Radio wave propagation simulation device

Abstract

To provide a radio wave propagation simulation device which improves prediction precision by using a prediction result obtained by ensemble learning of multiple types of radio wave propagation models.SOLUTION: A radio wave propagation simulation device uses a value obtained by synthesizing predicted values of received power calculated by a plurality of existing radio wave propagation models as an estimated equation predicted value of the received power, among the plurality of predicted values of the received power as learning data. A learning model construction unit 4 performs simulation of radio wave propagation by using the learning data, and a machine learning model constructed by machine-learning actual measurement data of the received power of a corresponding coordinate as correct answer data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a radio wave propagation simulation device that simulates radio wave propagation, and more particularly to a radio wave propagation simulation device that can improve prediction accuracy through ensemble learning of a radio wave propagation model. [Background technology]

[0002] [Prior Art] Conventionally, radio wave propagation models are used to simulate radio wave propagation and determine installation conditions for wireless devices. However, even if a relatively accurate model is used in this radio wave propagation model, some degree of error will occur.

[0003] [Related Technology] As a related prior art, there is Japanese Patent Application Laid-Open No. 2011-033583 entitled "Radio wave propagation estimation system, radio wave propagation estimation method, and radio wave propagation estimation program" (Patent Document 1). Patent Document 1 shows that radio wave propagation is estimated, and that the accuracy of estimation is improved by estimating propagation loss based on topographical information and the height of topographical objects between the transmitting point and the receiving point. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Laid-Open No. 2011-033583 Summary of the Invention [Problem to be solved by the invention]

[0005] However, in conventional radio wave propagation simulations, errors occur in the radio wave propagation models used to design the installation of wireless equipment, which means that the desired received power cannot be achieved after installation, and the installation conditions of the wireless equipment must be reviewed.

[0006] Incidentally, Patent Document 1 does not describe a configuration for improving prediction accuracy by utilizing prediction results obtained by ensemble learning of multiple types of radio wave propagation models.

[0007] The present invention has been made in consideration of the above-described circumstances, and aims to provide a radio wave propagation simulation device that improves prediction accuracy by utilizing prediction results obtained by ensemble learning of multiple types of radio wave propagation models. [Means for solving the problem]

[0008] The present invention, which aims to solve the problems of the above-mentioned conventional example, is a radio wave propagation simulation device that simulates radio wave propagation, and uses a machine learning model constructed by inputting a plurality of predicted values ​​of received power for each coordinate as learning data and using actual measured data of received power for the corresponding coordinate as correct answer data to perform machine learning. and a received power estimation formula predicted value obtained by combining a plurality of received power prediction values ​​calculated using a plurality of existing radio wave propagation models is used as learning data. It is characterized by:

[0010] The present invention is characterized in that in the radio wave propagation simulation device, the learning data uses ray tracing predicted values ​​of received power for each coordinate, the average delay time until reaching the predicted point, the distance to the predicted point, and visibility information on whether the predicted point is visible.

[0011] The present invention is characterized in that, in the radio wave propagation simulation device, a machine learning model is input with an estimation formula predicted value of a new received power for each coordinate, a ray tracing predicted value of a new received power, an average delay time until reaching the predicted point, a distance to the predicted point, and visibility information as to whether the predicted point is visible, and the machine learning model infers a predicted value of the received power for each coordinate and outputs the prediction result. [Effects of the Invention]

[0012] According to the present invention, a radio wave propagation simulation device that simulates radio wave propagation inputs a plurality of predicted values ​​of received power for each coordinate as learning data, and performs machine learning using a machine learning model that is constructed by using actual measured data of received power at the corresponding coordinate as correct answer data. and a received power estimation formula predicted value obtained by combining a plurality of received power prediction values ​​calculated using a plurality of existing radio wave propagation models is used as learning data. Since it is a radio wave propagation simulation device, it has the effect of improving prediction accuracy. [Brief explanation of the drawings]

