CN113678468B - Machine learning device - Google Patents
Machine learning device Download PDFInfo
- Publication number
- CN113678468B CN113678468B CN202080027864.6A CN202080027864A CN113678468B CN 113678468 B CN113678468 B CN 113678468B CN 202080027864 A CN202080027864 A CN 202080027864A CN 113678468 B CN113678468 B CN 113678468B
- Authority
- CN
- China
- Prior art keywords
- radio wave
- state
- learning
- machine learning
- wave propagation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000010801 machine learning Methods 0.000 title claims abstract description 82
- 238000011156 evaluation Methods 0.000 claims description 20
- 230000006870 function Effects 0.000 claims description 17
- 238000013507 mapping Methods 0.000 claims description 6
- 238000005259 measurement Methods 0.000 description 17
- 238000004088 simulation Methods 0.000 description 15
- 238000000034 method Methods 0.000 description 8
- 230000005540 biological transmission Effects 0.000 description 7
- 238000009434 installation Methods 0.000 description 7
- 238000004378 air conditioning Methods 0.000 description 6
- 238000013461 design Methods 0.000 description 6
- 238000004891 communication Methods 0.000 description 5
- 238000009423 ventilation Methods 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000012417 linear regression Methods 0.000 description 3
- 239000002184 metal Substances 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 238000002790 cross-validation Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000010438 heat treatment Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 229910000831 Steel Inorganic materials 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 229910052602 gypsum Inorganic materials 0.000 description 1
- 239000010440 gypsum Substances 0.000 description 1
- 230000000116 mitigating effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
Landscapes
- Air Conditioning Control Device (AREA)
- Selective Calling Equipment (AREA)
Abstract
A computer (10) is a machine learning device for learning the radio wave propagation state between a BLE module and another BLE module, and is provided with an acquisition unit (20) and a learning unit (30). An acquisition unit (20) acquires air conditioner arrangement information (21) and beam arrangement information (22) as information for acquiring state variables. These information are information about the BLE module and between other BLE modules. A learning unit (30) learns a state variable by associating the state variable with the radio wave propagation state between the BLE module and another BLE module.
Description
Technical Field
The present invention relates to a machine learning apparatus that learns a radio wave propagation state between a radio device and another radio device.
Background
Conventionally, there is a system in which equipment (hereinafter, HVAC equipment) for heating, ventilation, air conditioning, and the like is connected to a network and managed remotely. In this system, it is necessary to store a network address in association with information on the physical configuration of each HVAC equipment in a building. This association is often performed manually in a building, and the work takes a lot of time and cost.
In response to this problem, in recent years, there has been an effort to provide radio wave transceivers (radio devices) for respective HVAC devices, obtain information on radio wave intensities from the radio wave transceivers, and estimate the arrangement of the respective HVAC devices based on the information. Regarding simulation of a radio wave propagation state, for example, patent document 1 (japanese patent application laid-open publication 2016-208265) shows 1 methods.
Disclosure of Invention
Problems to be solved by the invention
Although the simulation of the radio wave propagation state can be performed by the conventional method, a simpler method or a more accurate method is required unlike the conventional method.
Means for solving the problems
The machine learning device of the first aspect is a machine learning device that learns a radio wave propagation state between a wireless device and another wireless device. The machine learning device includes an acquisition unit and a learning unit. The acquisition unit acquires the first information as information for acquiring a state variable. The first information is information about the wireless device and other wireless devices. For example, the first information includes information on a distance between the wireless device and another wireless device, information on an article located between the wireless device and another wireless device, and the like. The learning unit learns the state variable in association with the radio wave propagation state between the radio device and the other radio devices.
Here, the acquisition unit acquires first information on the wireless device and other wireless devices, and thus can obtain necessary state variables from the first information. Since the learning unit learns the state variable in association with the radio propagation state, the radio propagation state between the radio device and another radio device can be obtained by the machine learning apparatus.
The radio wave propagation state between the radio device and the other radio device can be expressed by any one of the attenuation amount of the radio wave between the radio device and the other radio device, the minimum transmission radio wave intensity that the other radio device can receive, the reception radio wave intensity of the other radio device for transmission from the radio device at a predetermined transmission radio wave intensity, and the like.
The machine learning apparatus of the second aspect is the machine learning apparatus of the first aspect, wherein the first information includes information of a distance between the wireless device and the other wireless device. The distance of the wireless device from other wireless devices is at least 1 of the state variables.
Here, since the distance between 2 wireless devices is used as the state variable, a radio wave propagation state with higher accuracy can be obtained.
The machine learning device according to the third aspect is the machine learning device according to the first aspect or the second aspect, wherein the first information includes article information on a predetermined article located between the wireless device and the other wireless device.
Here, the state variable can be obtained from information of the article located between the two wireless devices that affects the radio wave propagation state. This can obtain a radio wave propagation state with higher accuracy.
The machine learning device according to the fourth aspect is the machine learning device according to the third aspect, wherein the article information includes information on a predetermined number of articles.
Here, the state variable is obtained from information of the number of predetermined articles, which is relatively simple information. Therefore, the radio wave propagation state can be obtained more simply.
The machine learning device according to the fifth aspect is the machine learning device according to the third or fourth aspect, wherein the predetermined article is an air conditioner and/or a beam disposed in a space above the ceiling.
Here, in a narrow space on the ceiling where it is difficult to obtain a result with good accuracy in the conventional radio wave propagation simulator, a radio wave propagation state with relatively good accuracy can be obtained by the learning unit.
The machine learning device according to the sixth aspect is the machine learning device according to any one of the third to fifth aspects, wherein the article information includes information on a predetermined article size.
