Ray tracing actual measurement correction method and system based on multi-modal fusion and electronic equipment

Abstract

The invention discloses a ray tracing actual measurement correction method and system based on multi-modal fusion and electronic equipment, and mainly solves the problem of insufficient prediction precision of a traditional model in a complex propagation environment. According to the implementation scheme, the method comprises the following steps: respectively obtaining a convolutional layer training sample set, a test sample set and a self-attention layer training sample set comprising actual measurement data; constructing a multi-modal fusion network model comprising a convolutional neural network module, a self-attention mechanism module and a fusion module, wherein the modules are cascaded in sequence; using the training sample set to train the multi-modal fusion prediction network model; and performing forward propagation by taking the test sample set as the input of the trained prediction network model to obtain a radio wave propagation actual measurement receiving power prediction value. The method effectively reduces the error between the power measured value and the theoretical value of the receiving point in a complex scene, can process different types of data in a targeted manner, improves the prediction precision of the receiving power, meets the application requirements of a complex propagation environment, and can be used for electronic countermeasure, automatic command and other systems and electromagnetic wave propagation.

Description

Technical Field

[0001] The present invention belongs to the technical field of big data and machine learning applications, and specifically relates to a ray tracing measurement correction method, system and electronic device, which can be used in systems such as electronic countermeasures and automated command, as well as electromagnetic wave propagation. Background Art

[0002] The ray tracing algorithm is a branch of the radio wave propagation calculation model. Based on geometric optics and the theory of uniform diffraction, this algorithm simulates and emulates the radio wave propagation path by constructing specific scene information. Radio wave propagation models mainly include empirical models and deterministic models. Empirical models are mathematical models fitted based on a large amount of measured data and can only be used in specific scenarios. However, for complex environments such as small urban areas, factors such as building occlusion will cause the prediction accuracy of empirical models to be greatly reduced. Among deterministic models, the ray tracing method is widely used. Compared with empirical models, the ray tracing algorithm can more accurately reflect the characteristics of the propagation environment and significantly improve the calculation accuracy. However, due to the complexity of the radio wave propagation environment, there are still many environmental factors that cannot be reflected through spatial modeling. Therefore, there is still a certain error between its simulation results and the measured values at the receiving point.

[0003] The patent document with the publication number CN116401940A discloses a radio wave propagation simulation method based on big data fitting. It mainly predicts the transmission loss in the scene by training a radio wave propagation model and replaces the ray tracing method with a big data fitting radio wave propagation simulation method to reduce the computational complexity. However, since it does not consider the measured values at the receiving point, it may lead to a large difference between the prediction results and the measured results in complex scenarios.

[0004] The patent document with the publication number CN115828740A discloses a transmission loss prediction method for a radio wave propagation model based on a neural network. This method fine-tunes the simulation training model by introducing a small amount of measured data, making the model more in line with the actual situation. However, since it does not consider the propagation scene information, it may lead to the problem of difficult radio wave propagation prediction in complex scenarios. Summary of the Invention

[0005] The purpose of the present invention is to address the deficiencies of the above-mentioned existing technologies and propose a ray tracing measurement correction method, system and electronic device based on multi-modal fusion to make full use of measured data and improve the radio wave propagation prediction accuracy by combining multi-modal data and measured data.

[0006] To achieve the above purpose, the technical solutions adopted by the present invention include:

[0007] 1. A ray tracing measurement correction method based on multi-modal fusion, characterized by including:

[0008] Obtain the training sample set R of the convolutional layer respectively 1 and the training sample set R of the self-attention layer 2 , and the test sample set E, where R 2 includes measured data;

[0009] Construct a multi-modal fusion network model H including a convolutional neural network module H 1 , a self-attention mechanism module H 2 and a fusion module H 3 , and the H 1 module is connected in parallel with the H 2 module and then cascaded with the H 3 module;

[0010] Use the training sample set to train the prediction network model H of multi-modal fusion;

[0011] Take the test sample set E as the input of the trained prediction network model H * to perform forward propagation, and obtain the predicted values of the measured received power of M - K radio wave propagations.

