The swarm intelligence optimization algorithm is a type of optimization algorithm that simulates collaborative behavior between groups or individuals. It finds the optimal solution to the problem through information exchange and collaboration between group intelligent agents. Common algorithms include particle swarm optimization, ant colony optimization, etc.

[0149] Collect and preprocess historical base station switching data, including switching time, switching base station ID, signal strength, resource occupancy, etc. For example, collect base station switching data in a certain area in the past month, clean and filter abnormal data, and build a historical switching data set. Then, use this data set to train a reinforcement learning model, such as a deep Q network (DQN), to learn the characteristic patterns of the optimal switching path. The model input can be the current base station status, signal strength, resource availability, etc., and the output is the next best switching base station. By continuously interacting with the simulation environment, the reinforcement learning model can learn the strategy of selecting the best switching base station in different situations and generate an initial base station switching sequence. For example, the sequence is [base station A, base station B, base station C], indicating that the signal passes through these three base stations in sequence.

[0150] Define a switching cost function that includes three indicators: resource utilization, switching overhead, and signal quality, and assign a corresponding weight coefficient to each indicator. For example, the resource utilization weight is 0.4, the switching overhead weight is 0.3, and the signal quality weight is 0.3. Use an evolutionary algorithm, such as a genetic algorithm, to optimize these weight coefficients. Through crossover, mutation, and selection operations, continuously iterate and search for the optimal weight combination to minimize the switching cost function. For example, after optimization, the weight coefficients may become 0.5 for resource utilization, 0.2 for switching overhead, and 0.3 for signal quality.

[0151] Construct a fuzzy rule base for resource constraints. For example, "if the base station load is high, the switching feasibility is low", "if the base station has sufficient remaining resources, the switching feasibility is high". Use the fuzzy reasoning mechanism to evaluate the switching feasibility of each base station on the optimal transmission path. For example, the load of base station A is 80%, and according to the fuzzy rules, its switching feasibility score is 0.2; the load of base station B is 30%, and its switching feasibility score is 0.8. Based on the fuzzy reasoning results, filter out feasible base stations that meet the resource constraints to form a subset of feasible base stations. For example, filter out base stations with feasibility scores higher than 0.5 to form a subset of feasible base stations [base station B, base station C].

[0152] A swarm intelligence optimization algorithm, such as the ant colony algorithm, is used to generate a base station switching sequence. Each agent represents a possible switching path. Through the pheromone mechanism and the speed update mechanism, agents can exchange information and explore the optimal switching path. At the same time, a deep reinforcement learning model, such as the deep deterministic policy gradient (DDPG) algorithm, is established to optimize the time slot configuration strategy through continuous interaction with the network environment. The time slot length and time slot interval are adjusted according to the feedback reward signal of the switching performance, such as throughput, delay, etc. For example, if the throughput is improved after switching, the reward is increased to encourage the model to continue exploring similar switching strategies.

[0153] Use machine learning algorithms, such as time series analysis, to predict resource occupancy trends of each base station. For example, predict that the load of base station A will reach 90% in the next hour. When the risk of resource overload is predicted, adjust the base station switching sequence and resource allocation plan. For example, switch part of the signal traffic from base station A to base station B with lower load.

[0154] Determine whether the feasible base station subset is empty. If it is empty, match the historical handover failure cases based on the case reasoning mechanism and generate abnormal handling suggestions. For example, if the handover fails due to a base station failure, it is recommended to enable the backup base station. If it is not empty, send the generated base station handover sequence to each base station, and the control signal switches between the base stations according to the sequence.

[0155] In this embodiment, through intelligent base station switching and resource allocation, the load of each base station can be effectively balanced, and excessive concentration or idleness of resources can be avoided, thereby improving the overall resource utilization rate. By optimizing the switching cost function and the group intelligent optimization algorithm, the optimal switching path can be found, the number of unnecessary switching times can be reduced, and the overhead caused by switching, such as delay and signaling overhead, can be reduced. By combining signal quality indicators for optimization, it can be ensured that during the switching process, the signal quality remains within an acceptable range, avoiding signal interruption or quality degradation due to switching, and improving the user experience.

