Method, system and medium for base station positioning work based on complex environmental factors of GIS

Abstract

The application provides a system and method for base station positioning based on GIS complex environmental factors. The system integrates multi-source data and applies deep learning models to propose an intelligent and automated base station site selection and network optimization solution. First, multi-source data including geographic information, environmental characteristics and signal propagation characteristics are collected to construct a feature matrix. Through data preprocessing, the feature matrix is normalized. Then, a deep learning model is constructed using a convolutional neural network to automatically extract spatial features and predict the suitability of base station site selection and signal coverage. After that, GIS tools are used for visual analysis of base station site selection and signal coverage. Through this method, intelligent site selection of base station location, effective prediction and optimization of signal coverage, and rational scheduling of network resources can be achieved, especially suitable for network planning and base station deployment in areas with complex terrain and significant environmental factors.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of communication technology, in particular to a GIS-based base station positioning method and system for complex environmental factors and a medium. BACKGROUND

[0002] The mountainous environment collected by GIS needs to ensure the signal transmission of the propagation model in the 5G environment. In order to build more value and meet the evaluation accuracy of signal coverage in complex scenes, the point-grid discontinuous site selection problem in the previous base station planning needs to be expanded to a three-dimensional-continuous site selection problem that is more in line with the actual construction situation, taking into account various complex scenes such as mountains and hills, optimizing the early construction planning, and providing scientific basis and guidelines for the construction of 5G base stations in the future.

[0003] The point-grid two-dimensional site selection algorithm mainly deals with two-dimensional planar graphs. In this algorithm, the study area is considered as a two-dimensional plane, which is divided by regular grids. Each grid point represents a potential site selection point in the study area, and the algorithm determines the best site selection location by evaluating specific conditions or indicators on these points.

[0004] The processing method of two-dimensional planar graph is suitable for most ground projects and some basic spatial analysis tasks, such as urban planning, facility layout, and service area optimization. It simplifies complex three-dimensional spatial problems, making the implementation and calculation of the algorithm more intuitive and efficient. However, this simplification also means that the algorithm is difficult to directly handle problems related to height or depth, such as floor selection of high-rise buildings, layout of underground facilities, unless these problems are projected or converted into two-dimensional planar problems in some way. This method cannot consider the signal propagation in three-dimensional geographic space (x, y, z) and the impact of actual buildings, terrain, vegetation, and other real ground environments on the signal. It is not sufficient to simulate and emulate the actual coverage of base station signals. At the same time, the lack of three-dimensional data also makes it difficult to simulate and predict signal coverage.

[0005] For site selection problems that need to consider height or depth information, three-dimensional spatial analysis methods or height information as additional parameters need to be included in two-dimensional analysis to provide more comprehensive spatial solutions. This requires technical personnel in the field to solve the corresponding technical problems. SUMMARY

[0006] The present application aims to at least solve the technical problems existing in the prior art, and particularly innovatively proposes a GIS-based base station positioning system and method for complex environmental factors.

[0007] In order to achieve the above purpose of the present application, the present application provides a GIS-based base station positioning system and method for complex environmental factors, which comprises:

[0008] Information acquisition module: acquire MR data and CQT data;

[0009] Signal propagation modeling: Combine the ray tracing model with the WINNER II model to build a comprehensive model using the advantages of both models;

[0010] Feature engineering module: introduces signal features and environmental features and combines them into a total feature matrix;

[0011] Model training module: input the feature matrix into the convolutional neural network for deep learning training;

[0012] Base station optimization module: uses the training results of convolutional neural networks to optimize base station operations.

[0013] In a preferred embodiment of the present invention, the information collection and processing module includes:

[0014] Use periodic test data to perform MR data acquisition and CQT data acquisition operations,

[0015] MR data includes reference signal received power, reference signal received quality, and signal-to-interference-plus-noise ratio;

[0016] CQT data requires fixed-point call quality testing at specific locations to evaluate network performance by simulating user communication processes.

[0017] Preprocess the collected MR and CQT data, first perform data cleaning:

[0018] Missing value processing: Check whether there are missing values ​​in the data set and use mean filling, interpolation or discarding missing samples to handle them;

[0019] Outlier detection and processing: Identify obviously unreasonable outliers in the data and filter them using standard deviation analysis;

[0020] The data of reference signal received power, reference signal received quality, and signal interference plus noise ratio are normalized or standardized so that subsequent model analysis can be more effective. The normalization uses the Z-score method, and the normalization scales the data values ​​to the range of [0,1].