[0012] Furthermore, the obtaining of the training sample set R of the convolutional layer 1 is to splice the building height feature vector f 1 , the transmitting point position feature vector f 2 and the receiving point position feature vector f 3 in the channel dimension to generate a three-dimensional tensor R containing multi-modal information 1 ;

[0013] Furthermore, the obtaining of the training sample set R of the self-attention layer 2 is to first normalize the radio wave characteristic influence factors of the target points, the measured values of the auxiliary points and their influence factors, and then splice the M sets of target point vectors B after normalization with the N sets of auxiliary point vectors A respectively to obtain M spliced vectors C, and select K vectors from C as the training sample set R 2 ,

[0014] Furthermore, the obtaining of the test sample set E is to use the remaining M - K spliced vectors in the spliced vector C as the test sample set E.

[0015] Furthermore, the structures and functions of the modules in the multi-modal fusion network model H are as follows:

[0016] The convolutional neural network module H 1 , which is used to process map information, includes two consecutive convolutional-activation function-pooling layers, a flattening layer, a first fully connected-activation layer, and a first fully connected layer;

[0017] Self-attention mechanism module H 2 , which is used to process the radio wave influence factor information of the receiving point, and includes two self-attention - normalization layers, a second fully connected - activation function layer, and a second fully connected layer connected in sequence. The input and output ends of each self-attention layer are connected by a cross-layer identity path and input into the normalization layer;

[0018] Fusion module H 3 , which is used to fuse spatial information and radio wave influence factor information, and includes a splicing layer and a third fully connected layer;

[0019] The output of the above first fully connected layer and the output of the second fully connected layer are connected in parallel and then connected to the splicing layer to form the multi-modal fusion network model H.

[0020] 2. A ray tracing measured value correction system based on multi-modal fusion, comprising:

[0021] A data acquisition module, which is used to acquire ray tracing data and measured received power data, including radio wave propagation characteristic data such as the working frequency band, the positions of the transmitting point and the receiving point, and normalize them;

[0022] A network module, which is used to construct a multi-modal fusion network model, including a convolutional neural network module, a self-attention mechanism module, and a fusion module, which are respectively used to process map information, radio wave influence factor information, and fuse spatial information and radio wave propagation characteristics;

[0023] A training module, which is used to train the multi-modal fusion network model, including inputting data into the network, performing forward propagation, calculating the loss value, and iteratively updating the weights and parameters through backpropagation until the maximum number of iterations is reached or the loss value converges, to obtain the trained prediction network model;

[0024] A prediction module, which is used to predict the measured received power of an unknown receiving point by using the trained multi-modal fusion network model, so as to correct the ray tracing result to improve the accuracy.

[0025] 3. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the program, it implements any one of the steps of the above-mentioned ray tracing measured value correction method based on multi-modal fusion.

[0026] Compared with the prior art, the present invention has the following advantages:

[0027] First, since the present invention constructs a multi-modal fusion network, the prediction accuracy of radio wave propagation can be improved. By inputting the radio wave propagation influence factor data and map data into different network modules respectively, this network can process different types of data more specifically. Among them, the self-attention module is responsible for processing the radio wave propagation influence factors. By calculating the feature correlation and weight distribution, it can automatically identify the key features and suppress the secondary features, and can deeply explore the correlation between the target point and the auxiliary point. The map data is processed by the convolutional neural network module to learn the local patterns in the map, such as building occlusion and obstacle distribution, which can help the network understand the path distribution of electromagnetic waves in the propagation environment.

[0028] Second, based on the traditional ray tracing simulation data, the present invention adds the measured data of the application area, effectively solves the problem of large errors between the measured values and theoretical values of the received power at the receiving point in complex scenarios, makes up for the deficiencies of the traditional model, improves the prediction accuracy of the received power, and meets the application requirements of complex propagation environments. Brief Description of the Drawings

[0029] Figure 1 is the implementation flowchart of the ray tracing measurement correction method based on multi-modal fusion of the present invention;

[0030] Figure 2 is Figure 1 the structure diagram of the multi-modal fusion network constructed in

[0031] Figure 3 is the structural block diagram of the embodiment of the ray tracing measurement correction system based on multi-modal fusion of the present invention;

[0032] Figure 4 is the comparison diagram of the prediction results of the present invention and the traditional algorithm at 19 LOS points in the 5800 MHz band;

[0033] Figure 5 is the comparison diagram of the prediction results of the present invention and the traditional algorithm at 13 NLOS points in the 5800 MHz band Detailed Embodiments

[0034] The present invention will be further described below with reference to the drawings and specific embodiments.

