Construction method, system and equipment of outdoor microcell scene electromagnetic propagation prediction model and medium
Through GBDT algorithm and data preprocessing technology, an outdoor microcellular scene electromagnetic propagation prediction model was constructed, solving the problem of insufficient prediction accuracy of traditional models in complex environments, and achieving efficient and accurate electromagnetic propagation prediction.
Patent Information
- Application Number
- CN202411307512.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-19
- Publication Date
- 2024-12-27
AI Technical Summary
The prior art is difficult to accurately describe the complex propagation path of electromagnetic waves in outdoor microcellular scenarios, especially in urban dense environments. The prediction accuracy of traditional models is affected by multipath effect and shadow effect, and cannot adapt to different scenarios and data distributions. Electromagnetic propagation prediction task.
Gradient improvement decision tree (GBDT) algorithm is used for training, and the original data is obtained through CW tests, data cleaning, geographical averaging and standardization are performed, key features are extracted, hyperparameters are adjusted to build an electromagnetic propagation prediction model to achieve accurate prediction of complex environments. .
It significantly improves the accuracy and efficiency of electromagnetic propagation prediction, can accurately predict electromagnetic propagation paths in different scenarios and diversified data distribution, has high flexibility and high prediction accuracy, and is suitable for complex electromagnetic propagation environment modeling.
Smart Images
Figure CN119210633A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of radio wave propagation prediction model construction, and in particular to a method, system, device and medium for constructing an outdoor microcellular scene electromagnetic propagation prediction model. Background Art
[0002] With the rapid development of wireless communication technology, the coverage and capacity requirements of mobile communication networks are growing. Micro-cell base stations have become an important means to improve network performance and optimize user experience due to their small coverage and high capacity. However, the deployment and planning of micro-cell base stations face a complex propagation environment, especially in outdoor scenarios. Electromagnetic wave propagation is affected by multiple factors such as terrain, buildings, and vegetation, which makes the prediction of propagation models more difficult.
[0003] Ray tracing models based on deterministic algorithms came into being. RT algorithms are widely used as deterministic algorithms, and the algorithms rely on GO theory and UTD theory. With the development of science and technology, ray tracing algorithms can handle more complex terrain data and accurate building information, and then perform radio wave prediction. Compared with traditional empirical models, RT algorithms can perform more accurate radio wave predictions, but the prediction accuracy and efficiency are heavily dependent on the accuracy of set parameters and environmental information. Traditional electromagnetic propagation models, such as free space models and Okumura-Hata models, are usually used for predictions in macrocellular scenarios. These models perform well in large-scale environments, but their prediction accuracy is limited in microcellular scenarios. In particular, in dense urban environments, the signal propagation path is complex, and the multipath effect and shadow effect are significant. It is difficult for traditional models to accurately describe the actual propagation situation.
[0004] In recent years, machine learning technology has shown great potential in electromagnetic propagation prediction. Compared with traditional statistical methods or simple regression models, the GBDT (gradient boosted decision tree) model is a more advanced integrated learning method. GBDT gradually improves the prediction accuracy by constructing multiple weak prediction models (decision trees). GBDT, a highly flexible model, can adapt to electromagnetic propagation prediction tasks in different scenarios and different data distributions. It performs well in processing nonlinear relationships and high-dimensional features, is not easily limited by specific scenarios, and is suitable for modeling complex electromagnetic propagation environments.
[0005] In the conference proceedings "Localization of Microcell Models by Integrating Measured and Simulated Data" (April 2024, DOI: 10.1109 / PIERS62282.2024.10618362, Conference: 2024 Photonics & Electromagnetics Research Symposium (PIERS)), four propagation models, including the free space path loss model, the logarithmic distance path loss model, the free space reference distance path loss model and the multi-slope path loss model, were deeply analyzed. However, the above models have poor applicability and flexibility. After changing to a new scenario, a lot of measured data is required to support the model. They cannot adapt to electromagnetic propagation prediction tasks under different scenarios and different data distributions, and cannot be used for complex electromagnetic propagation environment modeling. Summary of the invention
[0006] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide a method, system, device and medium for constructing an electromagnetic propagation prediction model for an outdoor microcellular scenario, which adopts a gradient boosting decision tree algorithm for training, can effectively cope with the complex propagation environment in the microcellular scenario, provide accurate propagation prediction for network planning, and realize localized prediction of the actual propagation environment.
