Wireless repeater multi-band signal intensity prediction system based on artificial intelligence

Abstract

The invention relates to the technical field of wireless communication, in particular to a wireless repeater multi-band signal strength prediction system based on artificial intelligence, which comprises a multi-band signal acquisition and spatial decomposition module, a spatial-temporal feature extraction module, a channel modeling module and a multi-band signal strength prediction module. According to the method, environmental parameters such as multi-band original signals and air temperature of the wireless repeater are obtained in real time, a detailed feature matrix is generated through spatial signal processing, and a rich and accurate data source is provided for accurate prediction; an LSTM time sequence model is constructed to extract time features, a graph convolution model is used to extract spatial features, the time features and the spatial features are fused, an attention weight vector is optimized, and prediction precision is improved; and multi-band channel modeling is carried out, a path loss compensation model is constructed to correct a prediction result, and powerful support is provided for wireless repeater signal optimization management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless communication, in particular to a wireless repeater multi-band signal strength prediction system based on artificial intelligence. BACKGROUND

[0002] In view of the above problems, it is necessary to propose a wireless repeater multi-band signal strength prediction system based on artificial intelligence.

[0003] With the continuous evolution of 5G and even future 6G communication technology, wireless communication networks have been widely deployed and applied around the world. From traditional voice communication to today's high-speed data transmission, Internet of Things, intelligent transportation and other diversified services, wireless communication has become the core infrastructure of modern information interaction. In order to meet the growing demand for services, the scale of wireless communication networks continues to expand, the density of base stations continues to increase, and the signal propagation environment has become increasingly complex, including the multi-path effect caused by urban high-rise buildings, the interference of different frequency bands, and the dynamic changes of user distribution. However, the existing signal strength prediction system has many shortcomings.

[0004] The existing signal strength prediction theoretical model is usually based on some idealized assumptions, such as free space propagation, uniform medium, etc. In actual complex wireless communication environment, these assumptions are often difficult to establish, resulting in a large deviation between the prediction results and the actual situation. Although the empirical model considers the actual environmental factors, it is usually modeled for a specific area or specific scenario, and when the application environment changes, the accuracy of the model will decrease significantly, and a large amount of field measurement and model correction work needs to be done again.

[0005] Therefore, it is necessary to propose a wireless repeater multi-band signal strength prediction system based on artificial intelligence to solve the above problems. SUMMARY

[0006] The purpose of the present application is to solve the problems in the background art and propose a wireless repeater multi-band signal strength prediction system based on artificial intelligence.

[0007] The purpose of the present application can be achieved by the following technical solutions:

[0008] A wireless repeater multi-band signal strength prediction system based on artificial intelligence, comprising a multi-band signal acquisition and spatial decomposition module, a space-time feature extraction module, a channel modeling module and a multi-band signal strength prediction module.

[0009] The multi-band signal acquisition and spatial decomposition module acquires multi-band original signal data and environmental parameters in the coverage area of the wireless repeater in real time. The multi-band antenna array is accessed to acquire the center frequency, signal amplification and phase offset of the wireless repeater at each frequency band. Real-time environmental parameters are acquired through environmental sensors, including air temperature, air humidity and environmental electromagnetic noise signals.

[0010] The multi-band original signal data is acquired, and the specific process is as follows:

[0011] The multi-band original signal in the coverage area of the wireless repeater is acquired through the frequency band antenna array, and the center frequency fi of each frequency band i is acquired. Wherein i is the frequency band number, i = 1, 2,..., n; n is the total number of frequency bands.

[0012] As a preferred mode of the present application, the acquired multi-band original signal is subjected to spatial signal processing, the superimposed original signal is decomposed into multiple arrival paths, and the amplitude and phase of each path are decoupled to obtain the signal amplitude Ak(t) and phase offset of each path Wherein k is the path number, k = 1, 2,..., K, K is the total number of paths; the path delay τk(t) of each arrival path k with respect to each center frequency fi and the arrival angle θk(t) of each arrival path k are acquired.