[0035] Embodiment 1, a ray tracing measurement correction method based on multi-modal fusion.

[0036] Referring to Figure 1 , the implementation steps of this example are as follows:

[0037] Step 1, obtain the training sample set R of the convolutional layer 1 , the training sample set R of the self-attention layer 2 and the test sample set E.

[0038] 1.1) Obtain the convolutional layer training sample set R 1 :

[0039] Concatenate the building height feature vector f 1 , the transmitter location feature vector f 2 and the receiver location feature vector f 3 along the channel dimension to generate a three-dimensional tensor R containing multi-modal information 1 . Among them:

[0040] Each vector point of the building height feature vector f 1 represents the building height at a spatial position. If the vector point does not contain a building, its value is 0;

[0041] For the transmitter location feature vector f 2 , the vector point corresponding to the location of the transmitter and its surrounding vector points are set to 1, and the remaining vector points are set to 0;

[0042] For the receiver location feature vector f 3 , the vector point corresponding to the location of the receiver and its surrounding vector points are set to 1, and the remaining vector points are set to 0.

[0043] In this example, the map of the predicted area is divided into 587×512 grids, and the size of each grid is 2m×2m. Its

[0044] 1.2) Obtain the attention layer training sample R 2 :

[0045] Select the factors affecting radio wave propagation, including: transmission frequency, local coordinates of the receiver, distance between the transmitter and the receiver, ray tracing simulation power value, measured power value at the receiver, multi-path information, where 0 represents line-of-sight and 1 represents non-line-of-sight in the multi-path information;

[0046] Normalize the radio wave characteristic factors of the target points and the measured values and their influencing factors of the auxiliary points, and then concatenate the M sets of target point vectors B after normalization with the N sets of auxiliary point vectors A respectively to obtain M concatenated vectors C, and select K vectors from C as the attention layer training sample set R 2 , Among them:

[0047] The set of target point vectors B, the set of auxiliary point vectors A, and the concatenated vector C are respectively expressed as follows:

[0048] B={b m |1≤m≤M}, bm=[freqm,Losm,Xm,Ym,Zm,Distm,Pm simulation,0],

[0049] A={a nm |1≤n≤N,1≤m≤M}, anm=[freqnm,Losnm,Xnm,Ynm,Znm,Distnm,Pnm simulation,Pnm measurement],

[0050] C={c m |1≤m≤M}, c m =[a 1m ,a 2m ,...,a nm ,b m ]

[0051] Where b m is the mth predicted target point vector; a nm is the nth auxiliary point vector corresponding to the mth target point; freq m , Los m , X m , Y m , Z m 、Dist m , P m仿真 Respectively represent the normalized frequency, multipath information, local coordinates, transmitter-receiver distance, and ray tracing simulation power value of the target point; freq nm , Los nm , X nm , Y nm , Z nm 、Dist nm , P nm仿真 , P nm实测 They respectively represent the normalized frequency, multipath information, local coordinates, transmitter-receiver distance, ray tracing simulation power value, and measured received power value of the nth auxiliary point corresponding to the mth target point.

[0052] 1.3) Get the test sample set E:

[0053] The remaining MK concatenated vectors in the concatenated vector C are taken as the test sample set E.

[0054] In this example, it is assumed but not limited to M=636, N=3, and K=572.

[0055] Step 2: Construct a multimodal fusion network model H.

[0056] Reference Figure 2 , the implementation of this step includes the following:

[0057] 2.1) Establish a convolutional neural network module H consisting of two convolutional layers, three activation functions, two pooling layers, a flattening layer, and two fully connected layers 1, and the first convolutional layer, the first activation function, the first pooling layer, the second convolutional layer, the second activation function, the second pooling layer, the flattening layer, the first fully connected layer, the third activation function, and the second fully connected layer are connected in sequence to process map information, where:

[0058] In the first convolutional layer, the convolution operation uses 32 convolutional kernels of size 3×3;

[0059] In the second convolutional layer, the convolution operation uses 64 convolutional kernels of size 3×3;

[0060] Both the first pooling layer and the second pooling layer use max pooling, the pooling window size is set to 2×2, and the stride is 2; the number of neurons in the first fully connected layer is 256, and the number of neurons in the second fully connected layer is 1;

[0061] All three activation functions use ReLU.