[0013] [Figure 1] FIG. 2 is a block diagram of the configuration of the device. [Figure 2] FIG. 1 is a schematic diagram illustrating an example of explanatory variables. [Figure 3] FIG. 10 is a flowchart showing the processing of a preprocessing unit. [Figure 4] FIG. 10 is a flowchart showing the processing of an inference unit. DETAILED DESCRIPTION OF THE INVENTION

[0014] An embodiment of the present invention will be described with reference to the drawings. [Outline of the embodiment] A radio wave propagation simulation device (this device) according to an embodiment of the present invention simulates radio wave propagation using a machine learning model constructed by inputting multiple predicted values ​​of received power for each coordinate as learning data and using actual measured data of received power at the corresponding coordinate as ground truth data. Of the multiple predicted values ​​of received power used as learning data, a combination of predicted values ​​of received power calculated using multiple existing radio wave propagation models is used as the estimated formula predicted value of received power. Therefore, prediction accuracy can be improved by utilizing prediction results obtained by ensemble learning of multiple types of radio wave propagation models.

[0015] [This device: Figure 1] This device will be described with reference to Figure 1. Figure 1 is a block diagram of the device. As shown in FIG. 1, the present device basically includes a target variable storage unit 1, a simulation execution unit 2, an explanatory variable storage unit 3, and a learning model construction unit 4. The learning model construction unit 4 includes a preprocessing unit 5, a learning unit 6, and an inference unit .

[0016] [Parts of this device] Each part of this device will now be described in detail. [Objective variable storage unit 1] The objective variable storage unit 1 stores actual measurement data as correct answer data for constructing a machine learning model, and specifically stores data on received power values ​​(RSSI: Received Signal Strength Indicator) [dBm] collected using a measuring instrument or the like as objective variables.

[0017] [Simulation execution part 2] The simulation execution unit 2 executes a simulation based on an existing radio wave propagation model using an existing area simulator or the like, and calculates a predicted value of the received power value. Specifically, predicted values ​​are calculated using existing radio wave propagation models, which are estimated indoor radio wave propagation models: the "Multi Wall Model" (first existing model), the "Motley Keenan Model" (second existing model), and the "One Slope Model" (third existing model).

[0018] Furthermore, the simulation execution unit 2 calculates a predicted value (RSSI_ray trace predicted value) by the ray trace method. The ray tracing method is a method of searching for propagation paths by treating radio waves as light, and by tracing the radio waves as they travel from the transmitting point to the receiving point based on the theory of geometrical optics, it calculates propagation loss (received level), delay time, direction of emission, and direction of arrival.The calculation is constrained by the number of reflections, transmissions, and diffractions.

[0019] Furthermore, the simulation execution unit 2 calculates or sets data on the delay time for multiple rays (20 rays per point) calculated by the ray tracing method to reach the predicted point, the coordinates of the predicted point, and visibility information indicating whether or not it is visible in a straight line (line of sight [visible] / out of sight [not visible]).

[0020] [Explanatory variable memory section 3] The explanatory variable memory unit 3 stores, as learning data for explanatory variables, predicted value data of an existing radio wave propagation model calculated by the simulation execution unit 2, predicted value data of the ray tracing method (RSSI_ray tracing predicted value), data of each delay time until reaching the predicted point, data of the coordinates of the predicted point, and data of visibility information. Here, the "RSSI_ray tracing predicted value" is the sum of multiple rays. Details of the explanatory variables will be described later.

[0021] [Learning Model Construction Part 4] The learning model construction unit 4 includes a pre-processing unit 5 that performs pre-processing to shape the features of the learning data so that it can be used as input data for a machine learning model, a learning unit 6 that constructs a learning model using the shaped features, and an inference unit 7 that performs radio wave propagation simulation using the constructed learning model and infers the prediction results.