[0156] Figure 2 FIG. 1 is a schematic diagram of a dynamic monitoring and early warning system for mobile phone signals according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0157] The first unit is used to collect the original signal of the mobile phone terminal, adaptively segment the original signal, divide the signal into a long-period signal segment, a medium-period signal segment and a short-period signal segment, reconstruct the signal trajectory of the long-period signal segment, establish a signal trajectory attractor model, perform cycle analysis on the medium-period signal segment, extract the signal periodic feature vector, extract the signal transient feature of the short-period signal segment, combine the signal trajectory attractor model, the signal periodic feature vector and the signal transient feature to construct an initial feature matrix, perform nonlinear dimensionality reduction to obtain a signal evolution feature space, construct a signal dynamics equation based on chaos theory and solve the corresponding Lyapunov exponent to obtain a signal system stability index, and combine the signal system stability index with the signal evolution feature space to obtain a composite feature space;

[0158] The second unit is used to partition and map the composite feature space, establish a feature distribution density map, identify the cluster center and divergence area of ​​the signal feature on the feature distribution density map, use the cluster center as the signal steady-state area, and the divergence area as the signal transition area. In the signal steady-state area, a signal normal mode library based on recursive quantization analysis is established, and in the signal transition area, a signal abnormal mode library based on phase change theory is established. A hierarchical prediction model is trained based on the signal normal mode library and the signal abnormal mode library, and the data in the composite feature space is added to the trained hierarchical prediction model to obtain a prediction output result;

[0159] The third unit is used to construct a signal quality warning decision tree based on the prediction output result, wherein the warning decision tree includes a signal attenuation warning branch, a signal interference warning branch and a signal interruption warning branch. If the signal attenuation warning branch is triggered, the signal attenuation rate is calculated and the optimal signal compensation strategy is selected based on the signal attenuation rate. If the signal interference warning branch is triggered, the interference source characteristics are analyzed and an interference avoidance plan is generated. If the signal interruption warning branch is triggered, an emergency response mechanism is initiated. According to the emergency response mechanism, the optimal signal transmission path is calculated based on the composite feature space, and the available base station resources are evaluated on the optimal signal transmission path. According to the base station resource evaluation result, a base station switching sequence is generated and a control instruction containing a warning level, a base station switching sequence and an avoidance plan is sent to the control terminal, the warning response effect is monitored and the optimal disposal decision is dynamically generated.

[0160] According to a third aspect of the embodiments of the present invention,

[0161] An electronic device is provided, comprising:

[0162] processor;

[0163] a memory for storing processor-executable instructions;

[0164] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0165] A fourth aspect of the embodiments of the present invention is:

[0166] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0167] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0168] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dynamic monitoring and early warning method for mobile phone signals, characterized in that: include: Collect the original signal of the mobile phone terminal, adaptively segment the original signal, divide the signal into a long-period signal segment, a medium-period signal segment and a short-period signal segment, reconstruct the signal trajectory of the long-period signal segment, establish a signal trajectory attractor model, perform cycle analysis on the medium-period signal segment, extract the signal periodic feature vector, extract the signal transient feature of the short-period signal segment, combine the signal trajectory attractor model, the signal periodic feature vector and the signal transient feature to construct an initial feature matrix, perform nonlinear dimensionality reduction to obtain a signal evolution feature space, construct a signal dynamics equation based on chaos theory and solve the corresponding Lyapunov exponent to obtain a signal system stability index, and combine the signal system stability index with the signal evolution feature space to obtain a composite feature space; Partition mapping is performed on the composite feature space to establish a feature distribution density map, and the cluster center and divergence area of ​​the signal feature are identified on the feature distribution density map, and the cluster center is used as the signal steady-state area, and the divergence area is used as the signal transition area. In the signal steady-state area, a signal normal mode library based on recursive quantization analysis is established, and in the signal transition area, a signal abnormal mode library based on phase change theory is established. A hierarchical prediction model is trained based on the signal normal mode library and the signal abnormal mode library, and the data in the composite feature space is added to the trained hierarchical prediction model to obtain a prediction output result; A signal quality warning decision tree is constructed based on the prediction output result, wherein the warning decision tree includes a signal attenuation warning branch, a signal interference warning branch and a signal interruption warning branch. If the signal attenuation warning branch is triggered, the signal attenuation rate is calculated and the optimal signal compensation strategy is selected based on the signal attenuation rate. If the signal interference warning branch is triggered, the interference source characteristics are analyzed and an interference avoidance plan is generated. If the signal interruption warning branch is triggered, an emergency response mechanism is initiated. According to the emergency response mechanism, an optimal signal transmission path is calculated based on the composite feature space, and available base station resources are evaluated on the optimal signal transmission path. According to the base station resource evaluation result, a base station switching sequence is generated and a control instruction containing a warning level, a base station switching sequence and an avoidance plan is sent to the control terminal, the warning response effect is monitored and the optimal disposal decision is dynamically generated.