Here, information about the size of a predetermined object located between the wireless device and another wireless device is acquired as the first information. This can obtain a radio wave propagation state with higher accuracy.
The machine learning device according to the seventh aspect is the machine learning device according to any one of the third to sixth aspects, wherein the article information includes information on a position of a predetermined article.
Here, information on the position of a predetermined article located between the wireless device and another wireless device is acquired as first information. This can obtain a radio wave propagation state with higher accuracy.
The machine learning device according to the eighth aspect is the machine learning device according to any one of the third to seventh aspects, wherein the article information includes information on a predetermined orientation of the article.
Here, information on the orientation of a predetermined object located between the wireless device and another wireless device is acquired as the first information. This can obtain a radio wave propagation state with higher accuracy.
The machine learning device according to the ninth aspect is the machine learning device according to the first aspect or the second aspect, wherein the learning unit performs learning by associating the state variable with the radio wave propagation state based on the learning data set. The learning data set is composed of a result obtained by actually measuring the radio wave propagation state and a state variable at the time of actual measurement of the radio wave propagation state.
Here, the actual measurement of the radio wave propagation state is performed, and the learning unit learns using a learning data set composed of the state variable and the result at that time. This can obtain a radio wave propagation state with higher accuracy.
A tenth aspect of the machine learning device is the machine learning device according to any one of the third to eighth aspects, wherein the learning unit performs learning by associating a state variable with a radio wave propagation state based on the learning data set. The learning data set is composed of a result obtained by actually measuring the radio wave propagation state and a state variable obtained from the article information at the time of actually measuring the radio wave propagation state.
Here, the actual measurement of the radio wave propagation state is performed, and the learning unit learns using a learning data set composed of the state variable and the result at that time. This can obtain a radio wave propagation state with higher accuracy.
In the machine learning device according to the eleventh aspect, the learning unit learns the state variable and the radio wave propagation state by adjusting the coefficient of the linear model of the following equation 1 in the machine learning device according to the first aspect, the second aspect, or the ninth aspect.
Formula 1:
Wherein,
Is the result of the estimation of the propagation state of the electric wave,
W is a coefficient of the degree of freedom,
X is the distance of the wireless device from other wireless devices.
A twelfth aspect of the machine learning device is the machine learning device according to any one of the third to eighth aspects or the tenth aspect, wherein the learning unit learns the state variable in association with the radio wave propagation state by adjusting the coefficient of the linear model of the following equation 2.
Formula 2:
Wherein,
Is the result of the estimation of the propagation state of the electric wave,
W is a coefficient of the degree of freedom,
X is the distance of the wireless device from other wireless devices,
X' is the number of specified items that are between the wireless device and the other wireless devices.
The machine learning device according to the thirteenth aspect is the machine learning device according to any one of the first to twelfth aspects, further comprising an output unit and an update unit. The output unit outputs the estimation result of the radio wave propagation state. The updating unit updates the learning state of the learning unit by evaluating the difference between the estimation result of the radio wave propagation state and the result obtained by actually measuring the radio wave propagation state.
Here, the learning state of the learning unit is improved, and a radio wave propagation state with higher accuracy can be obtained.
The machine learning device according to the fourteenth aspect is the machine learning device according to the eleventh aspect, further comprising an output unit and an update unit. The output unit outputs the estimation result of the radio wave propagation state. The updating unit updates the coefficient w in the above-described expression 1 so as to reduce the following evaluation function 1, thereby updating the learning state of the learning unit.
Evaluation function 1:
Wherein,
Is the result of the estimation of the propagation state of the electric wave,
Y is a result obtained by actually measuring the radio wave propagation state.
Here, the learning state of the learning unit is improved, and a radio wave propagation state with higher accuracy can be obtained.
The machine learning device according to a fifteenth aspect is the machine learning device according to the twelfth aspect, further comprising an output unit and an update unit. The output unit outputs the estimation result of the radio wave propagation state. The updating unit updates the learning state of the learning unit by updating the coefficient w in the above-described expression 2 so as to reduce the following evaluation function 2.
Evaluation function 2:
Wherein,
Is the result of the estimation of the propagation state of the electric wave,
Y is a result obtained by actually measuring the propagation state of the radio wave,
Alpha is a regularization parameter.
Here, the learning state of the learning unit is improved, and a radio wave propagation state with higher accuracy can be obtained.
Drawings
Fig. 1 is a simple vertical sectional view showing a building in which a plurality of BLE modules for learning a radio wave propagation state by a machine learning device are arranged, and a beam and an air conditioner in a space existing on a ceiling of the room.
Fig. 2 is a plan view of a 1-floor portion of a building including a beam and an arrangement of air conditioners in a space existing on a ceiling.
FIG. 3 is a block diagram of an HVAC management system including a machine learning device.
Detailed Description
(1) Summary of HVAC management system
(1-1) Background of the need for an HVAC management System
With the increasing demands for energy saving and rapid development of intelligent control technology for buildings (office buildings, commercial facilities, etc.), intelligent systems for air conditioning and ventilation are being installed in the buildings. The system can provide feedback control corresponding to the need, such as adjusting the set temperature of the air conditioner based on the number of people in the room, by monitoring the temperature, humidity, CO 2 concentration, occupancy of the room, and the like.
In a specific region, in order to realize such feedback control, it is necessary to connect HVAC equipment and sensors that perform ventilation and air conditioning to a network, and map (map) a network address of the HVAC equipment or the like to a physical location of the HVAC equipment or the like. The mapping operation, which is also called address setting, has heretofore been performed by a manual operation. The work performed by the operator on site includes a work performed by using the control device to operate HVAC equipment 1 at a time, and a work of writing an address of HVAC equipment or the like displayed on a display of the control device on the layout.