[0007] To achieve the above object, the present invention provides the following technical solutions:
[0008] A method for constructing an outdoor microcellular scene electromagnetic propagation prediction model comprises the following steps:
[0009] Step S1, performing a power test on a single carrier signal on a preset route through a CW test to obtain an original training data set, wherein the original training data set includes but is not limited to frequency (MHz), transmitting antenna x coordinate (latitude and longitude coordinates), transmitting antenna y coordinate (latitude and longitude coordinates), transmitting antenna height (m), receiving antenna x coordinate (latitude and longitude coordinates), receiving antenna y coordinate (latitude and longitude coordinates), receiving antenna height (m) and power (dBm);
[0010] Step S2, performing data cleaning and geographic averaging preprocessing on the original training data set obtained in step S1;
[0011] Step S3, extracting key features that affect electromagnetic propagation according to the original training data set processed in step S2, the key features including but not limited to frequency (MHz), transmitting antenna x-coordinate (UTM coordinate), transmitting antenna y-coordinate (UTM coordinate), transmitting antenna height (m), receiving antenna x-coordinate (UTM coordinate), receiving antenna y-coordinate (UTM coordinate), receiving antenna height (m), whether line of sight (line of sight is 0, non-line of sight is 1), building density and power (dBm), using the extracted key features as a test data set, and performing standardization on the test data set;
[0012] Step S4, use the gradient boosted decision tree (GBDT) algorithm to train the test data set obtained in step S3, build an electromagnetic propagation prediction model, adjust the hyperparameters of the electromagnetic propagation prediction model, the hyperparameters include the number of trees, maximum depth and learning rate, and optimize the model performance; use the optimal hyperparameters to train to obtain the optimal GBDT model, and verify the optimal GBDT model.
[0013] The CW test in step S1 is to place a transmitting antenna in a set area, transmit a single carrier signal, and then perform a power test on a preset route.
[0014] The data cleaning in step S2 refers to filtering out data with power values less than -110dBm; the geographical averaging refers to performing geographical averaging on the measured power data according to Li's theorem, and taking the average of the received powers of the measured points in the same interval.
[0015] In step S4, the test data set obtained in step S3 is trained using a gradient boosted decision tree (GBDT) algorithm, and the steps of constructing an electromagnetic propagation prediction model include:
[0016] S4.1. Data loading and partitioning: Load the test data set and randomly divide it into two parts, one for training and one for validation. The input data X contains the feature values of the first 9 columns, and Y contains the label (target value) of the 10th column.
[0017] S4.2. Standardize the input data: Use the Z-score method, the specific formula is:
[0018]
[0019] Among them, X represents the original data, μ and σ represent the mean and standard deviation of the corresponding input dimension respectively; after Z-score standardization, the data will be expressed as its degree of deviation relative to the entire data set, so that the data of different feature dimensions have the same scale;
[0020] S4.3. Hyperparameter tuning: Hyperparameter grid search is used to traverse multiple hyperparameter combinations through nested loops, and the hyperparameters of the GBDT model are tuned using the 5-fold cross-validation method to minimize the mean square error (MSE).
[0021] S4.4. Best model training: Use the best parameters obtained in hyperparameter tuning to train the final GBDT model, save the model to a file, and verify the final model.
[0022] The 5-fold cross-validation method is specifically as follows: first, the training data is randomly divided into 5 non-overlapping subsets, one subset is selected as the validation set each time, and the remaining 4 subsets are used as training sets. The model is trained in turn and the minimized mean square error (MSE) on the validation set is calculated; the minimized mean square error (MSE) of the 5 validations is averaged as the evaluation result of the group of hyperparameters.
[0023] The formula for minimizing the mean square error (MSE) is:
[0024]
[0025] Where y i To test the power value, is the predicted power value, and n is the number of samples in the test set.
[0026] A system for constructing an electromagnetic propagation prediction model for an outdoor micro-cellular scenario, comprising:
[0027] A data test module is used to perform a power test on a single carrier signal on a preset route through a CW test to obtain an original training data set;
[0028] The data preprocessing module is used to perform data cleaning and geographic averaging preprocessing on the original training data set;
[0029] A data extraction and standardization module is used to extract key features that affect electromagnetic propagation, use the extracted key features as a test data set, and perform standardization on the test data set;
[0030] The training and prediction module is used to train the test data set, build an electromagnetic propagation prediction model, use the best hyperparameters to train the optimal GBDT model, and verify the optimal GBDT model.