[0022] Each part of the learning model construction unit 4 will now be described in detail. [Preprocessing section 5] The preprocessing unit 5 shapes the feature quantities of the data stored in the explanatory variable storage unit 3 so that the data can be used as input data for the machine learning model.

[0023] Specifically, first, the preprocessing unit 5 combines the predicted values ​​calculated using the existing radio wave propagation models (the first to third existing models). The three predicted values ​​basically produce similar prediction results and are therefore highly correlated with each other, so combining them effectively realizes ensemble learning. To combine them, the weights of each predicted value are calculated using linear regression or similar methods, and then combined into a single value called "RSSI_Estimation Formula Prediction Value."

[0024] Secondly, the preprocessing unit 5 obtains the "average delay time" by arithmetically averaging the delay times of a plurality of rays (approximately 20 rays) calculated by the ray tracing method until they reach the predicted point. Radio waves will arrive from the coordinates of the transmitting antenna to the coordinates of the predicted location via multiple predicted paths, and the average delay time for the shortest arrival time is calculated. Furthermore, the preprocessing unit 5 calculates the "distance" between two points from the coordinates of each predicted point and the coordinates of the transmitting antenna. The feature quantities of the data obtained in the preprocessing unit 5 are as shown in Fig. 2. The details of Fig. 2 will be described later.

[0025] [Study Section 6] The learning unit 6 uses LightGBM (Light Gradient Boosting Machine / LGBM) as a machine learning model, inputs the feature data shaped by the preprocessing unit 5 and the correct answer data from the objective variable storage unit 1, performs machine learning, and constructs a learning model. LGBM is a learning method based on a decision tree algorithm. Since the learning unit 6 uses LGBM, the preprocessing unit 5 does not standardize the feature amounts or interpolate missing values.

[0026] [Inference part 7] The inference unit 7 uses the learning model constructed by the learning unit 6 to infer predictions, inputs the newly acquired feature data shown in Figure 2, performs radio wave propagation simulation, and outputs prediction results. The prediction result is output as a predicted RSSI value for each coordinate of the location where the transmitting antenna is planned to be installed.

[0027] [Explanatory variables: Figure 2] Next, the explanatory variables used in this device will be described with reference to Fig. 2. Fig. 2 is a schematic diagram showing examples of explanatory variables. The explanatory variables are stored in the pre-processing unit 5 in a table format, and as shown in Figure 2, data on the RSSI_estimation formula predicted value, RSSI_ray trace predicted value, average delay time, distance, and line of sight are stored corresponding to the coordinates of the transmitting antenna to be installed. As described above, the preprocessing unit 5 receives data from the explanatory variable storage unit 3, shapes the feature quantities, and generates the explanatory variables shown in FIG.

[0028] The RSSI_estimation formula predicted value is a composite value of predicted values ​​calculated using the above-mentioned existing radio wave propagation models (existing models 1 to 3), and the RSSI_ray trace predicted value (total value of multiple rays), average delay time, distance, and visibility data are calculated using the ray tracing method. Note that visibility is set to "1" if the predicted point is visible from the transmitting antenna, and "0" if not.

[0029] In other words, in Figure 2, the coordinates of the transmitting antenna to be installed are set on the vertical axis, and the RSSI_estimation formula predicted value, RSSI_ray trace predicted value, average delay time, distance, and line of sight data corresponding to those coordinates are set on the horizontal axis. Then, one horizontal line for the coordinates is input to the learning model of the learning unit 6 in the case of machine learning, and is input to the trained model of the inference unit 7 in the case of inference.

[0030] [Preprocessing section: Figure 3] The processing of the pre-processing unit 5 will be described with reference to Fig. 3. Fig. 3 is a flow chart showing the processing of the pre-processing unit. As shown in FIG. 3, the preprocessing unit 5 combines predicted values ​​calculated using the existing radio wave propagation models (first to third existing models) for each coordinate of the transmitting antenna to calculate an "RSSI_estimation formula predicted value" (S1).