2. The method according to claim 1, characterized in that The original signal of the mobile phone terminal is collected, and the original signal is adaptively segmented to divide the signal into a long-period signal segment, a medium-period signal segment and a short-period signal segment. The signal trajectory of the long-period signal segment is reconstructed, and a signal trajectory attractor model is established. The cyclic period analysis of the medium-period signal segment is performed to extract the signal periodic feature vector, and the signal transient feature of the short-period signal segment is extracted. The signal trajectory attractor model, the signal periodic feature vector and the signal transient feature are combined to construct an initial feature matrix, and nonlinear dimensionality reduction is performed to obtain a signal evolution feature space. Based on chaos theory, a signal dynamics equation is constructed and the corresponding Lyapunov exponent is solved to obtain a signal system stability index. The signal system stability index is combined with the signal evolution feature space to obtain a composite feature space including: Collecting the original signal of the mobile terminal, wherein the original signal includes signal strength, signal quality, signal bandwidth and signal delay, performing adaptive segmentation processing on the original signal, and dividing the original signal into a long-period signal segment, a medium-period signal segment and a short-period signal segment based on the frequency characteristics and energy characteristics of the original signal, wherein the long-period signal segment corresponds to the channel fading trend of the original signal, the medium-period signal segment corresponds to the periodic fluctuation of the original signal, and the short-period signal segment corresponds to the burst interference of the original signal; Reconstructing the signal trajectory of the long-period signal segment, mapping the long-period signal segment to a high-dimensional phase space by time delay embedding, performing cluster analysis on the signal trajectory in the high-dimensional phase space to obtain a high-dimensional cluster center, and constructing the high-dimensional cluster center as a signal trajectory attractor model; Performing cyclic period analysis on the mesocycle signal segment, calculating the autocorrelation function and power spectrum of the mesocycle signal segment through autocorrelation analysis and Fourier transform, and extracting the period, amplitude and phase of the mesocycle signal segment based on the autocorrelation function and the power spectrum to form a signal period feature vector; Performing wavelet transform and empirical mode decomposition on the short-period signal segment, extracting the energy distribution, instantaneous frequency and instantaneous amplitude of the short-period signal segment on different time scales as signal transient features, and combining the signal trajectory attractor model, the signal periodic feature vector and the signal transient features to construct an initial feature matrix; The initial feature matrix is ​​subjected to nonlinear dimensionality reduction by the kernel principal component analysis method to obtain a signal evolution feature space, a signal dynamics equation is constructed based on chaos theory, the Lyapunov exponent spectrum of the signal dynamics equation is solved to obtain a signal system stability index, and the signal system stability index is combined with the signal evolution feature space to construct a composite feature space.