[0031] A device for constructing an outdoor micro-cellular scene electromagnetic propagation prediction model, comprising:
[0032] A memory for storing the acquired data and a computer program for storing the outdoor microcellular scene electromagnetic propagation prediction model constructed in steps S1 to S4;
[0033] A processor is used to implement the outdoor micro-cellular scenario electromagnetic propagation prediction described in steps S1 to S4 when executing the computer program.
[0034] A computer-readable storage medium stores a computer program, which can predict electromagnetic propagation in an outdoor micro-cellular scene when executed by a processor.
[0035] Compared with the prior art, the present invention has the following beneficial effects:
[0036] 1. The present invention establishes an original training data set containing a large amount of data through CW testing; then performs geographic averaging, data cleaning and data standardization on the original training data set, and then extracts key features that affect electromagnetic propagation, thereby amplifying sample features and using the gradient boosting decision tree algorithm for training. Through verification, the model trained by the present invention performs well in the test set and can be used to establish a radio wave propagation prediction model. The present invention effectively improves the accuracy and efficiency of electromagnetic propagation prediction through sophisticated data preprocessing and integrated machine learning algorithms, providing strong support for the optimization and planning of wireless networks.
[0037] 2. The present invention significantly improves the performance of the electromagnetic propagation prediction model in outdoor microcellular scenarios by introducing the GBDT model, systematic hyperparameter optimization and sophisticated data preprocessing. Compared with traditional methods, the model shows obvious advantages in prediction accuracy, generalization ability and adaptability. The GBDT model is highly flexible and can accurately predict electromagnetic propagation paths in different scenarios and diverse data distributions, especially when dealing with nonlinear relationships and high-dimensional features.
[0038] In summary, the present invention performs CW test on the signal, introduces the GBDT model, systematic hyperparameter optimization and sophisticated data preprocessing, optimizes the prediction model hyperparameters using the grid search method, and updates the model by minimizing the gradient of the loss function. It is capable of modeling complex electromagnetic propagation environments, is not easily restricted by specific scenarios, and has the characteristics of high precision and high prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 It is a flow chart of the present invention.
[0040] Figure 2 It is a main flow chart of a prediction model using a gradient boosting decision tree algorithm provided by an embodiment of the present invention.
[0041] Figure 3 It is a schematic diagram of a test system provided by an embodiment of the present invention.
[0042] Figure 4 It is a schematic diagram of the energy value of the test path provided by an embodiment of the present invention.
[0043] Figure 5 It is a schematic diagram of predicted path energy value provided by an embodiment of the present invention.
[0044] Figure 6 It is a schematic diagram comparing the actual measurement results and the model prediction results provided by the embodiment of the present invention.
[0045] Figure 7 It is a residual histogram provided by an embodiment of the present invention.
[0046] Figure 8 It is a schematic diagram comparing the actual measurement results and the prediction results of the neural network model provided in an embodiment of the present invention.
[0047] Fig. 9 It is a residual histogram of the actual measurement and neural network model prediction results provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0048] In order to more clearly illustrate the technical solution and advantages of the present invention, the present invention is described in detail below in combination with specific implementation steps. It should be noted that the implementation steps described below are only used to explain the present invention, rather than to limit the present invention.
[0049] Reference Figure 1 , a method for constructing an electromagnetic propagation prediction model for an outdoor micro-cellular scenario, specifically comprising the following steps:
[0050] Step S1, perform a power test on a single carrier signal on a preset route through a CW test to obtain an original training data set, where the original training data set includes but is not limited to frequency (MHz), transmitting antenna x-coordinate (latitude and longitude coordinates), transmitting antenna y-coordinate (latitude and longitude coordinates), transmitting antenna height (m), receiving antenna x-coordinate (latitude and longitude coordinates), receiving antenna y-coordinate (latitude and longitude coordinates), receiving antenna height (m) and power (dBm); based on the propagation model correction result of the CW test, accurate model prediction can be provided for network planning to achieve localized prediction of the propagation environment.
[0051] Step S2: Perform data cleaning and geographic averaging preprocessing on the original training data set obtained in step S1; since wireless signals are affected by factors such as multipath effects and shadow effects during propagation, the measurement data fluctuates. Data cleaning can eliminate extreme outliers caused by hardware failures and environmental interference to ensure data quality. Geographic averaging can smooth these fluctuations, making the test results more representative and stable.