These operations take time and a correction operation due to an artificial error is generated. In addition, the labor cost involved in the work performed by the operator on site becomes a large cost. For example, in the case of a large-scale building having a floor area of 50000 square meters, even if 2 operators perform work, the address setting requires 3 months.
(1-2) Basic ideas related to map job mitigation for HVAC management systems
In order to suppress the load of the mapping (address setting) operation, in the present embodiment, an HVAC management system for performing automatic mapping will be described. In this HVAC management system, as shown in fig. 1, an air conditioner a as HVAC equipment is provided with a BLE (Bluetooth Low Energy: bluetooth low energy) module M, and the RSSI (received signal strength indicator) of the BLE module M is used. Here, the installation position of the BLE modules M provided in the air conditioner a is estimated, and the radio wave propagation state between the BLE modules M is estimated. When estimating the radio wave propagation state between the respective BLE modules M, mapping can be easily performed based on these estimated values.
Specifically, the air conditioner a is equipped with BLE modules M, and performs packet (packet) communication with each other, thereby measuring the actual received signal strength of the packet transmitted from the transmitting side BLE module M to the receiving side BLE module M. On the other hand, the physical arrangement of the air conditioners a, the distance between the air conditioners a, the arrangement of obstacles that may affect the propagation of radio waves (signals), and the like are automatically read from the layout of the air conditioners a extracted from the design drawing (see fig. 3) provided by the designer of the building or the like. The radio propagation state between the BLE module M and the other BLE modules M, for example, the radio attenuation between the two modules M, the ratio of the received signal strength of the receiving side BLE module M to the transmitted signal strength of the transmitting side BLE module M, and the like can be estimated by using the linear model 35 (see fig. 3) of the learning unit 30. If the information of the arrangement of each air conditioner a and the obstacle read from the layout of each air conditioner a and the linear model 35 of the learning unit 30 trained by the huge learning data set obtained by sampling in the past in various buildings are used, the radio wave propagation state between the BLE module M and the other BLE modules M can be estimated with high accuracy.
As a result, the network address can be automatically mapped to the air conditioner a in the space provided on the ceiling of the building. This can reduce the time required for the initial setting operation of the air conditioning/ventilation intelligent system (HVAC management system) in the building and reduce the labor cost.
(2) Structure of HVAC management system
The HVAC management system is a system that manages HVAC equipment for heating, ventilation, air conditioning, and the like, but here, an air conditioner a as HVAC equipment is taken as an example, and description is made with reference to fig. 1 to 3.
(2-1) Installation site of air conditioner as HVAC Equipment
As shown in fig. 1, the air conditioner a is an air conditioning indoor unit provided in an interior space of a building 81. The plurality of air conditioners a are disposed in spaces on the ceiling of each floor (room) of the building 81. In fig. 1,3 air conditioners A1, A2, A3 are shown which are provided in a space S above a ceiling of a building 1 of a building 81. These air conditioners A1, A2, A3 are air conditioners A1, A2, A3 shown in fig. 2, which are plan views including a space S on the ceiling of the 1 st floor. In the space of the space S above the ceiling of the 1 st floor, a plurality of beams B extend in the horizontal direction. As shown in fig. 1 and 2, a beam B1 exists between the air conditioner A2 and the air conditioner A3.
(2-2) BLE Module
The air conditioner a has a BLE module M built therein. The BLE module M has an RSSI, and can measure the intensity of the received radio wave (received signal intensity) in addition to the transmission of the radio wave.
As shown in fig. 1, the air conditioner A1 has a BLE module M1 built therein, the air conditioner A2 has a BLE module M2 built therein, and the air conditioner A3 has a BLE module M3 built therein.
(2-3) Computer as a machine learning device
The computer 10 functioning as a machine learning device is composed of 1 or more computers, and is connected to HVAC equipment such as an air conditioner a of each building 81 via a communication network 80 such as the internet. The computer 10 executes cloud computing services built by a service provider of the HVAC management system in order to provide various services. The hardware configuration of the computer 10 need not be housed in 1 case, nor need it be provided as a whole.
As shown in fig. 3, the computer 10 mainly includes an acquisition unit 20, a learning unit 30, an output unit 40, an input unit 50, and an update unit 60. The computer 10 includes a control arithmetic device and a storage device. The control arithmetic device can use a processor such as a CPU or GPU. The control arithmetic device reads out a program stored in the storage device, and performs predetermined image processing and arithmetic processing according to the program. Further, the control arithmetic device can write the arithmetic result into the storage device or read out the information stored in the storage device according to the program. The acquisition unit 20, learning unit 30, output unit 40, input unit 50, and update unit 60 shown in fig. 3 are various functional blocks realized by a control arithmetic device. These functional blocks are presented by controlling the arithmetic means to execute a modeling program.
(2-3-1) Obtaining portion
The acquisition unit 20 acquires air conditioner arrangement information (first information) 21 and beam arrangement information (first information) 22 of the space behind the ceiling as information for obtaining the state variables from the external plan view database 70. The design database 70 stores design drawings of each floor of the building 81 and the like. The air conditioner arrangement information 21 is information about a place where the air conditioner a is arranged as shown in fig. 1 and 2. The air conditioner arrangement information 21 includes data of X coordinates and Y coordinates in a plan view of each air conditioner a, a distance between the air conditioners a, and the like. The beam arrangement information 22 includes data of X coordinates and Y coordinates at both ends of each beam B, information indicating which air conditioner a is located between, and the like.