[0021] In a preferred embodiment of the present invention, the signal propagation modeling includes:

[0022] Path tracing: Use ray tracing models to calculate all paths from the emission point to the reception point, including direct paths, reflection paths, and diffraction paths;

[0023] For each path i, calculate the path loss due to reflection, refraction and diffraction:

[0024]

[0025] is the total path loss of the i-th path, is the free space loss of the i-th path, and are the reflection, diffraction and refraction caused path losses of the i-th path, respectively;

[0026] The WINNER II model is used to generate the multipath component, fast fading and shadow fading characteristics of each path;

[0027] For each path i, the multipath component loss in the WINNER II model is calculated as:

[0028]

[0029] where, is the result of the superposition of the multipath statistical characteristics of the i-th path, A i is the intercept related to the path i, e i is the path loss exponent of the i-th path, which depends on the environment the path goes through, d i is the distance of the i-th path, is the shadow fading factor of the i-th path;

[0030] The ray tracing calculated path loss is combined with the statistical characteristics of the WINNER II model:

[0031]

[0032] is the comprehensive loss of the i-th path, including the accurate physical loss of ray tracing and the statistical loss of WINNER II;

[0033] Considering the arrival time delay of different paths, the losses of all paths are integrated to obtain the received signal power through multipath superposition:

[0034]

[0035] P r is the received signal power, N is the total number of signal measurement points P t is the transmitted power, G t is the gain of the transmitting antenna, G r is the gain of the receiving antenna, λ is the signal wavelength, is the attenuation of the i-th path signal caused by the comprehensive loss;

[0036] In order to simulate the fast fading effect, the multipath interference and fast fading characteristics in WINNER II are used, and the Rice distribution is used to calculate the superposition effect of different paths at the receiving point;

[0037] The fast fading part of the received signal is expressed as:

[0038]

[0039] h(t) represents the fast fading value of the signal, a i represents the amplitude of the i-th path, which obeys the Rice distribution, φ i is the phase of the i-th path.

[0040] In a preferred embodiment of the present invention, the feature engineering module includes:

[0041] Introducing the joint features of signal strength and location information: Combining signal strength with geographic location information to create more spatial features:

[0042] F location =[P r ,x,y];

[0043] F location is the joint location feature, used to represent the combination of signal strength and location, P r is the received signal power, x and y are the geographical coordinates of the signal measurement point;

[0044] Introducing time-varying features: Introducing the time dimension, calculating the signal strength variation trend within a window at regular intervals to represent network performance in different time periods;

[0045]

[0046] Where Δt is the time interval, Indicates the rate of change of signal strength over time, P r (t) is the received signal power at time t, P r (t+1) is the received signal power at time t+1;

[0047] Introducing spatial clustering of signal coverage areas: clustering different measurement points to find areas with similar signal strengths, which represent similar signal coverage conditions;

[0048] C=kmeans([x,y,P r ]);

[0049] C is the clustering result, indicating similar signal coverage areas. kmeans(*) is a clustering algorithm used to divide data points into several clusters.

[0050] Introducing environmental context: Introducing more environmental information as features, including road types, building density, and user density, obtained through GIS datasets or field surveys;

[0051] F environment =[P r ,B d ,user];

[0052] F environment It is the environmental characteristic, indicating the signal strength and its influence by the surrounding environment. d is the building density, user is the user density,

[0053] Signal stability and fluctuation characteristics: Calculate the standard deviation of the signal strength at the measurement point to indicate the signal's fluctuation and stability;

[0054]

[0055] is the signal strength fluctuation, n represents the number of measurement points, represents the received signal power at the i-th measurement point, is the mean signal power at the measurement point;

[0056] Frequency domain characteristics of the signal: Perform Fourier transform on the time series data of the signal to extract frequency domain features. These features are used to describe the changing frequency of the signal and its periodic fluctuations.

[0057] F frequency =FFT ( P r ′(t));

[0058] F frequency It is the frequency characteristic of the signal, which is used to describe the characteristics of the signal in the frequency domain. r ′(t) is a time series signal, which represents the time domain characteristics of the received signal power. FFT(*) is a fast Fourier transform algorithm used to convert the signal from the time domain to the frequency domain.

[0059] Fusion of signal features into a feature matrix:

[0060]

[0061] The matrix X is the fusion feature matrix.