[0013] The environmental data in the coverage area of the wireless repeater is acquired, including the air temperature T(xj, tj, zj), air humidity H(xj, tj, zj) and real-time environmental electromagnetic noise signal η(xj, tj, zj) of each spatial point (xj, tj, zj), wherein (xj, tj, zj) is the monitoring point coordinate, j is the environmental data monitoring point sequence number, j = 1, 2,..., m; m is the total number of environmental data monitoring points.

[0014] The frequency domain data of the original signal with respect to each center frequency is generated according to the center frequency of each frequency band, the signal amplitude and phase offset of each path:

[0015]

[0016] Wherein j is the imaginary unit.

[0017] The real-time environmental parameters are acquired, and the original signal data and real-time environmental parameters are subjected to data standardization processing.

[0018] The path characteristic vector containing signal amplitude, phase offset, center frequency, path delay and arrival angle is generated:

[0019]

[0020] The environmental characteristic matrix containing air temperature, air humidity and electromagnetic noise signal is generated:

[0021]

[0022] The path feature vector and the environment feature matrix are sent to the space-time feature extraction module as source data for time feature and space feature extraction.

[0023] The space-time feature extraction module extracts the time feature and the space feature of signal propagation according to the time sequence data of the original signal, and constructs a dynamic channel model. The time feature is extracted through an LSTM model, and the space feature is extracted through graph convolution operation.

[0024] The time feature is extracted through an LSTM model, and the specific process is as follows:

[0025] The LSTM time sequence model is constructed, including:

[0026] The forget gate:

[0027]

[0028] wherein W f，k and b f，k are the weight matrix and the bias matrix of the forget gate, and σ is the sigmoid activation function; wherein Vk(t) is the path feature vector, and ε(t) is the environment feature matrix;

[0029] The input gate:

[0030]

[0031] wherein W i，k and b i，k are the weight matrix and the bias matrix of the input gate;

[0032] The candidate state:

[0033]

[0034] wherein tanh is the hyperbolic sine activation function, and wherein W c，k and b c，k are the weight matrix and the bias matrix of the candidate state, respectively;

[0035] The cell state:

[0036]

[0037] wherein is the Hadmard convolution operation; C t-1 is the cell state at the previous time;

[0038] The output gate:

[0039]

[0040] where W O，k and b O，k are the weight matrix and bias matrix of the output gate;

[0041] Hidden state:

[0042]

[0043] The path feature vector and the environment feature matrix are input into the LSTM time sequence model, and the final output hidden state h t is recorded. t Specifically, it is an n-dimensional vector, and the elements represent the signal intensity of each frequency band i.

[0044] The spatial features are extracted by graph convolution operation, and the specific process is as follows:

[0045] The adjacency matrix in the environment feature matrix is extracted:

[0046] The clustering eigenvalue A j1j2 of the elements p j1m and p j2m in the mth column of the j1th row and the j2th row is calculated by the formula:

[0047]

[0048] j1j2 . Where m is the column number in the environment feature matrix, m = 1, 2, 3; j is the row number in the environment feature matrix. Where is the standard deviation of the mth column element.

[0049] A graph convolution model is constructed, and the output formula of each layer l is:

[0050]

[0051] Where σ is an activation function, and is an added term, that is, a set composed of the path feature vector and the environment feature matrix. Where is the degree matrix, where U (l) is the node feature of the lth layer; where w (l) is the weight factor of the lth layer in the graph convolution model, and U (l+1 ) is the output of the l+1th layer in the graph convolution model.

[0052] The path feature vector and the environment feature matrix are input into the graph convolution model, and the output value U of the last layer is obtained, U is specifically an n-dimensional vector, and the elements represent the signal intensity of each frequency band i.

[0053] The hidden state h tThe U output by the graph convolution model is sent to the channel modeling module as source data for fusion of time characteristics and space characteristics.

[0054] The channel modeling module acquires signal strength, multipath delay, Doppler frequency shift and signal attenuation factor, performs multi-band channel modeling, performs fusion of time characteristics and space characteristics data, calculates theoretical attenuation based on a modified Friis formula, constructs a path loss compensation model, simulates attenuation of multi-band original signals along a propagation path, and optimizes a signal strength prediction result.