[0062] 2.2) Establish a self-attention mechanism module H composed of two self-attention layers, two normalization layers, two fully connected layers, and one activation function 2 , and the first self-attention layer, the first normalization layer, the second self-attention layer, the second normalization layer, the first fully connected layer, the activation function, and the second fully connected layer are connected in sequence. The input and output ends of each self-attention layer are connected by a cross-layer identity path and input to the normalization layer to process the radio wave influence factor information of the receiving point, where:

[0063] The number of neurons in the first fully connected layer is 32;

[0064] The number of neurons in the second fully connected layer is 1;

[0065] The activation function uses Sigmoid,

[0066] In the self-attention layer, the input feature matrix R 2 is linearly transformed to generate the value matrix V = R 2 W V , the key matrix K = R 2 W K and the query matrix Q = R 2 W Q , and the output of the self-attention layer is:

[0067] Output = softmax(QK T )·V

[0068] In the formula, are weight matrices with three different parameters respectively, K T is the transpose matrix of K,; softmax is a layer normalization operation, d is the number of features of each target vector, and in this example, d = 8;

[0069] 2.3) Establish a fusion module H composed of sequentially connected splicing layers and fully connected layers 3 , which is used to fuse spatial information and radio wave influence factor information, where: the number of neurons in the fully connected layer is 1;

[0070] 2.4) After paralleling the convolutional neural network module H 1 and the self-attention mechanism module H 2 , cascade them with the H 3 module to form a multi-modal fusion network model H.

[0071] Step 3: Use the training sample set to train the multi-modal fusion prediction network model H.

[0072] 3.1) Set the maximum number of training times to T≥500. The prediction network model for the t-th training is denoted as H t , and its weight parameters are set as w t , and the bias parameters are set as b t , and let t = 1;

[0073] 3.2) Use the R 1 and R 2 in Step 1 as the inputs of the convolutional neural network module H 1 and the self-attention module H 2 respectively. After paralleling their outputs, pass them through the fusion module H 3 to output K measured power prediction values where is the e-th measured power prediction value;

[0074] 3.3) Calculate the loss value L t of H t using the mean square error function MSE, and calculate the network parameter gradient of H t through L t . Then, use the Adam optimizer to update the weight w t and the bias parameter b t with the network parameter gradient to obtain the prediction network model H t for this iteration, where:

[0075] Calculate the loss value L t of H t The formula is: In the formula is the true value of the e-th measured power;

[0076] Update the weight parameter w t and the bias b t , and the calculation formula is: P t∈{ω t ,b t}, where α is the learning rate and ε is a constant;

[0077] is the correction of v t , and v t = β 1 ·v t-1 +(1 - β 1 )·g t is the first - order moment estimation of the gradient of the H t network parameters, is the gradient at iteration t, and β 1 is the exponential decay rate of the first - order moment, is the t - th power of β 1 ;

[0078] is the correction of s t , is the second - order moment estimation of the gradient of the H t network parameters, and β 2 is the exponential decay rate of the second - order moment, are respectively the t - th powers of β 2 ;

[0079] 3.4) Repeat steps 3.2) - 3.3) until the maximum number of iterations is reached or the loss value converges to obtain the trained prediction network model H * ;

[0080] 3.5) Input the test sample set E into the trained prediction network model H * for forward propagation to obtain M - K predicted values of the measured received power of radio wave propagation

[0081] Example 2, a ray - tracing measured correction system based on multi - modal fusion.

[0082] Referring to Figure 3 , this example includes: a data acquisition module 1, a network module 2, a training module 3, and a prediction module 4.

[0083] The data acquisition module 1 is used to obtain ray - tracing data and measured received power data, including radio wave propagation characteristic data such as the working frequency band, the positions of the transmitting point and the receiving point, and normalize them to provide standardized input for the training module 3;

[0084] The network module 2 is used to construct a multi - modal fusion network model, including a convolutional neural network module, a self - attention mechanism module, and a fusion module, which are respectively used to process map information, radio wave propagation influence factor information, and fuse spatial information and radio wave propagation characteristics;

[0085] A training module 3 for training the network module 2, including inputting the data obtained by the data acquisition module 1 into the network module 2, performing forward propagation, calculating the loss value, and iteratively updating the weights and parameters through backpropagation until the maximum number of iterations is reached or the loss value converges, so as to obtain a trained prediction network model;

[0086] A prediction module 4 for predicting the measured received power of an unknown receiving point by using the prediction network model obtained by the training module 3, so as to correct the ray tracing result and improve the prediction accuracy of radio wave propagation.