[0062] In a preferred embodiment of the present invention, the model training module includes:

[0063] Input features include signal propagation characteristics, geographical characteristics, user density, signal fluctuation level, and frequency characteristics;

[0064] Normalize the feature matrix to ensure that all features are in the numerical range of [0,1];

[0065] Model construction:

[0066] Convolutional layer: The core of the convolutional neural network is the convolutional layer, which uses different convolution kernels to scan the input data and automatically learn features. These convolution kernels can extract local spatial features of the data;

[0067] Activation layer: Add an activation layer after the convolution layer and use the ReLU (Rectified Linear Unit) activation function to truncate negative values ​​to zero, introducing nonlinear characteristics so that the model can learn more complex features;

[0068] Pooling layer: In order to reduce the amount of computation and extract the main features in the data, a pooling layer is added to downsample the convolution output, making the feature map smaller but retaining the main information;

[0069] Fully connected layer: After the convolutional layer and pooling layer, the features are flattened and connected to the fully connected layer for final output prediction. The fully connected layer can integrate the features extracted by the convolutional layer for classification or regression;

[0070] Calculate the gradient of the loss function with respect to the network parameters, use the backpropagation algorithm to pass the gradient back to each neuron, update the weights in the network to minimize the loss function, and use the gradient descent method to update the parameters;

[0071] Finally, the training results of the convolutional neural network f model (X) is used for subsequent base station optimization.

[0072] In a preferred embodiment of the present invention, the base station optimization module includes:

[0073] Deep learning model output: Base station location suitability score S j and signal strength coverage prediction P r (x,y);

[0074] Base station site optimization uses data trained by CNN convolutional neural network to optimize the optimal base station location:

[0075] S=[S1,S2,.....,S j ,.....,S J ];

[0076] S j =f model (X j );

[0077] Where S is the set of base station locations in a given area, J is the total number of base stations in a given area, and f model (X j ) is the base station location suitability score calculated by the CNN model, X j Feature matrix at the candidate base station location;

[0078] User resource scheduling:

[0079] Use deep learning models to analyze user mobility patterns and needs, and output user density hotspot distribution;

[0080] U m =f model (X user );

[0081] U m is the user density at the mth position, X user It is the user data feature;

[0082] Visualization tools combined with:

[0083] Combine the prediction results of the deep learning model with GIS tools to display the predicted base station locations and signal coverage on a map;

[0084] Predicting P using signal intensity distribution c , generate a contour map of the coverage to visualize the coverage of each base station:

[0085] P c =f model (X coverage );

[0086] Among them, X coverage are characteristics of different locations in the region;

[0087] Check in GIS whether the location of each base station is affected by terrain and buildings, and use 3D visualization tools to analyze the propagation effect of base station signals in complex terrain;

[0088] Using signal strength coverage to predict P r (x,y) generates three-dimensional graphics:

[0089] Contour(P r (x,y)=P threshold );

[0090] Contour(P r (x,y)=P threshold ) represents the generation of a three-dimensional graph based on signal intensity, P threshold The signal strength threshold is used to distinguish between effective coverage and unavailable coverage areas.

[0091] The present invention also discloses a computer system, comprising:

[0092] a memory for storing processor-executable instructions;

[0093] The processor is configured to implement a base station positioning system based on complex environmental factors based on GIS when executing the executable instructions.

[0094] The present invention also discloses a computer-readable storage medium, comprising:

[0095] a memory having a computer program stored thereon;

[0096] The processor is used to execute the program in the memory to realize a base station positioning system based on complex environmental factors of GIS.

[0097] The present invention also discloses a method for performing base station positioning based on complex environmental factors based on GIS, comprising:

[0098] S1, information acquisition: obtaining MR data and CQT data;

[0099] S2, signal propagation modeling: combining the ray tracing model with the WINNER II model to build a comprehensive model using the advantages of both models;

[0100] S3, feature engineering: introduce signal features and environmental features and combine them into a total feature matrix;

[0101] S4, model training: input the feature matrix into the convolutional neural network for deep learning training;

[0102] S5, base station optimization: Utilize the training results of convolutional neural networks to optimize base station operations.

[0103] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0104] Collecting this data can help researchers and network engineers accurately analyze the coverage and quality of existing networks, thereby determining the optimal location for new base stations. MR provides network performance metrics from the perspective of user devices, while CQT provides coverage and performance data at both macro and micro levels.