[0055] The hidden state h t The output value U of the last layer of the graph convolution model is cross-modality feature spliced, an attention weight vector W i = (W 1, W 2,..., W n ) is arranged, the attention weight elements W 1, W 2,..., W n in the attention weight vector respectively represent attention to the hidden state h t The n th element in the hidden state h

[0056] Through the out-of-control attention optimization model:

[0057]

[0058] Solving the condition that the attention weight element combination is maximum when the attention parameter W i is maximum,

[0059] The elements in the attention weight vector are optimized to improve prediction accuracy.

[0060] Where q T Is a preset attention query vector, representing the influence of adjacent elements on the element, and Z k is a preset attention weight matrix, and the elements in the matrix represent the error value of the actual signal strength in the i th frequency band with respect to the corresponding element in the hidden state h t And the output value U of the last layer of the graph convolution model.

[0061] As a preferred mode of the present application, a multi-path loss compensation model is constructed based on a modified Friis formula to calculate theoretical attenuation, simulate attenuation of multi-band original signals along each propagation path, and further correct the signal strength prediction result, and the specific process is as follows:

[0062] Through the spatial free loss

[0063] The spatial loss L k (i) from the signal source (x 0, y 0, z 0 ) to any point (x, y, z) in space is calculated, where wherein C is the speed of light, dk is the distance from the signal source (x0, y0, z0) to any point (x, y, z) in space.

[0064] wherein e α×T(x，y，z) +e β×H(x，y，z) is the signal attenuation caused by temperature and humidity, wherein a and b are preset atmospheric attenuation coefficients and water vapor absorption attenuation coefficients, respectively;

[0065] wherein Gk(0k(t), f) is the incident angle reflection loss attenuation, d1 and d2 are preset incident angle loss and frequency attenuation coefficients, respectively.

[0066] The multi-band signal strength prediction module combines the time characteristics, the space characteristics, the attention weight vector and the multi-path loss compensation model to complete the final signal prediction, corrects the signal strength prediction, and predicts the signal strength distribution of each frequency band of each space point in a preset future time.

[0067] The time characteristics are extracted, i.e., the hidden state h t output by the LSTM time sequence model t ={hi}={h1, h2,..., hm};

[0068] The space characteristics are extracted, i.e., the output U of the lth layer of the graph convolution model

[0069] The time characteristics, the space characteristics and the attention weight vector are data fused through a preset formula to obtain the preliminary prediction result of the signal strength in each frequency band i

[0070] As a preferred mode of the present application, the preliminary prediction result of the signal strength in each frequency band i and the multi-path loss compensation model are data fused through a preset formula to obtain the final signal strength prediction value of each space point (x, y, z)

[0071] Compared with the prior art, the present application has the following beneficial effects:

[0072] 1. The present application acquires the multi-band original signal data and the environmental parameters in the coverage area of the wireless repeater in real time. The real-time environmental parameters such as air temperature, humidity and environmental electromagnetic noise signals are acquired by using the environmental sensor, and are subjected to space signal processing to decompose the superimposed signals, decouple the amplitude and the phase, acquire the path delay and the angle of arrival, generate a detailed path characteristic vector and an environmental characteristic matrix, and provide rich and accurate data sources for subsequent accurate prediction.

[0073] 2、The application extracts the time and space characteristics of signal propagation. By constructing an LSTM time sequence model, the time characteristics are effectively extracted by using structures such as a forgetting gate, an input gate, a candidate state, a cell state and an output gate; and by graph convolution operation, the adjacent matrix in the environmental feature matrix is extracted, and a graph convolution model is constructed to extract the space characteristics. Then the output results of the two are sent to the channel modeling module for data fusion, and the attention weight vector is arranged and optimized to improve the prediction accuracy, so that the system can more accurately grasp the space-time characteristics of signal propagation.