[0052] Step S3, extracting key features that affect electromagnetic propagation according to the original training data set processed in step S2, the key features including but not limited to frequency (MHz), transmitting antenna x-coordinate (UTM coordinate), transmitting antenna y-coordinate (UTM coordinate), transmitting antenna height (m), receiving antenna x-coordinate (UTM coordinate), receiving antenna y-coordinate (UTM coordinate), receiving antenna height (m), whether there is line of sight (line of sight is 0, non-line of sight is 1), building density and power (dBm), using the extracted key features as a test data set, and standardizing the test data set; standardization unifies the numerical ranges of different features, which can improve the performance and training effect of the model.
[0053] Step S4, use the gradient boosted decision tree (GBDT) algorithm to train the test data set obtained in step S3, build an electromagnetic propagation prediction model, adjust the hyperparameters of the electromagnetic propagation prediction model, the hyperparameters include the number of trees, maximum depth and learning rate, and optimize the model performance; use the optimal hyperparameters to train to obtain the optimal GBDT model, and verify the optimal GBDT model.
[0054] The CW test in step S1 is to place a transmitting antenna in a set area, transmit a single carrier signal, and then perform a power test on a preset route; obtaining sufficient data through the CW test and processing it using the mean method can ensure that the error between the sample point mean and the actual mean is within an acceptable range. In the actual operation of the cellular mobile communication system, Li's theorem should be followed, that is, the sampling length should be 40 wavelengths, the sampling interval should be 0.8 wavelengths, and the number of sampling points should be between 36 and 50. In addition, the ratio of the vehicle speed to the device sampling rate should be equal to the sampling interval.
[0055] The data cleaning in step S2 refers to filtering out data with power values less than -110dBm; the geographical averaging refers to performing geographical averaging on the measured power data according to Li's theorem, and taking the average of the received powers of the measured points in the same interval.
[0056] In step S4, the test data set obtained in step S3 is trained using a gradient boosted decision tree (GBDT) algorithm, and the steps of constructing an electromagnetic propagation prediction model include:
[0057] S4.1. Data loading and partitioning: Load the test data set and randomly divide it into two parts, one for training and one for validation. The input data X contains the feature values of the first 9 columns, and Y contains the label (target value) of the 10th column.
[0058] S4.2. Standardize the input data: Use the Z-score method, the specific formula is:
[0059]
[0060] Among them, X represents the original data, μ and σ represent the mean and standard deviation of the corresponding input dimension respectively; after Z-score standardization, the data will be expressed as its degree of deviation relative to the entire data set, so that the data of different feature dimensions have the same scale;
[0061] S4.3. Hyperparameter tuning: Hyperparameter grid search is used to traverse multiple hyperparameter combinations through nested loops, and the hyperparameters of the GBDT model are tuned using the 5-fold cross-validation method to minimize the mean square error (MSE).
[0062] S4.4. Best model training: Use the best parameters obtained in hyperparameter tuning to train the final GBDT model, save the model to a file, and verify the final model.
[0063] The 5-fold cross-validation method is specifically as follows: first, the training data is randomly divided into 5 non-overlapping subsets, one subset is selected as the validation set each time, and the remaining 4 subsets are used as training sets. The model is trained in turn and the minimized mean square error (MSE) on the validation set is calculated; the minimized mean square error (MSE) of the 5 validations is averaged as the evaluation result of the group of hyperparameters.
[0064] The formula for minimizing the mean square error (MSE) is:
[0065]
[0066] Where y i To test the power value, is the predicted power value, and n is the number of samples in the test set.
[0067] The above description is only a specific embodiment of the present invention and does not constitute any limitation to the present invention. Obviously, for professionals in this field, after understanding the content and design principles of the present invention, they may make various modifications and changes in form and details based on the principles and structures of the present invention. However, these modifications and changes based on the ideas of the present invention are still within the scope of protection of the claims of the present invention.
[0068] Example
[0069] GBDT is an ensemble learning method for regression and classification problems. In regression problems, GBDT uses decision trees as basic learners and gradually improves the accuracy of the model through gradient boosting. Decision tree is a tree-based machine learning algorithm that divides data into different categories or predicts the value of a target variable through a series of decision nodes. In GBDT, decision trees are often used as basic weak learners, and each tree is responsible for capturing part of the pattern in the data. By using the ensemble learning technique of gradient boosting, a series of weak learners (CART regression tree model) are iteratively trained, and each step is optimized on the residual of the previous step. The model is updated by minimizing the gradient of the loss function to gradually reduce the residual and thus improve the accuracy of the model. The main process of predicting the model using the gradient boosting decision tree algorithm is as follows: Figure 2 The specific operation process is as follows:
[0070] The input is the training set sample T = {(x1,y1),(x2,y2),...(x m ,y m )}, maximum number of iterations K, loss function L(y,f(x)), and the output is a strong learner f(x).