In other words, the air conditioner configuration information 21 and the beam configuration information 22 are information about a certain BLE module M and other BLE modules M. The information on the distance between the air conditioners a is information on the distance between the BLE modules M of the 2 air conditioners a. Further, from the data of the X-coordinate and the Y-coordinate in the plan view of each air conditioner a and the data of the X-coordinate and the Y-coordinate at both ends of each beam B, the information of the number of beams B and air conditioners a located between a certain BLE module M and other BLE modules M can be calculated.
Here, the obtaining unit 20 obtains the distance x between any 2 BLE modules M and the number x' of the air conditioners a and the beams B located on the line segment connecting the 2 BLE modules M as state variables from the layout of the air conditioners a and the beams B extracted from the design drawing.
For example, between BLE module M1 and another BLE module M2 shown in fig. 1 and 2, there are 1 beam B1 and 1 air conditioner A3. There are 1 beam B1 between BLE module M2 and other BLE module M3, and no air conditioner a is present. Between BLE module M1 and other BLE modules M3, neither beam B1 nor air conditioner a is present.
(2-3-2) Learning portion
The learning unit 30 learns the state variable in association with the radio wave propagation state between the BLE module M and the other BLE module M. The learning unit 30 performs learning by associating a state variable with a radio propagation state based on the learning data set. The learning data set includes the actual measurement result 55 of the radio wave propagation state, which is the result of actual measurement of the radio wave propagation state, and the state variable at the time of actual measurement of the radio wave propagation state.
As shown in fig. 3, the actual measurement result 55 of the radio wave propagation state is information collected from BLE modules M of air conditioners a installed in a building 81 through an input unit 50 described later. Specifically, the state variable at the time of the actual radio wave propagation state is a value obtained from information of each air conditioner a and beam B installed in the building 81. Here, the distance between any 2 BLE modules M and the number of beams B and air conditioners a located on the line segment connecting the 2 BLE modules M are used as state variables in the actual time of the radio wave propagation state, based on the layout of the air conditioners a and beams B extracted from the design drawing.
More specifically, the learning unit 30 learns the state variable in association with the radio wave propagation state by adjusting the coefficient of the linear model 35 of the following equation 12.
Formula 12:
Wherein,
Is the result of the estimation of the propagation state of the electric wave,
W is a coefficient of the degree of freedom,
X is the distance of the BLE module from the other BLE modules,
X' is the number of beams and air conditioners between the BLE module and the other BLE modules.
(2-3-3) Output portion
The output unit 40 outputs the estimation result 45 of the radio wave propagation state obtained by the linear model 35 of the learning unit 30.
(2-3-4) Input portion
The input unit 50 collects the actual measurement result 55 of the radio wave propagation state from the BLE module M of each air conditioner a installed in the building 81 via the communication network 80. The input unit 50 may collect the actual measurement result 55 of the radio wave propagation state from the user terminal 90 via the communication network 80.
(2-3-5) Updating part
The updating unit 60 obtains the difference between the estimated result 45 of the radio wave propagation state output from the output unit 40 and the actual measurement result 55 of the radio wave propagation state input to the input unit 50. Then, the updating unit 60 evaluates the difference to update the learning state of the learning unit 30.
Specifically, the updating unit 60 updates the coefficient w in the above-described expression 12 so as to reduce the following evaluation function 12, thereby updating the learning state of the learning unit 30.
Evaluation function 12:
Wherein,
Is the result of estimation 45 of the propagation state of the electric wave,
Y is the actual measurement result 55 of the radio wave propagation state,
Alpha is a regularization parameter.
(3) Features of a computer (machine learning device) of an HVAC management system
(3-1)
In general, when the distance between the transmitter and the receiver is a factor of propagation path loss, the propagation of radio waves in free space can be described by a fries transmission formula (Friis Transmission Equation). But the space on the ceiling of a building is typically a height of 0.5m or 1.5m, which is a very limited space. In addition, in the space above the ceiling, there are beams, HVAC equipment for maintaining the strength of the building. In such a complicated space, in addition to the distance, there is a possibility that an obstacle on the radio wave propagation path has a large influence on propagation path loss due to reflection and refraction. Therefore, the obstacle also needs to be considered as a variable of the model. In particular, 2 obstacles, beams (beams) and HVAC equipment, which are made of metal and have a relatively large volume, should be considered as variables of the linear model.
In view of this, in the computer 10 as the machine learning device of the present embodiment, the distance between the BLE module M and the other BLE modules M, and the number of beams B and air conditioners a located between the BLE module M and the other BLE modules M are brought into focus.
In the computer 10 of the HVAC management system according to the present embodiment, the acquisition unit 20 acquires air conditioner arrangement information (first information) 21 concerning the BLE module M and the other BLE modules M, and beam arrangement information (first information) 22 of the space behind the ceiling. Therefore, the computer 10 can obtain the state variables required for the learning unit 30 from these pieces of information. Further, since the learning unit 30 learns the state variable in association with the radio propagation state, the radio propagation state between the BLE module M and the other BLE modules M can be obtained by the computer 10.
Further, since the distance (x) between the BLE module M and the other BLE module M is used as a state variable, a radio wave propagation state with high accuracy can be obtained.
In this way, according to the computer 10, in the narrow space S on the ceiling where it is difficult to obtain a result of good accuracy in the conventional radio wave propagation simulator, a radio wave propagation state of relatively good accuracy can be obtained by the learning unit 30.