[0074] 3、The application obtains multiple signal parameters for multi-band channel modeling, and after data fusion of the time characteristics and the space characteristics, the theoretical attenuation is calculated based on the modified Friis formula, a path loss compensation model is constructed, and the attenuation of the original signal along the propagation path is simulated. The multi-band signal strength prediction module combines the time characteristics, the space characteristics, the attention weight vector and the multi-path loss compensation model to complete the final signal prediction, and the prediction result is further corrected, which provides strong support for the signal optimization and management of the wireless repeater. BRIEF DESCRIPTION OF DRAWINGS

[0075] In order to facilitate those skilled in the art to understand, the application will be further described below in conjunction with the drawings:

[0076] Figure 1 The system block diagram of the application. DETAILED DESCRIPTION

[0077] The technical solutions of the application will be described below in conjunction with the embodiments, obviously, the described embodiments are only a part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor belong to the protection scope of the application.

[0078] Please refer to Figure 1 As shown in the figure, a wireless repeater multi-band signal strength prediction system based on artificial intelligence includes a multi-band signal acquisition and spatial decomposition module, a space-time feature extraction module, a channel modeling module and a multi-band signal strength prediction module.

[0079] The multi-band signal acquisition and spatial decomposition module acquires multi-band original signal data and environmental parameters in the coverage area of the wireless repeater in real time. Access the multi-band antenna array, collect the center frequency, signal amplification and phase shift of the wireless repeater at each frequency band. Real-time environmental parameters are obtained through environmental sensors, including air temperature, air humidity and environmental electromagnetic noise signals.

[0080] The multi-band original signal data is collected, and the specific process is as follows:

[0081] The multi-band original signals in the coverage area of the wireless repeater are acquired by the frequency band antenna array, and the center frequency fi of each frequency band i is acquired. Wherein i is the frequency band number, i = 1, 2,..., n; n is the total number of frequency bands.

[0082] Further, the spatial signal processing is performed on the acquired multi-band original signals, the superimposed original signals are decomposed into multiple arrival paths, and the amplitude and phase decoupling is performed on each path to obtain the signal amplitude Ak(t) and phase offset Wherein k is the path number, k = 1, 2,..., K, K is the total number of paths; the path delay τk(t) of each arrival path k with respect to each center frequency fi and the arrival angle θk(t) of each arrival path k are acquired.

[0083] The environmental data in the coverage area of the wireless repeater is acquired, including the air temperature T(xj, tj, zj), air humidity H(xj, tj, zj) and real-time environmental electromagnetic noise signal η(xj, tj, zj) of each spatial point (xj, tj, zj), wherein (xj, tj, zj) is the monitoring point coordinate, j is the environmental data monitoring point sequence number, j = 1, 2,..., m; m is the total number of environmental data monitoring points.

[0084] The frequency domain data of the original signal with respect to each center frequency is generated according to the center frequency of each frequency band, the signal amplitude and phase offset of each path:

[0085]

[0086] Wherein j is the imaginary unit.

[0087] The real-time environmental parameters are acquired, and the original signal data and the real-time environmental parameters are subjected to data standardization processing.

[0088] The path feature vector containing the signal amplitude, phase offset, center frequency, path delay and arrival angle is generated:

[0089]

[0090] The environmental feature matrix containing the air temperature, air humidity and electromagnetic noise signal is generated:

[0091]

[0092] The path feature vector and the environmental feature matrix are sent to the space-time feature extraction module as the source data for time feature and space feature extraction.

[0093] The space-time feature extraction module extracts the time feature and the space feature of signal propagation according to the time sequence data of the original signal, and constructs a dynamic channel model. The time feature is extracted through an LSTM model, and the space feature is extracted through graph convolution operation.