[0031] Next, the preprocessing unit 5 acquires the "RSSI_ray tracing predicted value" calculated by the ray tracing method from the explanatory variable storage unit 3 in correspondence with the coordinates (S2). Then, the preprocessing unit 5 calculates the average of the delay times of the rays from the coordinates calculated by the ray tracing method until they reach the predicted point as the "average delay time" (S3).

[0032] Next, the preprocessing unit 5 calculates the "distance" between the two points from the coordinates of each predicted point and the coordinates of the transmitting antenna (S4). Then, the preprocessing unit 5 acquires visibility data corresponding to the coordinates from the explanatory variable storage unit 3 (S5). In this way, the feature data as learning data is shaped to obtain the explanatory variable data shown in Figure 2.

[0033] [Inference processing: Figure 4] The processing of the inference unit 7 will be described with reference to Fig. 4. Fig. 4 is a flow chart showing the processing of the inference unit. As shown in Figure 4, the inference unit 7 inputs the feature data of Figure 2 obtained by newly executing a simulation (at different coordinates) into the learning model constructed by the learning unit 6 in order to actually infer a prediction (S11). That is, the simulation execution unit 2 executes a simulation at a new time, the explanatory variable storage unit 3 stores the simulation result data, the preprocessing unit 5 formats the feature data, and the data is input to the inference unit 7.

[0034] Next, the inference unit 7 performs a radio wave propagation simulation using the input feature data in the learning model constructed by the learning unit 6, and outputs the prediction results (predicted RSSI values ​​for each coordinate of the location where the transmitting antenna is planned to be installed) (S12). The prediction results will be useful for optimizing the placement of wireless devices.

[0035] [Effects of the embodiment] According to this device, the simulation execution unit 3 calculates multiple predicted values ​​of received power for each coordinate and inputs them as learning data to the learning model construction unit 4. The learning model construction unit 4 then simulates radio wave propagation using a machine learning model constructed by machine learning the actual measured data of the received power at the corresponding coordinates as correct answer data. Of the multiple predicted values ​​of received power used as learning data, a combination of predicted values ​​of received power calculated using multiple existing radio wave propagation models is used as the estimated formula predicted value of received power. This has the effect of enabling the prediction accuracy to be improved by utilizing the prediction results obtained by ensemble learning of multiple types of radio wave propagation models. [Industrial Applicability]

[0036] The present invention is suitable for a radio wave propagation simulation device that improves prediction accuracy by utilizing prediction results obtained by ensemble learning of multiple types of radio wave propagation models. [Explanation of symbols]

[0037] 1...Objective variable storage unit, 2...Simulation execution unit, 3...Explanatory variable storage unit, 4...Learning model construction unit, 5...Preprocessing unit, 6...Learning unit, 7...Inference unit

Claims

1. A radio wave propagation simulation device that simulates radio wave propagation, A machine learning model is constructed by inputting multiple predicted values ​​of received power for each coordinate as learning data and using actual measured data of received power at the corresponding coordinate as correct answer data, and a radio wave propagation simulation is performed using the model. A radio wave propagation simulation device characterized in that a received power estimation formula predicted value obtained by combining a plurality of received power predicted values ​​calculated using a plurality of existing radio wave propagation models is used as the learning data.

2. 2. The radio wave propagation simulation device according to claim 1, wherein the learning data uses, in addition to the estimation-formula predicted value of the received power, a ray-trace predicted value of the received power for each coordinate, an average delay time until reaching the predicted point, a distance to the predicted point, and line-of-sight information indicating whether the predicted point is visible.

3. The radio wave propagation simulation device according to claim 2, characterized in that the machine learning model is input with an estimation-formula predicted value of a new received power for each coordinate, a ray-trace predicted value of a new received power, an average delay time until reaching the predicted point, a distance to the predicted point, and visibility information indicating whether the predicted point is visible, to infer a predicted value of received power for each coordinate, and output a prediction result.

Images