3. The method according to claim 2, characterized in that The initial feature matrix is ​​subjected to nonlinear dimension reduction by the kernel principal component analysis method, and the signal evolution feature space obtained includes: Performing data center preprocessing on the initial feature matrix, where the number of rows of the initial feature matrix is ​​the number of samples and the number of columns is the feature dimension, calculating the mean vector of each feature, and subtracting the mean of the corresponding feature from each element in the initial feature matrix to obtain a centralized feature matrix; Constructing a kernel function matrix, selecting a Gaussian radial basis kernel function as the kernel function, calculating the kernel matrix based on the centralized feature matrix, wherein the number of rows and columns of the kernel matrix are equal to the number of samples, each element of the kernel matrix is ​​composed of a Gaussian radial basis kernel function value between two random sample vectors, and the bandwidth parameter of the Gaussian radial basis kernel function is selected to have an optimal value by a cross-validation method; The kernel matrix is ​​centralized, a row mean matrix and a column mean matrix are subtracted from the kernel matrix, and a global mean matrix is ​​added to obtain a centralized kernel matrix, an eigenvalue decomposition is performed on the centralized kernel matrix to obtain an eigenvalue diagonal matrix and an eigenvector matrix, an eigenvector corresponding to the maximum eigenvalue in the eigenvalue diagonal matrix is ​​selected based on the cumulative variance contribution rate, the eigenvector is normalized, and a projection matrix is ​​constructed by multiplying the normalized eigenvector with the inverse of the square root of the corresponding eigenvalue; The centered kernel matrix is ​​multiplied by the projection matrix to obtain a signal evolution feature space after dimensionality reduction, wherein the number of rows of the signal evolution feature space is equal to the number of samples, and the number of columns is equal to the feature dimension after dimensionality reduction.

4. The method according to claim 1, characterized in that: Partition mapping is performed on the composite feature space to establish a feature distribution density map, and the cluster center and divergence area of ​​the signal feature are identified on the feature distribution density map. The cluster center is used as the signal steady-state area, and the divergence area is used as the signal transition area. In the signal steady-state area, a signal normal mode library based on recursive quantization analysis is established, and in the signal transition area, a signal abnormal mode library based on phase change theory is established. A hierarchical prediction model is trained based on the signal normal mode library and the signal abnormal mode library, and the data in the composite feature space is added to the hierarchical prediction model after training, and the prediction output results obtained include: Partition mapping is performed on the composite feature space, and the composite feature space is mapped to a two-dimensional plane by a self-organizing feature mapping method to obtain a feature distribution density map, in which a cluster center is determined based on a local maximum point of data density, and a divergent region is determined based on a local minimum point of data density, wherein the cluster center corresponds to a signal steady-state region, and the divergent region corresponds to a signal transition region; A signal normal mode library is established in the signal steady-state region, and a recursive quantization analysis method is used to perform vector quantization and recursive clustering on feature samples in the signal steady-state region, and the feature samples are divided into multiple subsets according to the distance metric between the feature samples, and the mean vector of each subset is calculated to construct a quantization codebook, and the mean vector in the quantization codebook is used to replace the original feature sample for iterative clustering until the size of the quantization codebook reaches a preset threshold, and the final quantization codebook is used as the signal normal mode library; Establishing a signal anomaly pattern library in the signal transition region, using a hidden Markov model to perform state modeling on feature samples in the signal transition region, training a state transfer matrix and an emission matrix of the hidden Markov model, and constructing the signal anomaly pattern library based on the state transfer matrix and the emission matrix; A hierarchical prediction model is trained based on the signal normal mode library and the signal abnormal mode library, wherein the hierarchical prediction model includes a first prediction layer and a second prediction layer, wherein the first prediction layer uses a quantization codebook in the signal normal mode library to perform normal state discrimination on feature data in a composite feature space, and the second prediction layer uses a state transfer matrix and a transmission matrix in the signal abnormal mode library to perform abnormal mode recognition on feature data discriminated as abnormal by the first prediction layer; The feature data in the composite feature space is input into the hierarchical prediction model to obtain a signal state discrimination result and an abnormal pattern recognition result and combine them into a prediction output result.