[0105] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0106] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments with reference to the accompanying drawings, in which:

[0107] Figure 1 It is a schematic diagram of the system of the present invention. DETAILED DESCRIPTION

[0108] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0109] like Figure 1 As shown, the present invention discloses a base station positioning system based on complex environmental factors based on GIS, including the following:

[0110] Indoor environments are complex, and radio waves are blocked by obstacles, causing reflection, refraction, or scattering, resulting in non-line-of-sight (NLOS) propagation, which severely impacts positioning accuracy. Furthermore, the layout and topology of indoor production environments are susceptible to human influence, leading to changes in acoustic, optical, and electrical propagation, thereby reducing the performance of feature-matching-based positioning technologies. The presence of walls, furniture, and other obstacles complicates wireless signal propagation paths, leading to multipath effects. In multipath propagation, wireless signals travel along multiple distinct paths to reach the receiver.

[0111] 1. Information acquisition and processing module, which uses periodic test data to acquire MR data and CQT data.

[0112] MR data includes reference signal received power, reference signal received quality, and signal-to-interference-plus-noise ratio;

[0113] CQT data requires fixed-point call quality testing at specific locations to evaluate network performance by simulating user communication processes.

[0114] The collected MR and CQT data were preprocessed and data cleaning was performed first.

[0115] Missing value processing: Check whether there are missing values ​​in the dataset. Use mean filling and interpolation to handle them.

[0116] Outlier detection and handling: Identify outliers in the data that are obviously unreasonable (such as measurements with abnormally high or low RSRP), which can be filtered out using statistical methods or thresholds based on domain knowledge.

[0117] The reference signal received power, reference signal received quality, and signal-to-interference-plus-noise ratio data are normalized or standardized to make subsequent model analysis more effective. Normalization typically uses the Z-score method, which scales data values ​​to the range of [0, 1].

[0118] 1.1 Signal propagation modeling:

[0119] Combining the ray tracing model with the WINNER II model takes full advantage of the advantages of both models: the physical accuracy of the ray tracing model and the statistical flexibility of the WINNER II model. The goal of combining these two models is to combine the precise geometric path analysis of ray tracing with the statistical channel characteristics of the WINNER II model to capture all important characteristics of signal propagation.

[0120] Path tracing: Use ray tracing models to calculate all paths from the emission point to the reception point, including direct paths, reflection paths, and diffraction paths;

[0121] For each path i, calculate the path loss due to reflection, refraction and diffraction:

[0122]

[0123] is the total path loss, is the free space loss, and are the path losses caused by reflection, diffraction, and refraction, respectively.

[0124] The WINNER II model is used to generate the multipath components, fast fading, and shadow fading characteristics of each path;

[0125] For each path i, calculate the multipath component loss in the WINNER II model:

[0126]

[0127] in, is the result of the calculation of the superposition of multipath statistical characteristics, A i is the intercept associated with path i, e i is the path loss exponent, which depends on the environment the path passes through, d i is the distance of the i-th path, is the shadow fading factor of the i-th path

[0128] Combining the path loss calculated by ray tracing with the statistical properties of the WINNER II model:

[0129]

[0130] is the total loss of the i-th path, including the exact physical loss of ray tracing and the statistical loss of WINNER II.

[0131] The loss of all paths is synthesized, and the received signal power is obtained by superposition of multiple paths.

[0132] The received signal power is calculated by superposition of the power of all paths, considering the arrival time delay of different paths:

[0133]

[0134] P r is the received signal power, N is the total number of signal measurement points, P t is the transmit power, G t is the gain of the transmit antenna, G r is the gain of the receive antenna, λ is the signal wavelength, is the attenuation of the i-th path signal caused by the synthesized loss;

[0135] In order to simulate the fast fading effect, the multipath interference and fast fading characteristics in WINNER II are used. The superposition effect of different paths at the receiving point is calculated using the Rice distribution;

[0136] The fast fading part of the received signal is represented as:

[0137]

[0138] h(t) represents the fast fading value of the signal, a i represents the amplitude of the i-th path, which follows the Rice distribution, φ i is the phase of the i-th path. The superscript j represents what it means

[0139] Combining ray tracing and WINNER II model makes full use of the advantages of both models:

[0140] The ray tracing model provides accurate physical path analysis of the signal, which can capture the characteristics of reflection, refraction, diffraction, etc. in complex geometric environment.

[0141] The WINNER II model can provide actual measurement data support for multipath statistical characteristics, fast fading and shadow fading, and is suitable for statistical modeling of modern communication systems.