[0094] The time feature is extracted through an LSTM model, and the specific process is as follows:

[0095] The LSTM time sequence model is constructed, including:

[0096] The forget gate is:

[0097]

[0098] wherein W f，k and b f，k are the weight matrix and the bias matrix of the forget gate, and sigma is a sigmoid activation function; wherein Vk(t) is a path feature vector, and wherein e(t) is an environment feature matrix;

[0099] The input gate is:

[0100]

[0101] wherein W i，k and b i，k are the weight matrix and the bias matrix of the input gate;

[0102] The candidate state is:

[0103]

[0104] wherein tanh is a hyperbolic sine activation function, wherein W c，k and b c，k are the weight matrix and the bias matrix of the candidate state, respectively;

[0105] The cell state is:

[0106]

[0107] wherein is a Hadmard convolution operation; C t-1 is the cell state at the last time;

[0108] The output gate is:

[0109]

[0110] wherein W O，k and b O，k are the weight matrix and the bias matrix of the output gate;

[0111] The hidden state is:

[0112]

[0113] The path feature vector and the environment feature matrix are input into the LSTM time sequence model, and the final output hidden state h is recorded t The hidden state h t Specifically, an n-dimensional vector, where the elements represent the signal intensity of each frequency band i.

[0114] The spatial features are extracted by graph convolution operation, and the specific process is as follows:

[0115] The adjacency matrix in the environment feature matrix is extracted:

[0116] The clustering eigenvalue A of the elements p

[0117]

[0118] The clustering eigenvalue A of the elements p j1m and p j2m is calculated. j1j2 Where m is the column number in the environment feature matrix, m = 1, 2, 3; j is the row number in the environment feature matrix. Where is the standard deviation of the mth column element.

[0119] The graph convolution model is constructed, and the output formula of each layer l is:

[0120]

[0121] Where σ is the activation function, and is the added term, that is, the set composed of the path feature vector and the environment feature matrix. Where is the degree matrix, Where U (l ) is the node feature of the lth layer; where w (l) is the weight factor of the lth layer in the graph convolution model, and U (l+1) is the output of the l+1th layer in the graph convolution model.

[0122] The path feature vector and the environment feature matrix are input into the graph convolution model, and the output value U of the last layer is obtained, U is an n-dimensional vector, where the elements represent the signal intensity of each frequency band i.

[0123] The hidden state h output by the LSTM time sequence model t and the U output by the graph convolution model are sent to the channel modeling module as the source data for time feature and spatial feature data fusion and fusion.

[0124] ​The channel modeling module obtains signal strength, multipath delay, Doppler shift and signal attenuation factor, performs multi-band channel modeling, performs time characteristic and space characteristic data fusion, calculates theoretical attenuation based on a modified Friis formula, constructs a path loss compensation model, simulates the attenuation of multi-band original signals along the propagation path, and optimizes the signal strength prediction result.

[0125] The hidden state h output by the LSTM time series model t The output value U of the last layer of the graph convolution model is cross-modal feature spliced, and an attention weight vector Wi=(W1, W2,..., Wn) is arranged, wherein the attention weight elements W1, W2,..., Wn in the attention weight vector respectively represent attention to the hidden state h t The nth element in the hidden state h

[0126] Through the out-of-control attention optimization model:

[0127]

[0128] Solving the condition that the attention weight element combination of the attention parameter Wi is maximum,

[0129] The elements in the attention weight vector are optimized to improve the prediction accuracy.

[0130] Where q T is a preset attention query vector, representing the influence of adjacent elements on the element, wherein Zk is a preset attention weight matrix, and the elements represent the error value of the actual signal strength in the ith frequency band to the hidden state h t and the corresponding element in the output value U of the last layer of the graph convolution model.

[0131] Further, based on the modified Friis formula to calculate the theoretical attenuation, a multi-path loss compensation model is constructed to simulate the attenuation of multi-band original signals along each propagation path, and the signal strength prediction result is further corrected, and the specific process is as follows:

[0132] Through the spatial free loss

[0133] Calculate the spatial loss Lk(i) from the signal source (x0, y0, z0) to any point (x, y, z) in space, wherein is the path loss, wherein C is the speed of light, and dk is the distance from the signal source (x0, y0, z0) to any point (x, y, z) in space.

[0134] Where e α×T(x，y，z) +e β×H(x，y，z)wherein a and b are preset atmospheric attenuation coefficient and water vapor absorption attenuation coefficient respectively;

[0135] wherein Gk(0k(t), fi) is the incidence angle reflection loss attenuation, d1 and d2 are preset incidence angle loss and frequency attenuation coefficient respectively.