(3-2)
In the computer 10 as the machine learning device of the present embodiment, the number (x') of the beam B and the air conditioner a located between the BLE module M and the other BLE module M is also adopted as the state variable. Therefore, the computer 10 can obtain a radio wave propagation state with higher accuracy.
(3-3)
In the computer 10 as the machine learning device of the present embodiment, actual measurement of the radio propagation state is performed by each BLE module M in the building 81, and the learning unit 30 learns using a learning data set composed of the state variable and the actual measurement result 55 of the radio propagation state at that time. Specifically, the updating unit 60 updates the learning state of the learning unit 30 by evaluating the difference between the estimated result 45 of the radio wave propagation state and the actually measured result 55 of the radio wave propagation state. Thus, the coefficient w of the linear model 35 is a proper value that matches the arrangement of the beam B and the air conditioner a in the space on the ceiling of the building 81.
(4) One example of a comparison of a model with 1 variable to a model with 3 variables
Examples
The computer (machine learning device) 10 focuses on beams and air conditioners, which are 2 kinds of obstacles made of metal and having a relatively large volume. First, based on a layout extracted from the design drawing, an arbitrary distance (distance) between 2 BLE modules, and the number of beams and air conditioners on a line segment connecting 2 BLE modules (#beam and #machine) are obtained. Then, next, a prediction variable set composed of 3 variables (distance, # beam, # machine) is generated.
Here, in order to evaluate the influence of #beam and #machine, the learning unit uses
1 Variable (1-feature), distance,
3 Variables (3-feature; distance, #beam, #machine)
Both of which construct a linear model and their differences are compared.
For the 1-variable model, OLS regression with polynomials is chosen (Ordinary Linear Regression; usual least squares). Linear regression approximates a linear model with coefficients that minimize the sum of squares of residuals between observed responses and responses predicted by linear approximation within the dataset. Mathematically, the model can be shown as follows.
Wherein,
X is a variable that is used to determine the degree of freedom,
Is the result of the estimation of the propagation state of the electric wave,
W is a coefficient.
The sum of squares of the residuals (sometimes also referred to as the loss function) is represented by the following equation.
Y is the RSSI attribute and,
X is a polynomial of the distance, is
d,d2,…,dn。
On the other hand, for a model with 3 variables [ distance, # beam, # machine ], ridge Regression (Ridge Regression) using a polynomial was selected. Ridge regression is also 1 of generalized linear regression. In contrast to OLS regression, there is a penalty function accompanying an additional penalty term in ridge regression.
The larger the value of α, the larger the penalty, and therefore the smaller the coefficient size.
In the initial trial of training of the model, α=10 was applied to the model with respective 3 (distance, #beam, #machine) variables.
For both OLS and ridge regression, the degree of the polynomial is changed from 1 to 4, scaling is applied to the variables before regression. By using the function of the machine learning library, the variables are initially transformed into a normal distribution, and then the maximum absolute value of each variable is scaled to 1.0.
The accuracy of linear model prediction is mainly evaluated by RMSE (Root Mean Square Error; root mean square error) and R 2 (decision coefficient). Here, RMSE is employed. The RMSE is calculated using the absolute difference between the estimated RSSI and the measured RSSI.
Here, RMSE calculation was performed by the following 2 methods.
1) Both model training and testing (RMSE calculations) were performed on all data sets (100% data sets).
2) The K-split cross-validation (k=10) set was randomly split into 10 sets, the model trained on 9 sets and tested on the remaining 1 set. The test group was repeated to test the model and calculate 10 RMSE. Then, the average value of 10 RMSE was used as an evaluation index.
In the experiment, data sampling was performed in a building (office building) in which all beams were made of metal and typically have a steel skeleton structure.
The height of the space above the ceiling (from below the gypsum board to the floor above) was 0.85m,
The maximum height of the beam is 0.7m,
The average height of the air conditioner was 0.3m.
Each floor was a flat ceiling of 18m x 18 m. The 26 BLE modules are arranged in floors 1 and 2. The distance between 2 BLE modules is 1.3m to 20m.
The data sampling process is performed for 1 week during which the BLE module exchanges communication packets and collects packet records. The packet record includes a time stamp, a transmitter ID, a receiver ID, transmission power (always set to 8dBm for each BLE module), and RSSI. Wi-Fi-induced wireless interference within a building becomes serious due to operating time. With this in mind, nighttime (22 to 7 points) and RSSI for weekends for model training were selected. The RSSI is further resampled at 5 minute intervals for each pair of BLE modules, with the average of the RSSI in each interval being used as the 100% dataset described above.
The results of the model evaluation are shown in table 1.
TABLE 1
RMSE of model
Here, 2 RMSE were calculated. In K-fold cross-validation, RMSE for data sets greater than 100% is shown. Any RMSE shows the same tendency.
Among the models having the same function, the higher the number of times, the lower the RMSE. The model adds the number of beams and the number of air conditioners as variables in addition to the distance, thereby improving the accuracy of the radio wave propagation model in the space above the ceiling.
Next, table 2 shows the comparison of the RSSI shown by the model with the RSSI of the actual BLE module. Of the 8 sets of models, the 4-degree model of 3 variables shows the best estimation accuracy. In table 2, the difference between pairs (not shown) of all BLE modules associated with the BLE module of "102A" is shown.
TABLE 2
Estimation of BLE 102 pair
/>
To improve the process of network address setting of HVAC equipment, it is useful to utilize the RSSI of the BLE module described above. It was found through experiments that the RMSE of RSSI predictions has the following tendency.