[0142] 2 Data integration and feature engineering:

[0143] Introduce joint features of signal strength and location information: combine signal strength and geographic location information to create more spatial features:

[0144] Flocation =[P r ,x,y]

[0145] F location is the joint location feature, used to represent the combination of signal strength and location, P r is the received signal power, x and y are the geographical coordinates of the signal measurement point;

[0146] Introducing time-varying features: Consider introducing a time dimension. For example, calculate the signal strength variation trend within a window at regular intervals to represent network performance at different time periods.

[0147]

[0148] Where Δt is the time interval, Indicates the rate of change of signal strength over time, P r (t) is the received signal power at time t, P r (t+1) is the received signal power at time t+1

[0149] Introducing spatial clustering of signal coverage areas: clustering different measurement points to find areas with similar signal strengths, which represent similar signal coverage conditions;

[0150] C=k-means([x,y,P r ]);

[0151] C is the clustering result, indicating similar signal coverage areas. k-means(*) is a clustering algorithm used to divide data points into several clusters.

[0152] Introducing environmental context: Introducing more environmental information as features, such as road type and building density, obtained through GIS datasets or field surveys;

[0153] F environment =[P r ,B d ,user]

[0154] F environment It is the environmental characteristic, indicating the signal strength and its influence by the surrounding environment. d is the building density, user is the user density,

[0155] Signal stability and fluctuation characteristics: Calculate the standard deviation of the signal strength at the measurement point to indicate the signal's fluctuation and stability;

[0156]

[0157] is the signal strength fluctuation, N is the number of measurement points, represents the received signal power at the nth measurement point, is the mean signal power at the measurement point.

[0158] Frequency domain features of the signal: Perform Fourier transform on the signal's time series data to extract frequency domain features. These features are used to describe the signal's changing frequency and periodic fluctuations.

[0159] F frequency =FFT(P r ′(t));

[0160] F frequency It is the frequency characteristic of the signal, which is used to describe the characteristics of the signal in the frequency domain. r ′(t) is a time series signal, which represents the time domain characteristics of the received signal power. FFT(*) is a fast Fourier transform algorithm used to convert the signal from the time domain to the frequency domain.

[0161] Fusion of signal features into a feature matrix:

[0162]

[0163] The matrix X is the fusion feature matrix.

[0164] 3. Deep Learning Model Training and Optimization

[0165] Input features include signal propagation characteristics (signal strength, path loss), geographical characteristics (building density, height, etc.), user density, signal fluctuation degree, and frequency characteristics.

[0166] The feature matrix is ​​normalized to ensure that all features are in the range of [0, 1].

[0167] Model construction:

[0168] Convolutional layer: The core of a convolutional neural network is the convolutional layer, which uses different convolution kernels (filters) to scan the input data and automatically learn features. These convolution kernels can extract local spatial features of the data, such as terrain changes around base stations and building layouts.

[0169] Activation layer: Add an activation layer after the convolution layer and use the ReLU (Rectified Linear Unit) activation function to truncate negative values ​​to zero, introducing nonlinear characteristics so that the model can learn more complex features.

[0170] Pooling layer: In order to reduce the amount of computation and extract the main features in the data, a pooling layer (maximum pooling or average pooling) is added to downsample the convolution output to make the feature map smaller but retain the main information.

[0171] Fully connected layer: After the convolutional and pooling layers, the features are flattened and connected to the fully connected layer for the final output prediction. The fully connected layer can integrate the features extracted by the convolutional layer for classification or regression.

[0172] Calculate the gradient of the loss function with respect to the network parameters, use the backpropagation algorithm to pass the gradient back to each neuron, and update the weights in the network to minimize the loss function. Use gradient descent to update the parameters.

[0173] Finally, the training results of the convolutional neural network f model (X) is used for subsequent base station optimization.

[0174] The specific advantages of applying CNN models to base station optimization:

[0175] CNN automatically extracts effective features from input data without the need to manually define the parameters of the signal propagation model (such as path loss exponent, etc.), and is particularly suitable for processing complex geographical environments and multivariate features.

[0176] The convolutional layer can effectively identify patterns in spatial data, such as building distribution and terrain changes. The impact of these factors on signal propagation is complex, and CNN can better analyze and understand the relationship between these factors.

[0177] Through a large amount of training data, the CNN model has good generalization ability and can be applied in different environments to adapt to different base station deployment requirements.