[0071] (1) Initialize weak learners
[0072]
[0073] (2) For k = 1, 2, ..., K
[0074] a) For i = 1, 2, ..., M, calculate
[0075]
[0076] b) Using r ki Fit a CART regression tree and obtain the leaf node region R of the kth regression tree kj , where j=1,2...J, J is the number of leaf nodes of regression tree k.
[0077] c) For leaf region j=1,2...,J, calculate the best fit value of the sample
[0078]
[0079] d) Update the strong learner
[0080]
[0081] (3) Output the final model
[0082]
[0083] The signal generator used in the present invention is a 2W signal generator with a test frequency band range of 20 to 6000 MHz and an amplitude of -10 to +33 dBm. The receiving antenna used is a vertically polarized, multi-band magnetic antenna of PCTEL model OP451. The PCTEL frequency sweeper is wirelessly connected to the test tablet computer PAD via Bluetooth, the RF omnidirectional test antenna is connected to RF1, and the GPS antenna is connected to the GPS interface. The experimental parameters are set through the test tablet computer PAD and monitored during the test. The spectrum analyzer uses the N9951A FieldFox handheld microwave analyzer, 44 GHz. The transmitting antenna uses a biconical antenna model SZ-80400 / P with a gain of 0 dB. The test system is as follows Figure 3 shown.
[0084] After setting up the test system, power measurements were performed in an outdoor micro-cell scenario. The transmit power was set to 10 dBm and the frequency was set to 1260 MHz. The transmit antenna height was 1.8 m and the receive antenna height was 1.67 m. The energy values of the test paths are as follows: Figure 4 shown.
[0085] The final prediction model is obtained by training the test data set using the GBDT model. The power prediction of the test data set is performed using the final model, and the predicted path energy value is as follows: Figure 5 The predicted results of the GBDT model are compared with the measured power values, as shown in Figure 6 As shown in the figure, the horizontal axis represents the receiving antenna position number and the vertical axis represents the power value. The mean square error (MSE) calculation formula is:
[0086]
[0087] Where y i To test the power value, is the predicted power value, and n is the number of samples in the test set.
[0088] The residual is obtained by subtracting the predicted power value from the measured power value. The result is as follows: Figure 7 As shown in the figure, the vertical axis represents the receiving antenna position number, and the vertical axis represents the residual value. The residual value is basically distributed in the range of -10dB to 10dB, and the proportion of the residual absolute value less than 5dB is 95.43%.
[0089] like Figure 8As shown in the figure, a four-layer neural network model is used to predict radio wave propagation, and the network structure is 9×20×20×1. The input layer of the neural network model has the same number of elements as the GBDT algorithm test data set, and the data is also processed using Z-score standardization. The test data set is also divided into two parts according to the ratio of 7:3, one for training and the other for verification; the input layer X contains the feature values of the first 9 columns of the test data set, and the output layer Y contains the label (target value) of the 10th column of the test data set. The mean square error between the neural network prediction and the measured data is better than Figure 5 The mean square error of the GBDT shown is about 1.1dB higher than that of the measured data. Fig. 9 As shown, the residual values are basically distributed in the range of -15dB to 10dB, and the proportion of residual absolute values less than 5dB is 85.27%. By comparison, it can be seen that the GBDT algorithm proposed in the present invention can perform higher-precision radio wave propagation prediction.