[0087] Embodiment 3: An electronic device.

[0088] The electronic device provided by the present invention includes a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor, communication interface, and memory complete communication with each other through a communication bus. The processor can call the logical instructions in the memory to execute the method for correcting the ray tracing measurement by multi-modal fusion. The method includes: obtaining a training sample set and a test sample set; constructing a multi-modal fusion network model including a convolutional neural network module, a self-attention mechanism module, and a fusion module; training the prediction network model of multi-modal fusion by using the training sample set; and performing forward propagation by using the test sample set as the input of the trained prediction network model to obtain the predicted value of the measured received power of radio wave propagation.

[0089] The effect of the present invention can be further illustrated by the following simulation results.

[0090] I. Simulation experiment conditions

[0091] Use a GPU with the model number NVIDIA RTX 2080 to run the prediction model, and the prediction model adopts the prediction model H in Embodiment 1 * , select the mean squared error function MSE as the loss function, and select the prediction area size as 1174m×1024m.

[0092] II. Simulation experiment content and results

[0093] In Simulation Experiment 1, under the above conditions, the power of the line-of-sight type receiving point is predicted at the 5800 MHz frequency band by using the present invention and the traditional ray tracing method respectively, and the prediction results are compared. The results are as Figure 4 shown.

[0094] From Figure 4 it can be seen that the average error between the result obtained by the prediction network model of the present invention and the measured value is only 3 dB, while the average error between the traditional ray tracing method and the measured value is as high as 13 dB. Compared with the traditional ray tracing method, the prediction accuracy of the present invention is improved by 10 dB.

[0095] Simulation experiment 2. Under the above conditions, the power of the non-line-of-sight type receiving point is predicted at the 5800 MHz frequency band by using the present invention and the traditional ray tracing method respectively, and the prediction results are compared. The results are as Figure 5 shown.

[0096] As can be seen from Figure 5 the figure, the average error between the result obtained by the prediction network model of the present invention and the measured value is only 2.44 dB, while the average error between the traditional ray tracing method and the measured value is as high as 17.44 dB. Compared with the traditional ray tracing method, the prediction accuracy of the present invention is improved by 15 dB.

[0097] The simulation results show that the prediction network model of the present invention effectively solves the problem of large error between the measured value and the theoretical value of the receiving point power in complex scenarios, makes up for the deficiencies of the traditional model, improves the prediction accuracy of the receiving power, and meets the application requirements of complex propagation environments.

[0098] The above description is only several specific examples of the present invention and does not constitute any limitation to the present invention. Obviously, for professionals in the field, after understanding the content and principle of the present invention, various modifications and changes in form and details may be made without departing from the principle and structure of the present invention. However, these corrections and changes based on the idea of the present invention are still within the protection scope of the claims of the present invention.

[0099] It should be noted that the step numbers in the specification and claims of the present invention are only for a clear description of the embodiments of the present invention for easy understanding, and their sequence numbers are not limited.

Claims

1. A ray tracing measurement correction method based on multi-modal fusion, characterized in that: include: Obtain the convolutional layer training sample set R1, the self-attention layer training sample set R2, and the test sample set E respectively, where R2 includes measured data; Constructing a multimodal fusion network model H including a convolutional neural network module H1, a self-attention mechanism module H2 and a fusion module H3, wherein the H1 module is connected in parallel with the H2 module and then cascaded with the H3 module; The multimodal fusion prediction network model H is trained using the training sample set; The test sample set E is used as the trained prediction network model H * The input is forward propagated to obtain MK predicted values ​​of the actual received power of radio wave propagation.

2. The method according to claim 1, characterized in that: The convolutional layer training sample set R1 is obtained by concatenating the three feature vectors of the building height feature vector f1, the transmitting point position feature vector f2 and the receiving point position feature vector f3 in the channel dimension to generate a three-dimensional tensor R1 containing multimodal information; The self-attention layer training sample set R2 is obtained by first normalizing the influence factors of the target point's electric wave characteristics and the measured values ​​of the auxiliary points and their influence factors, and then splicing the normalized M target point vector sets B with the N auxiliary point vector sets A to obtain M spliced ​​vectors C, and selecting K vectors in C as the training sample set R2. The method of obtaining the test sample set E is to use the remaining MK splicing vectors in the splicing vector C as the test sample set E.