[0178] 4. Base station optimization

[0179] Deep learning model output: Base station location suitability score S j and signal coverage prediction

[0180] Base station site optimization uses data trained by CNN convolutional neural network to optimize the optimal base station location:

[0181] S=[S1,S2,.....,S j ,.....,S J ];

[0182] S j =f model (X j );

[0183] Where S is the set of base station locations in a given area, J is the total number of base stations in a given area, and f model (X j ) is the base station location suitability score calculated by the CNN model, Xj is the feature matrix at the candidate base station location;

[0184] User resource scheduling:

[0185] Use deep learning models to analyze user mobility patterns and needs, and output user density hotspot distribution;

[0186] U m =f model (X user );

[0187] U m is the user density at the mth position, X user It is the user data feature;

[0188] Visualization tools combined with:

[0189] The prediction results of the deep learning model are combined with GIS tools to display the predicted base station site selection and signal coverage on the map.

[0190] Predicting P using signal intensity distribution c , generate a contour map of the coverage to visualize the coverage of each base station:

[0191] P c =f model (X coverage );

[0192] Among them, X coverage are characteristics of different locations in the region;

[0193] Use GIS to check whether the location of each base station is affected by terrain and buildings, and use 3D visualization tools to analyze the propagation effect of base station signals in complex terrain.

[0194] Using signal strength coverage to predict P r (x,y) generates three-dimensional graphics:

[0195] Contour(P r (x,y)=P threshold );

[0196] Contour(P r (x,y)=P threshold ) represents the generation of a three-dimensional graph based on signal intensity, P threshold The signal strength threshold is used to distinguish between effective coverage and unavailable coverage areas.

[0197] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the claims and their equivalents.

Claims

1. A base station positioning system based on complex environmental factors based on GIS, characterized by: include: Information acquisition and processing module: acquiring MR data and CQT data, and processing the data; Signal propagation modeling: Combine the ray tracing model with the WINNER II model to build a comprehensive model using the advantages of both models; Feature engineering module: introduces signal features and environmental features and combines them into a total feature matrix; The feature engineering module includes: Introducing the joint features of signal strength and location information: Combining signal strength with geographic location information to create more spatial features: F location =[P r ,x,y]; F location is a joint location feature, used to represent the combination of signal strength and location, P r is the received signal power, x and y are the geographical coordinates of the signal measurement point; Introducing time-varying features: Introducing the time dimension, calculating the signal strength variation trend within a window at regular intervals to represent network performance in different time periods; Where Δt is the time interval, Indicates the rate of change of signal strength over time, P r (t) is the received signal power at time t, P r (t+1) is the received signal power at time t+1; Introducing spatial clustering of signal coverage areas: clustering different measurement points to find areas with similar signal strengths, which represent similar signal coverage conditions; C=kmeans([x,y,P r ]); C is the clustering result, indicating similar signal coverage areas. kmeans(*) is a clustering algorithm used to divide data points into several clusters. Introducing environmental context: Introducing more environmental information as features, including road types, building density, and user density, obtained through GIS datasets or field surveys; F environment =[P r ,B d ,user]; F environment It is the environmental characteristic, indicating the signal strength and its influence by the surrounding environment. d is the building density, user is the user density, Signal stability and fluctuation characteristics: Calculate the standard deviation of the signal strength at the measurement point to indicate the signal's fluctuation and stability; is the signal strength fluctuation, n represents the number of measurement points, represents the received signal power at the i-th measurement point, is the mean signal power at the measurement point; Frequency domain characteristics of the signal: Perform Fourier transform on the time series data of the signal to extract frequency domain features. These features are used to describe the changing frequency of the signal and its periodic fluctuations. F frequency =FFT(P r ′(t)); F frequency It is the frequency characteristic of the signal, which is used to describe the characteristics of the signal in the frequency domain. r ′(t) is a time series signal, which represents the time domain characteristics of the received signal power. FFT(*) is a fast Fourier transform algorithm used to convert the signal from the time domain to the frequency domain. Fusion of signal features into a total feature matrix: Matrix X is the total feature matrix; Model training module: input the total feature matrix into the convolutional neural network for deep learning training; Base station optimization module: Utilizes the training results of the convolutional neural network to optimize the base station operation; the base station optimization module includes: CNN model output: base station location suitability score S j and signal strength coverage prediction P r (x,y); Base station site optimization uses data trained by CNN convolutional neural network to optimize the optimal base station location: S=[S1,S2,.....,S j ,.....,S J ]; S j =f model (X j ); Where S is the set of base station locations in a given area, J is the total number of base stations in a given area, and f model (X j ) is the base station location suitability score calculated by the CNN model, X j is the feature matrix at the candidate base station location; User resource scheduling: Use CNN models to analyze user mobility patterns and needs, and output user density hotspot distribution; U m =f model (X user ); U m is the user density at the mth position, X user It is the user data feature; Visualization tools combined with: Combine the CNN model's prediction results with GIS tools to display the predicted base station locations and signal coverage on a map; Predicting P using signal intensity distribution c , generate a contour map of the coverage to visualize the coverage of each base station: P c =f model (X coverage ); Among them, X coverage are characteristics of different locations in the region; Check in GIS whether the location of each base station is affected by terrain and buildings, and use 3D visualization tools to analyze the propagation effect of base station signals in complex terrain; Using signal strength coverage to predict P r (x,y) generates three-dimensional graphics: Contour(P r (x,y)=P threshold ); Contour(P r (x,y)=P threshold ) represents the generation of a three-dimensional graph based on signal intensity, P threshold The signal strength threshold is used to distinguish between effective coverage and unavailable coverage areas.