Claims
1. A method for constructing an electromagnetic propagation prediction model for an outdoor microcellular scenario, characterized in that: The following steps are involved: Step S1, performing a power test on a single carrier signal on a preset route through a CW test to obtain an original training data set, wherein the original training data set includes but is not limited to frequency (MHz), transmitting antenna x coordinate (latitude and longitude coordinates), transmitting antenna y coordinate (latitude and longitude coordinates), transmitting antenna height (m), receiving antenna x coordinate (latitude and longitude coordinates), receiving antenna y coordinate (latitude and longitude coordinates), receiving antenna height (m) and power (dBm); Step S2, performing data cleaning and geographic averaging preprocessing on the original training data set obtained in step S1; Step S3, extracting key features that affect electromagnetic propagation according to the original training data set processed in step S2, the key features including but not limited to frequency (MHz), transmitting antenna x-coordinate (UTM coordinate), transmitting antenna y-coordinate (UTM coordinate), transmitting antenna height (m), receiving antenna x-coordinate (UTM coordinate), receiving antenna y-coordinate (UTM coordinate), receiving antenna height (m), whether line of sight (line of sight is 0, non-line of sight is 1), building density and power (dBm), using the extracted key features as a test data set, and performing standardization on the test data set; Step S4, use the gradient boosted decision tree (GBDT) algorithm to train the test data set obtained in step S3, build an electromagnetic propagation prediction model, adjust the hyperparameters of the electromagnetic propagation prediction model, the hyperparameters include the number of trees, maximum depth and learning rate, and optimize the model performance; use the optimal hyperparameters to train to obtain the optimal GBDT model, and verify the optimal GBDT model.
2. The method for constructing an outdoor microcellular scene electromagnetic propagation prediction model according to claim 1, characterized in that: The CW test in step S1 is to place a transmitting antenna in a set area, transmit a single carrier signal, and then perform a power test on a preset route.
3. The method for constructing an outdoor microcellular scene electromagnetic propagation prediction model according to claim 1, characterized in that: The data cleaning in step S2 refers to filtering out data with power values less than -110dBm; the geographical averaging refers to performing geographical averaging on the measured power data according to Li's theorem, and taking the average of the received powers of the measured points in the same interval.
4. The method for constructing an outdoor microcellular scene electromagnetic propagation prediction model according to claim 1, characterized in that: In step S4, the test data set obtained in step S3 is trained using a gradient boosted decision tree (GBDT) algorithm, and the steps of constructing an electromagnetic propagation prediction model include: S4.
1. Data loading and partitioning: Load the test data set and randomly divide it into two parts, one for training and one for validation. The input data X contains the feature values of the first 9 columns, and Y contains the label (target value) of the 10th column. S4.
2. Standardize the input data: Use the Z-score method, the specific formula is: Among them, X represents the original data, μ and σ represent the mean and standard deviation of the corresponding input dimension respectively; after Z-score standardization, the data will be expressed as its degree of deviation relative to the entire data set, so that the data of different feature dimensions have the same scale; S4.
3. Hyperparameter tuning: Hyperparameter grid search is used to traverse multiple hyperparameter combinations through nested loops, and the hyperparameters of the GBDT model are tuned using the 5-fold cross-validation method to minimize the mean square error (MSE). S4.
4. Best model training: Use the best parameters obtained in hyperparameter tuning to train the final GBDT model, save the model to a file, and verify the final model.
5. The method for constructing an outdoor microcellular scene electromagnetic propagation prediction model according to claim 4, characterized in that: The 5-fold cross-validation method is specifically as follows: first, the training data is randomly divided into 5 non-overlapping subsets, one subset is selected as the validation set each time, and the remaining 4 subsets are used as training sets. The model is trained in turn and the minimized mean square error (MSE) on the validation set is calculated; the minimized mean square error (MSE) of the 5 validations is averaged as the evaluation result of the group of hyperparameters. The formula for minimizing the mean square error (MSE) is: Where y i To test the power value, is the predicted power value, and n is the number of samples in the test set.
6. A system for constructing an electromagnetic propagation prediction model for an outdoor microcellular scenario, characterized in that: include: The data test module is used to perform a power test on the signal on a preset route through a CW test to obtain an original training data set; The data preprocessing module is used to perform data cleaning and geographic averaging preprocessing on the obtained test data set; A data extraction and standardization module is used to extract key features that affect electromagnetic propagation, use the extracted key features as a test data set, and perform standardization on the test data set; The training and prediction module is used to train the test data set, build an electromagnetic propagation prediction model, use the best hyperparameters to train the optimal GBDT model, and verify the optimal GBDT model.
7. A device for constructing an outdoor microcellular scene electromagnetic propagation prediction model, characterized in that: include: A memory for storing the acquired data and a computer program for storing the outdoor microcellular scene electromagnetic propagation prediction model constructed in steps S1 to S4; A processor is used to implement the outdoor micro-cellular scenario electromagnetic propagation prediction described in steps S1 to S4 when executing the computer program.
8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, is capable of predicting electromagnetic propagation in an outdoor microcellular scenario.