5. The method according to claim 4, characterized in that The recursive quantization analysis method is used to perform vector quantization on the characteristic samples in the steady-state region of the signal as shown in the following formula: ； Where D(X, C) represents the feature samples after vector quantization, X represents the set of all feature samples, C represents the codebook set used in vector quantization, K represents the number of vectors in the codebook, N represents the number of feature samples, i represents the feature index, L represents the number of feature dimensions, k represents the feature dimension index, and w represents the feature dimension index. k represents the weight factor on dimension k, GP k (x i ) represents the sample x i The result after Gaussian process mapping at scale k, c j is the codebook vector, representing the jth codeword in the codebook, λ represents the balancing weight parameter, Represents sample x i The value after Gaussian process mapping and the codebook vector c j The distance between.

6. The method according to claim 1, characterized in that A signal quality warning decision tree is constructed based on the prediction output result, wherein the warning decision tree includes a signal attenuation warning branch, a signal interference warning branch and a signal interruption warning branch. If the signal attenuation warning branch is triggered, the signal attenuation rate is calculated and the optimal signal compensation strategy is selected based on the signal attenuation rate. If the signal interference warning branch is triggered, the interference source characteristics are analyzed and an interference avoidance plan is generated. If the signal interruption warning branch is triggered, an emergency response mechanism is started. According to the emergency response mechanism, the optimal signal transmission path is calculated based on the composite feature space, and the available base station resources are evaluated on the optimal signal transmission path. According to the base station resource evaluation result, a base station switching sequence is generated and a control instruction containing a warning level, a base station switching sequence and an avoidance plan is sent to the control terminal. Monitoring the warning response effect and dynamically generating the optimal disposal decision include: Based on the prediction output result, a signal quality warning decision tree including a signal attenuation warning branch, a signal interference warning branch and a signal interruption warning branch is constructed; Collect signal strength time series data corresponding to the original signal, perform first-order difference calculation on the signal strength time series data to obtain an attenuation rate, determine whether to trigger the signal attenuation warning branch based on the attenuation rate, and if triggered, compare the attenuation rate with a preset attenuation threshold. If the attenuation rate is greater than the attenuation threshold, switch the signal transmission channel; if the attenuation rate is less than the attenuation threshold, increase the signal transmission power; Collect interference signal data, determine whether the interference warning branch is triggered based on the interference signal data, and if triggered, extract spectrum parameters, amplitude parameters and modulation parameters of the interference signal data, determine frequency band switching parameters and modulation scheme parameters based on the spectrum parameters, amplitude parameters and modulation parameters, switch the signal to the target frequency band according to the frequency band switching parameters, and adjust the signal modulation mode according to the modulation scheme parameters; Based on the prediction output result, it is queried whether there is a preset external interruption event. If so, the signal interruption warning branch is triggered, the emergency response mechanism is started and the signal data of multiple alternative transmission paths are collected, and the signal strength index, signal-to-noise ratio index and transmission delay index of each alternative transmission path are calculated. The signal strength index, the signal-to-noise ratio index and the transmission delay index are input into a preset scoring model, and the quality score of each alternative transmission path is calculated in combination with the composite feature space, and the alternative transmission path with the highest quality score is selected as the optimal signal transmission path; Detecting the number of channel resources, the number of time slot resources and the power resource capacity of each base station on the optimal signal transmission path, generating a base station switching sequence according to the number of channel resources, the number of time slot resources and the power resource capacity, sending the base station switching sequence to each base station, and controlling the signal to switch between the base stations according to the base station switching sequence; Collect signal quality data after switching, calculate the quality index of the signal quality data, compare the quality index with the preset signal quality threshold, if the quality index is less than the signal quality threshold, recalculate the quality score of the alternative transmission path and update the base station switching sequence, monitor the early warning response effect based on the updated base station switching sequence and dynamically generate the optimal disposal decision according to the early warning response effect.