2. The base station positioning system based on complex environmental factors based on GIS according to claim 1 is characterized in that: The information collection and processing module includes: Use periodic test data to perform MR data acquisition and CQT data acquisition operations, MR data includes reference signal received power, reference signal received quality, and signal-to-interference-plus-noise ratio; CQT data requires fixed-point call quality testing at specific locations to evaluate network performance by simulating user communication processes. Preprocessing the collected MR and CQT data, including missing value processing and / or outlier detection and processing and / or normalization or standardization; Missing value processing: Check whether there are missing values ​​in the data set and use mean filling, interpolation or discarding missing samples to handle them; Outlier detection and processing: Identify obviously unreasonable outliers in the data and filter them using standard deviation analysis; The data of reference signal received power, reference signal received quality, and signal interference plus noise ratio are normalized or standardized so that subsequent model analysis can be more effective. The normalization uses the Z-score method, and the normalization scales the data values ​​to the range of [0,1].

3. The base station positioning system based on complex environmental factors based on GIS according to claim 1 is characterized in that: The signal propagation modeling includes: Path tracing: Use ray tracing models to calculate all paths from the emission point to the reception point, including direct paths, reflection paths, and diffraction paths; For each path i, calculate the path loss due to reflection, refraction and diffraction: is the total path loss of the ith path, is the free space loss of the ith path, and are the path losses caused by reflection, diffraction, and refraction of the i-th path, respectively; The WINNER II model is used to generate the multipath components, fast fading, and shadow fading characteristics of each path; For each path i, calculate the multipath component loss in the WINNER II model: in, is the calculation result of the superposition of the multipath statistical characteristics of the i-th path, A i is the intercept associated with path i, e i is the path loss exponent of the ith path, which depends on the environment through which the path passes, d i is the distance of the i-th path, is the shadow fading factor of the i-th path; Combining the path loss calculated by ray tracing with the statistical properties of the WINNER II model: is the comprehensive loss of the i-th path, including the accurate physical loss of ray tracing and the statistical loss of WINNER II; Considering the arrival delays of different paths, the losses of all paths are combined, and the received signal power is obtained by multipath superposition: P r is the received signal power, N is the total number of signal measurement points, P t is the transmit power, G t is the gain of the transmitting antenna, G r is the gain of the receiving antenna, λ is the signal wavelength, is the attenuation of the signal on the i-th path due to the comprehensive loss; In order to simulate the fast fading effect, the multipath interference and fast fading characteristics in WINNER II are used, and the Rice distribution is used to calculate the superposition effect of different paths at the receiving point; The fast fading part of the received signal is expressed as: h(t) represents the fast fading value of the signal, a i represents the amplitude of the i-th path, which obeys the Rice distribution, φ i is the phase of the i-th path.

4. The base station positioning system based on complex environmental factors based on GIS according to claim 1 is characterized in that: Its model training modules include: Input features include signal propagation characteristics, geographical characteristics, user density, signal fluctuation level, and frequency characteristics; Normalize the feature matrix to ensure that all features are in the numerical range of [0,1]; Model construction: Convolutional layer: The core of the convolutional neural network is the convolutional layer, which uses different convolution kernels to scan the input data and automatically learn features. These convolution kernels can extract local spatial features of the data; Activation layer: Add an activation layer after the convolution layer, use the ReLU activation function, truncate negative values ​​to zero, introduce nonlinear characteristics, and enable the model to learn more complex features; Pooling layer: In order to reduce the amount of computation and extract the main features in the data, a pooling layer is added to downsample the convolution output, making the feature map smaller but retaining the main information; Fully connected layer: After the convolutional layer and pooling layer, the features are flattened and connected to the fully connected layer for final output prediction. The fully connected layer can integrate the features extracted by the convolutional layer for classification or regression; Calculate the gradient of the loss function with respect to the network parameters, use the backpropagation algorithm to pass the gradient back to each neuron, update the weights in the network to minimize the loss function, and use the gradient descent method to update the parameters; Finally, the training results of the convolutional neural network f model (X) is used for subsequent base station optimization.