[0136] The multi-band signal strength prediction module combines the time feature, the space feature, the attention weight vector and the multi-path loss compensation model to complete the final signal prediction, corrects the signal strength prediction, and predicts the signal strength distribution of each frequency band at each spatial point in the future preset time.

[0137] The time feature is extracted, i.e. the hidden state h of the LSTM time sequence model final output t , h t ={hi}={h1, h2,..., hm};

[0138] The space feature is extracted, i.e. the output U of the lth layer of the graph convolution model U={Ui}={U1, U2,..., Un};

[0139] The time feature, the space feature and the attention weight vector are data fused through a preset formula to obtain the preliminary prediction result of the signal strength in each frequency band i

[0140] Further, the preliminary prediction result of the signal strength in each frequency band i and the multi-path loss compensation model are data fused through a preset formula to obtain the final signal strength prediction value of each spatial point (x, y, z)

[0141] It should be understood that the terms "comprise" and "comprising" used in the specification and claims of the present disclosure indicate the presence of the described features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0142] It should also be understood that the terms used in the specification of the present disclosure are only for the purpose of describing specific embodiments and are not intended to limit the present disclosure. As used in the specification and claims of the present disclosure, the singular forms "a", "an" and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should be further understood that the term "and / or" used in the specification and claims of the present disclosure means any combination of one or more of the associated listed items and all possible combinations thereof;

[0143] The preferred embodiments of the application disclosed above are only to facilitate the elucidation of the application. The preferred embodiments do not describe all the details of the application and limit the application to the specific embodiments. Obviously, many modifications and variations can be made in light of the teachings above. The description is chosen and described in order to provide the best illustration of the application and its practical application to those skilled in the art and to enable those skilled in the art to best utilize the application. The application is limited only by the claims and their full scope and equivalents.

Claims

1. An artificial intelligence-based wireless repeater multi-band signal strength prediction system, comprising a multi-band signal acquisition and spatial decomposition module, a spatiotemporal feature extraction module, and a channel modeling module, characterized by: The multi-band signal acquisition and spatial decomposition module acquires multi-band raw signal data and environmental parameters in the wireless repeater coverage area in real time; accesses the multi-band antenna array to collect multi-band raw signal data, and collects the center frequency, signal gain, and phase offset of the wireless repeater in each frequency band; and obtains real-time environmental parameters, including air temperature, air humidity, and environmental electromagnetic noise signals, through environmental sensors; The spatiotemporal feature extraction module extracts the temporal and spatial features of signal propagation based on the time series data of the original signal and constructs a dynamic channel model. It extracts temporal features through the LSTM model and spatial features through graph convolution operations. The channel modeling module obtains signal strength, multipath delay, Doppler shift and signal attenuation factor to perform multi-band channel modeling; The temporal and spatial feature data are fused, and the theoretical attenuation is calculated based on the modified Friis formula. A path loss compensation model is constructed to simulate the attenuation of multi-band original signals along the propagation path and optimize the signal strength prediction results.

2. The artificial intelligence-based wireless repeater multi-band signal strength prediction system according to claim 1, characterized in that: It also includes a multi-band signal strength prediction module; The multi-band signal strength prediction module combines temporal features, spatial features, attention weight vectors and multi-path loss compensation models to complete the final signal prediction, correct the signal strength prediction, and predict the signal strength distribution of each frequency band at each spatial point within a preset time in the future.

3. The artificial intelligence-based wireless repeater multi-band signal strength prediction system according to claim 1, characterized in that: The specific process of collecting multi-band raw signal data is as follows: The multi-band original signal within the wireless repeater coverage area is acquired through the frequency band antenna array to obtain the center frequency of each frequency band; spatial signal processing is performed on the acquired multi-band original signal to decompose the superimposed original signal into multiple arrival paths, and amplitude and phase decoupling is performed on each path to obtain the signal amplitude and phase offset of each path; the path delay of each arrival path with respect to each center frequency and the arrival angle of each arrival path are obtained; Acquire real-time environmental parameters and perform data standardization on the original signal data and real-time environmental parameters.