7. The method according to claim 6, characterized in that Detecting the number of channel resources, the number of time slot resources, and the power resource capacity of each base station on the optimal signal transmission path, generating a base station switching sequence according to the number of channel resources, the number of time slot resources, and the power resource capacity, sending the base station switching sequence to each base station, and controlling the signal to switch between the base stations according to the base station switching sequence includes: Establishing a reinforcement learning model by analyzing historical switching data, the reinforcement learning model is used to learn the characteristic pattern of the optimal switching path, and generating an initial base station switching sequence based on the reinforcement learning model, the initial base station switching sequence is used to store the base stations passed in sequence during the switching process; An evolutionary algorithm is used to optimize the weight coefficient of the switching cost function, wherein the switching cost function includes a resource utilization weight, a switching overhead weight, and a signal quality weight, and an optimal weight combination is searched through a crossover operation, a mutation operation, and a selection operation; Constructing a fuzzy rule base of resource constraint conditions, using a fuzzy reasoning mechanism to evaluate the switching feasibility of each base station, and selecting feasible base stations that meet the resource constraints from all base stations on the optimal transmission path based on the fuzzy reasoning mechanism to form a subset of feasible base stations; A swarm intelligence optimization algorithm is used to generate a base station switching sequence. The swarm intelligence optimization algorithm includes multiple agents. The agents exchange information through a pheromone mechanism and a speed update mechanism, explore the optimal switching path, and establish a deep reinforcement learning model. The deep reinforcement learning model optimizes the time slot configuration strategy through continuous interaction with the network environment and adjusts the time slot length and time slot interval according to the feedback reward signal of the switching performance; A machine learning algorithm is used to predict the resource occupancy trend of each base station, and when a risk of resource overlimit is predicted, the base station switching sequence and resource allocation scheme are adjusted; Determine whether the feasible base station subset is empty. If the feasible base station subset is empty, match historical switching failure cases based on the case reasoning mechanism to generate exception handling suggestions. If the feasible base station subset is not empty, send the base station switching sequence to each base station, and control the signal to switch between the base stations according to the base station switching sequence.

8. A dynamic monitoring and early warning system for mobile phone signals, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used to collect the original signal of the mobile phone terminal, adaptively segment the original signal, divide the signal into a long-period signal segment, a medium-period signal segment and a short-period signal segment, reconstruct the signal trajectory of the long-period signal segment, establish a signal trajectory attractor model, perform cycle analysis on the medium-period signal segment, extract the signal periodic feature vector, extract the signal transient feature of the short-period signal segment, combine the signal trajectory attractor model, the signal periodic feature vector and the signal transient feature to construct an initial feature matrix, perform nonlinear dimensionality reduction to obtain a signal evolution feature space, construct a signal dynamics equation based on chaos theory and solve the corresponding Lyapunov exponent to obtain a signal system stability index, and combine the signal system stability index with the signal evolution feature space to obtain a composite feature space; The second unit is used to partition and map the composite feature space, establish a feature distribution density map, identify the cluster center and divergence area of ​​the signal feature on the feature distribution density map, use the cluster center as the signal steady-state area, and the divergence area as the signal transition area. In the signal steady-state area, a signal normal mode library based on recursive quantization analysis is established, and in the signal transition area, a signal abnormal mode library based on phase change theory is established. A hierarchical prediction model is trained based on the signal normal mode library and the signal abnormal mode library, and the data in the composite feature space is added to the trained hierarchical prediction model to obtain a prediction output result; The third unit is used to construct a signal quality warning decision tree based on the prediction output result, wherein the warning decision tree includes a signal attenuation warning branch, a signal interference warning branch and a signal interruption warning branch. If the signal attenuation warning branch is triggered, the signal attenuation rate is calculated and the optimal signal compensation strategy is selected based on the signal attenuation rate. If the signal interference warning branch is triggered, the interference source characteristics are analyzed and an interference avoidance plan is generated. If the signal interruption warning branch is triggered, an emergency response mechanism is initiated. According to the emergency response mechanism, the optimal signal transmission path is calculated based on the composite feature space, and the available base station resources are evaluated on the optimal signal transmission path. According to the base station resource evaluation result, a base station switching sequence is generated and a control instruction containing a warning level, a base station switching sequence and an avoidance plan is sent to the control terminal, the warning response effect is monitored and the optimal disposal decision is dynamically generated.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images