5. A method for base station positioning based on complex environmental factors based on GIS, characterized in that: include: S1, information acquisition: obtaining MR data and CQT data; S2, signal propagation modeling: combining the ray tracing model with the WINNER II model to build a comprehensive model using the advantages of both models; S3, feature engineering: introduce signal features and environmental features and combine them into a total feature matrix; Step S3 includes: Introducing the joint features of signal strength and location information: Combining signal strength with geographic location information to create more spatial features: F location =[P r ,x,y]; F location is a joint location feature, used to represent the combination of signal strength and location, P r is the received signal power, x and y are the geographical coordinates of the signal measurement point; Introducing time-varying features: Introducing the time dimension, calculating the signal strength variation trend within a window at regular intervals to represent network performance in different time periods; Where Δt is the time interval, Indicates the rate of change of signal strength over time, P r (t) is the received signal power at time t, P r (t+1) is the received signal power at time t+1; Introducing spatial clustering of signal coverage areas: clustering different measurement points to find areas with similar signal strengths, which represent similar signal coverage conditions; C=kmeans([x,y,P r ]); C is the clustering result, indicating similar signal coverage areas. kmeans(*) is a clustering algorithm used to divide data points into several clusters. Introducing environmental context: Introducing more environmental information as features, including road types, building density, and user density, obtained through GIS datasets or field surveys; F environment =[P r ,B d ,user]; F environment It is the environmental characteristic, indicating the signal strength and its influence by the surrounding environment. d is the building density, user is the user density, Signal stability and fluctuation characteristics: Calculate the standard deviation of the signal strength at the measurement point to indicate the signal's fluctuation and stability; is the signal strength fluctuation, n represents the number of measurement points, represents the received signal power at the i-th measurement point, is the mean signal power at the measurement point; Frequency domain characteristics of the signal: Perform Fourier transform on the time series data of the signal to extract frequency domain features. These features are used to describe the changing frequency of the signal and its periodic fluctuations. F frequency =FFT(P r ′(t)); F frequency It is the frequency characteristic of the signal, which is used to describe the characteristics of the signal in the frequency domain. r ′(t) is a time series signal, which represents the time domain characteristics of the received signal power. FFT(*) is a fast Fourier transform algorithm used to convert the signal from the time domain to the frequency domain. Fusion of signal features into a total feature matrix: Matrix X is the total feature matrix; Model training module: input the total feature matrix into the convolutional neural network for deep learning training; S4, model training: input the total feature matrix into the convolutional neural network for deep learning training; S5, base station optimization: using the training results of the convolutional neural network to optimize the base station operation; step S5 includes: CNN model output: base station location suitability score S j and signal strength coverage prediction P r (x,y); Base station site optimization uses data trained by CNN convolutional neural network to optimize the optimal base station location: S=[S1,S2,.....,S j ,.....,S J ]; S j =f model (X j ); Where S is the set of base station locations in a given area, J is the total number of base stations in a given area, and f model (X j ) is the base station location suitability score calculated by the CNN model, X j is the feature matrix at the candidate base station location; User resource scheduling: Use CNN models to analyze user mobility patterns and needs, and output user density hotspot distribution; U m =f model (X user ); U m is the user density at the mth position, X user It is the user data feature; Visualization tools combined with: Combine the CNN model's prediction results with GIS tools to display the predicted base station locations and signal coverage on a map; Predicting P using signal intensity distribution c , generate a contour map of the coverage to visualize the coverage of each base station: P c =f model (X coverage ); Among them, X coverage are characteristics of different locations in the region; Check in GIS whether the location of each base station is affected by terrain and buildings, and use 3D visualization tools to analyze the propagation effect of base station signals in complex terrain; Using signal strength coverage to predict P r (x,y) generates three-dimensional graphics: Contour(P r (x,y)=P threshold ); Contour(P r (x,y)=P threshold ) represents the generation of a three-dimensional graph based on signal intensity, P threshold The signal strength threshold is used to distinguish between effective coverage and unavailable coverage areas.

Images