1) By taking into account information about obstacles such as the number of beams and the number of HVAC equipment such as air conditioners, RMSE is reduced.
2) As the degree of the polynomial increases, RMSE decreases.
(5) Modification examples
(5-1)
In the above-described computer (machine learning device) 10, the number of obstacles (beam B and air conditioner a) located on the line segment connecting the 2 BLE modules M is adopted as the state variables related to these obstacles. However, the size of the obstacle may be used instead of or in addition to this as a state variable.
(5-2)
In the above-described computer (machine learning device) 10, the number of obstacles (beam B and air conditioner a) located on the line segment connecting the 2 BLE modules M is adopted as the state variables related to these obstacles. Alternatively or in addition, however, the position of the obstacle may be used as a state variable.
(5-3)
In the above-described computer (machine learning device) 10, the number of obstacles (beam B and air conditioner a) located on the line segment connecting the 2 BLE modules M is adopted as the state variables related to these obstacles. Alternatively or in addition, however, the orientation of the obstacle may be used as a state variable.
(5-4)
In the computer (machine learning device) 10, the learning unit 30 adjusts the coefficient of the linear model 35 of the expression 12 to thereby learn the state variable in association with the radio wave propagation state. Instead, the learning unit 30 may learn by adjusting the coefficient of the linear model of the following expression 11 so as to correlate the state variable with the radio propagation state.
Formula 11:
Wherein,
Is the result of the estimation of the propagation state of the electric wave,
W is a coefficient of the degree of freedom,
X is the distance of the BLE module from the other BLE modules.
The updating unit 60 may update the learning state of the learning unit 30 by updating the coefficient w in the above-described expression 11 so that the following evaluation function 11 is smaller than the above-described evaluation function 12.
Evaluation function 11:
Wherein,
Is the result of estimation 45 of the propagation state of the electric wave,
Y is the actual measurement result 55 of the radio wave propagation state.
Even when the learning state of the learning unit 30 is updated by the coefficient w of the linear model of the update 11, a radio wave propagation state with high accuracy can be obtained.
(5-5)
In the HVAC management system described above, the BLE module M is provided at the air conditioner a as the HVAC equipment, but other wireless equipment may be provided instead of the BLE module. For example, a ZigBee module may also be employed.
(6) Comparison of an estimated value of radio wave intensity using a machine learning device with an estimated value of radio wave intensity using a conventional simulation
(6-1)
As described above, in order to reduce the effort (address setting effort) of mapping the network address of the HVAC equipment or the like to the physical arrangement location of the HVAC equipment or the like, the computer 10 as the machine learning device is useful for estimating the radio wave propagation state between the BLE modules M. If the installation position of each air conditioner can be detected and address setting can be automated, it is expected that feedback control, energy saving in a building, and development of intelligent control technology can be expected.
Therefore, the air conditioner is equipped with BLE modules, respectively, and the above-described machine learning device uses the reception intensity of the radio wave outputted from the BLE modules. The reception intensity of radio waves (radio wave intensity) is one of the indexes indicating the propagation state of radio waves.
The radio wave intensities between 2 points tend to be weaker as the distance increases due to the attenuation of the radio wave by the distance. However, if there is an object that blocks the propagation of the radio wave between these 2 points, the attenuation of the radio wave intensity increases due to the influence thereof. Conventionally, simulation based on a physical model has been used for estimating the propagation intensity of radio waves, but input conditions (input contents) are numerous and complicated. In address setting using a conventional physical model, if the input conditions are narrowed to such an extent that they can be used as normal traffic, there is a problem in that accuracy is greatly deteriorated.
In view of this, in the above-described machine learning apparatus, machine learning is used to construct a radio wave propagation model that predicts the radio wave intensity measured based on the distance between 2 BLE modules and the obstacle located between 2 BLE modules that blocks the radio wave propagation.
In the following, in order to verify the validity of a model using machine learning, a comparison result with a conventional physical model based on simulation is shown.
(6-2) Comparative evaluation
First, in prediction by machine learning, the radio wave intensity of the BLE module is used as a specified variable. The distance between BLE modules and the number of air conditioners and beams are described as explanatory variables (state variables). After a radio wave propagation model is learned by adding an actual radio wave intensity measurement value of a BLE module mounted on the back of the ceiling of a building, the radio wave intensity is estimated using the learning model. In the prediction accuracy evaluation of machine learning, accuracy evaluation was performed using the radio wave intensities between 15 BLE modules installed on the rear surface of the ceiling of a building.
In the prediction of simulation, a 3D model of the actual environment (including models of air conditioners and beams) after the ceiling of a building is created, reference values are input to material parameters, and radio wave propagation simulation is performed to estimate the radio wave intensity. The simulation software uses software in which an expansion module for performing high-definition radio wave propagation is combined with a commercially available discrete phenomenon simulator. In the simulation by this software, simulation in a fine radio wave propagation environment can be performed in consideration of influences of a building or the like on reflection, shielding, and diffraction of radio waves. In the prediction accuracy evaluation based on the simulation, the radio wave intensity between 5 BLE modules included in the building portion in which the 3D model is constructed is set as the evaluation target.
In addition, in the prediction of simulation, in addition to the input material parameters, in the case where there is no BIM model (digital model of a 3D building), it is necessary to create the model, and therefore, more complicated work is required than in the case where a learning model is used.
The results of the precision evaluation performed in the above-described order are shown in table 3 below.
TABLE 3
In the simulation, although the standard error of the actual measurement value of the radio wave intensity is small, the estimation error is large.