3. The method according to claim 2, characterized in that The convolutional layer training sample includes dividing the map of the predicted area into 587×512 grids, and the size of each grid is 2m×2m.

4. The method according to claim 2, characterized in that: The parameter settings of each feature vector in the convolutional layer training sample include the following: The building height feature vector f1, each vector point of which represents the building height at a spatial position, and if the vector point does not contain a building, the value is 0; The emission point position feature vector f2 is to set the vector point corresponding to the emission point position and its surrounding vector points to 1, and the remaining vector points to 0; The receiving point position feature vector f3 sets the vector point corresponding to the receiving point position and its surrounding vector points to 1, and sets the remaining vector points to 0.

5. The method according to claim 2, characterized in that: The target point vector set B, the auxiliary point vector set A and the splicing vector C are respectively expressed as follows: B={b m |1≤m≤M},A={a nm |1≤n≤N,1≤m≤M}, C={c m |1≤m≤M},cm=[a1m,a2m,...,anm,bm]; Among them, b m is the mth predicted target point vector, a nm is the nth auxiliary point vector corresponding to the mth target point.

6. The method according to claim 1, characterized in that The structure and function of each module in the multimodal fusion network model H are as follows: The convolutional neural network module H1 is used to process map information, and includes two convolution-activation function-pooling layers, a flattening layer, a first fully connected-activation layer, and a first fully connected layer connected in sequence; The self-attention mechanism module H2 is used to process the radio wave influence factor information of the receiving point, which includes two self-attention-normalization layers, a second fully connected-activation function layer and a second fully connected layer connected in sequence, and the input and output ends of each self-attention layer are connected by an identical path across layers and input to the normalization layer; The fusion module H3 is used to fuse the spatial information and the radio wave impact factor information, and includes a splicing layer and a third fully connected layer; The output of the first fully connected layer and the output of the second fully connected layer are connected in parallel and then connected to the concatenation layer to form a multimodal fusion network model H.

7. The method according to claim 6, characterized in that: The activation function in the first fully connected-activated layer is ReLU, and the number of fully connected neurons is 256; The activation function in the second fully connected-activated layer is Sigmoid, and the number of fully connected neurons is 32; The number of neurons in the first fully connected layer is 1; The number of neurons in the second fully connected layer is 1.

8. The method according to claim 1, characterized in that The multimodal fusion prediction network model H is trained using the training sample set, and its implementation includes the following: 8a) Set the maximum number of training times to T ≥ 500; the prediction network model of the tth training is denoted as H t , whose weight parameter is w t , the bias parameter is b t , and let t = 1; 8b) R1 and R2 are used as inputs of the convolutional neural network module H1 and the self-attention module H2 respectively, and their outputs are connected in parallel and sent to the fusion module H3 to obtain K measured power prediction values 8c) H calculated using the mean square error function MSE t Loss value L t , and through L t Calculate H t The network parameter gradient is then used to optimize the weight w by using the network parameter gradient. t and the bias parameter b t Update to get the prediction network model H for this iteration t ; 8d) Repeat steps 8b)-8c) until the maximum number of iterations is reached or the loss value converges to obtain a trained prediction network model.

9. A ray tracing measurement correction system based on multi-modal fusion, comprising: The data acquisition module is used to obtain ray tracing data and measured received power data, including radio wave propagation characteristic data such as working frequency band, transmitting point and receiving point position, and normalize them; The network module is used to build a multimodal fusion network model, including a convolutional neural network module, a self-attention mechanism module, and a fusion module, which are used to process map information, radio wave influencing factor information, and integrate spatial information with radio wave propagation characteristics respectively; The training module is used to train the multimodal fusion network model, including inputting data into the network, performing forward propagation, calculating the loss value, and iteratively updating the weights and parameters through back propagation until the maximum number of iterations is reached or the loss value converges, thereby obtaining the trained prediction network model; The prediction module is used to predict the measured received power of unknown receiving points using the trained multimodal fusion network model, and to correct the ray tracing results to improve the accuracy.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the ray tracing measurement correction method for multimodal fusion as described in any one of claims 1 to 8 is implemented.

Images