4. The artificial intelligence-based wireless repeater multi-band signal strength prediction system according to claim 3, characterized in that: The specific process of obtaining real-time environmental parameters and standardization is as follows: Obtain the air temperature, air humidity and real-time environmental electromagnetic noise signals at each spatial point, where is the coordinate of the monitoring point; Generate the frequency domain data of the original signal about each center frequency based on the center frequency of each frequency band, the signal amplitude of each path, and the phase offset: Generate a path characteristic vector containing signal amplitude, phase offset, center frequency, path delay, and arrival angle; Generate an environmental feature matrix containing air temperature, air humidity and electromagnetic noise signals; The path feature vector and the environment feature matrix are sent to the spatiotemporal feature extraction module as the source data for temporal feature and spatial feature extraction.

5. The artificial intelligence-based wireless repeater multi-band signal strength prediction system according to claim 1, characterized in that: The specific process of extracting time features through the LSTM model is: Build an LSTM time series model, including: forget gate, input gate, candidate state, cell state, output gate and hidden state; Among them, the inputs of the forget gate, input gate, candidate state and output gate are the path feature vector and the environment feature matrix; The path feature vector and the environment feature matrix are input into the LSTM time series model, and the final output of the hidden state is recorded, which is specifically a multidimensional vector whose elements represent the signal strength of each frequency band.

6. The artificial intelligence-based wireless repeater multi-band signal strength prediction system according to claim 1, characterized in that: The specific process of extracting spatial features through graph convolution operation is as follows: Extract the adjacency matrix from the environmental feature matrix, calculate the difference between the square of the difference between any two elements in the same column but different rows and twice the variance of all elements in the column where the two elements are located, obtain result one, calculate the specific value of the exponential function of result one with the natural exponential as the base, and obtain the clustering eigenvalue of the elements in each row; Construct a graph convolution model, and let the calculation process of each layer be: The output of the previous layer is multiplied by the preset weight factor of this layer to obtain the result 1; The matrix penalty result of the set consisting of the preset degree matrix, the path feature vector and the environment feature matrix is ​​recorded as result 2; the product of result 1 and result 2 is input into the preset activation function to obtain the output of this layer, which is used as the input of the next layer; The path feature vector and the environment feature matrix are input into the graph convolution model to obtain the output value of the last layer. The output value of the last layer is specifically a multidimensional vector, the elements of which represent the signal strength of each frequency band.

7. The artificial intelligence-based wireless repeater multi-band signal strength prediction system according to claim 1, characterized in that: The specific process of optimizing the signal strength prediction results is as follows: Perform cross-modal feature concatenation on the hidden state output by the LSTM time series model and the output value of the last layer of the graph convolutional model to arrange the attention weight vector; the elements in the attention weight vector represent the attention to the elements represented by the corresponding positions in the hidden state output by the time series model, and the value range is 0 to 1; Preset the attention weight matrix, where the elements represent the error values ​​of the corresponding elements in the hidden state output of the LSTM time series model and the output value of the last layer of the graph convolution model for the actual signal strength in the frequency band; use the attention weight matrix as a known parameter to optimize the elements in the attention weight vector, and solve the elements in the attention weight vector to meet the condition: the combination of attention weight elements when the attention parameter is maximized to improve the prediction accuracy; The theoretical attenuation is calculated based on the modified Friis formula, and a multipath loss compensation model is constructed to simulate the attenuation of multi-band original signals along various propagation paths. The signal strength prediction results are further corrected to calculate the spatial loss from the signal source to any point in space.

8. The artificial intelligence-based wireless repeater multi-band signal strength prediction system according to claim 2, characterized in that: The specific process of correcting the signal strength prediction is as follows: Extract time features, that is, the hidden state of the final output of the LSTM timing model; Extract spatial features, which are the output of the graph convolutional model; The temporal features, spatial features and attention weight vectors are fused using a preset formula to obtain preliminary prediction results of the signal strength in each frequency band and the final signal strength prediction value of each spatial point.

Images