From the above results, it is clear that the method using machine learning can predict the radio wave intensity with good accuracy as compared with the method using simulation. Here, although learning is performed using fewer state variables than the input conditions of the simulation, high accuracy can be obtained using the learning model.
(7) Matching between installation position of each air conditioner and BLE module
In the following steps, the radio wave intensity as the radio wave propagation state obtained by the above-described machine learning apparatus is used in a matching algorithm of the device setup position and the BLE module for determining the position of the BLE module. In matching the device installation position with the BLE module, first, an undirected graph GE is obtained in which the estimated received radio wave intensity obtained by the above-described machine learning method is set as a value of a side and the position ID of the air conditioner is set as a vertex. Next, BLE modules of air conditioners incorporated in a building (site) as a target property are transmitted to each other. And collecting the measured received radio wave intensity of the BLE module, and making an undirected graph GM taking the measured received radio wave intensity as a value of an edge and taking a vertex as an ID of the transmitting BLE module and the receiving BLE module. However, the estimated value and the measured value of the radio wave intensity are inevitably subject to errors. Therefore, by setting the error tolerance (slot value), it is determined that there is a possibility that the edge in the undirected graph GM and the edge in the undirected graph GE have the same error below the tolerance. In addition, by performing a matching algorithm for the undirected graph GM to the undirected graph GE, BLE module(s) to be matching candidates are determined for each installation position of the air conditioner.
If the estimated radio wave intensity value using simulation is used in the matching described above, when the BLE module candidate is determined for each installation position using the same algorithm and the same error tolerance value, the case where the correct solution is not included in the determined BLE module candidate becomes more likely. In addition, conversely, in the case where a larger error tolerance value is used so that a correct solution is included, the number of candidates of the determined BLE module increases, and screening of the correct solution becomes difficult.
(8)
While the embodiment of the HVAC management system having the computer 10 as the machine learning device has been described above, it should be understood that various changes in form and details may be made without departing from the spirit and scope of the present disclosure as set forth in the claims.
Description of the reference numerals
10: A computer (machine learning device); 20: an acquisition unit; 21: air conditioner configuration information (first information; article information); 22: beam arrangement information (first information; article information) of the space behind the ceiling; 30: a learning unit; 35: a linear model; 40: an output unit; 45: an estimation result of the radio wave propagation state; 55: actual measurement results of radio wave propagation states; 60: an updating unit; a (A1, A2, A3): air conditioner (specified articles); b (B1): a beam (a prescribed article); m (M1, M2, M3): BLE module (wireless device).
Prior art literature
Patent literature
Patent document 1: japanese patent laid-open publication 2016-208265
Claims (15)
1. A machine learning apparatus (10) that learns a radio wave propagation state between a wireless device (M1) and another wireless device (M2) and assists a mapping operation between network addresses and physical arrangement locations of the wireless device and the other wireless device, the machine learning apparatus (10) comprising:
an acquisition unit (20) that acquires first information (21, 22) regarding the wireless device and the other wireless devices as information for obtaining a state variable; and
A learning unit (30) for learning by associating the state variable with the radio wave propagation state,
The wireless device and the other wireless device are configured to measure an actual radio wave propagation state therebetween.
2. The machine learning device of claim 1 wherein,
The first information comprises information (21) of the distance of the wireless device from other wireless devices,
The distance of the wireless device from other of the wireless devices is at least 1 of the state variables.
3. The machine learning device of claim 1 wherein,
The first information includes item information (21, 22) relating to a prescribed item located between the wireless device and the other wireless devices.
4. The machine learning apparatus of claim 3, wherein,
The item information includes information of the number of the specified items.
5. The machine learning apparatus of claim 3, wherein,
The predetermined article is an air conditioner (A) and/or a beam (B) disposed in a space above a ceiling.
6. The machine learning apparatus of claim 3, wherein,
The item information includes information about a size of the prescribed item.
7. The machine learning apparatus of claim 3, wherein,
The article information includes information about a position of the prescribed article.
8. The machine learning apparatus of claim 3, wherein,
The article information includes information about an orientation of the prescribed article.
9. The machine learning device of claim 1 wherein,
The learning unit learns the state variable and the radio wave propagation state based on a learning data set including a result (55) obtained by actually measuring the radio wave propagation state and the state variable at the time of actually measuring the radio wave propagation state.
10. The machine learning apparatus of claim 3, wherein,
The learning unit learns the state variable and the radio wave propagation state based on a learning data set composed of a result (55) obtained by actually measuring the radio wave propagation state and the state variable obtained from the article information at the time of actually measuring the radio wave propagation state.
11. The machine learning device according to any one of claims 1, 2, 9, wherein,
The learning unit learns the state variable by adjusting the coefficient of the linear model (35) of the following formula 1 in association with the radio wave propagation state,
Formula 1:
Wherein,
Is the result of the estimation of the propagation state of the electric wave,
W is a coefficient of the degree of freedom,
X is the distance of the wireless device from other wireless devices.
12. The machine learning device according to any one of claims 3 to 8, 10, wherein,
The learning unit learns the state variable by adjusting the coefficient of the linear model (35) of the following formula 2 in association with the radio wave propagation state,
Formula 2:
Wherein,
Is the result of the estimation of the propagation state of the electric wave,
W is a coefficient of the degree of freedom,
X is the distance of the wireless device from other wireless devices,
X' is the number of the prescribed items between the wireless device and the other wireless devices.
13. The machine learning apparatus according to any one of claims 1 to 10, wherein,
The machine learning device further includes:
an output unit (40) that outputs the estimation result of the radio wave propagation state; and
An updating unit (60) for updating the learning state of the learning unit,
The updating unit evaluates the difference between the estimated result (45) of the radio wave propagation state and the result (55) obtained by actually measuring the radio wave propagation state, thereby updating the learning state of the learning unit.
14. The machine learning device of claim 11 wherein,
The machine learning device further includes:
an output unit (40) that outputs the estimation result of the radio wave propagation state; and
An updating unit (60) for updating the learning state of the learning unit,
The updating section updates the coefficient w in the expression 1 so as to make the following evaluation function 1 smaller, thereby updating the learning state of the learning section,
Evaluation function 1:
Wherein,
Is the result of the estimation of the propagation state of the electric wave,
Y is a result obtained by actually measuring the radio wave propagation state.
15. The machine learning device of claim 12 wherein,
The machine learning device further includes:
an output unit (40) that outputs the estimation result of the radio wave propagation state; and
An updating unit (60) for updating the learning state of the learning unit,
The updating section updates the coefficient w in the expression 2 so as to make the following evaluation function 2 small, thereby updating the learning state of the learning section,
Evaluation function 2:
Wherein,
Is the result of the estimation of the propagation state of the electric wave,
Y is a result obtained by actually measuring the propagation state of the radio wave,
Alpha is a regularization parameter.
Applications Claiming Priority (5)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2019-076604 | 2019-04-12 | ||
| JP2019076604 | 2019-04-12 | ||
| JP2020-021199 | 2020-02-12 | ||
| JP2020021199A JP7460886B2 (en) | 2019-04-12 | 2020-02-12 | Machine Learning Equipment |
| PCT/JP2020/016192 WO2020209380A1 (en) | 2019-04-12 | 2020-04-10 | Machine learning device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113678468A CN113678468A (en) | 2021-11-19 |
| CN113678468B true CN113678468B (en) | 2024-05-10 |
Family
ID=72831713
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202080027864.6A Active CN113678468B (en) | 2019-04-12 | 2020-04-10 | Machine learning device |
Country Status (2)
| Country | Link |
|---|---|
| JP (1) | JP7460886B2 (en) |
| CN (1) | CN113678468B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023145515A1 (en) * | 2022-01-31 | 2023-08-03 | 京セラ株式会社 | Information processing system, information processing method, and information processing program |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2016217773A (en) * | 2015-05-15 | 2016-12-22 | 株式会社日立産機システム | Radio wave measurement system and base station |
| CN106707035A (en) * | 2016-12-19 | 2017-05-24 | 西华大学 | Method for estimating field strength parameters of radio environment |
| CN107667552A (en) * | 2015-03-24 | 2018-02-06 | 开利公司 | The study and registration based on floor plan of distributed devices |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2672288B1 (en) | 2011-01-31 | 2021-05-19 | Mitsubishi Electric Corporation | Locating apparatus, locating system, locating method and program |
| JP6010204B1 (en) * | 2015-10-26 | 2016-10-19 | ファナック株式会社 | Machine learning apparatus and method for learning predicted life of power element, life prediction apparatus and motor driving apparatus equipped with machine learning apparatus |
| JP2019036901A (en) | 2017-08-21 | 2019-03-07 | パナソニックＩｐマネジメント株式会社 | Communication availability determination device, communication availability determination system, communication availability determination method, and program |
-
2020
- 2020-02-12 JP JP2020021199A patent/JP7460886B2/en active Active
- 2020-04-10 CN CN202080027864.6A patent/CN113678468B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107667552A (en) * | 2015-03-24 | 2018-02-06 | 开利公司 | The study and registration based on floor plan of distributed devices |
| JP2016217773A (en) * | 2015-05-15 | 2016-12-22 | 株式会社日立産機システム | Radio wave measurement system and base station |
| CN106707035A (en) * | 2016-12-19 | 2017-05-24 | 西华大学 | Method for estimating field strength parameters of radio environment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113678468A (en) | 2021-11-19 |
| JP2020174343A (en) | 2020-10-22 |
| JP7460886B2 (en) | 2024-04-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10635055B2 (en) | Building control system with break even temperature uncertainty determination and performance analytics | |
| EP3324316A1 (en) | Parameter estimation apparatus, air-conditioning system evaluation apparatus, parameter estimation method, and program | |
| JP2005528622A (en) | A stochastic model for positioning techniques. | |
| CN109740265B (en) | Urban outdoor electromagnetic environment situation prediction method based on MoM-UTD | |
| US20180279078A1 (en) | Confirming work supporting device, confirming work supporting system, and computer program product | |
| JP7231403B2 (en) | Air conditioning control system and method | |
| Pan et al. | Indoor scenario-based UWB anchor placement optimization method for indoor localization | |
| CN109981195B (en) | Method and device for processing wireless signal strength | |
| CN108242962B (en) | Indoor signal propagation loss calculation method and device based on measurement report | |
| CN112347705B (en) | Method and system for establishing transformer substation factory boundary noise model | |
| CN113678468B (en) | Machine learning device | |
| CN112135246B (en) | RSSI (received Signal Strength indicator) updating indoor positioning method based on SSD (solid State disk) target detection | |
| JP2015010927A (en) | Observation value processing device, observation value processing method, and observation value processing program | |
| WO2020209380A1 (en) | Machine learning device | |
| CN101345559A (en) | Method for route loss estimation of base station signal strength | |
| CN113778042B (en) | Industrial control equipment instruction simulation operation communication system based on informatization modeling | |
| CN104635206A (en) | Wireless positioning method and device | |
| Hosseini et al. | NSGA-II based optimal Wi-Fi access point placement for indoor positioning: A BIM-based RSS prediction | |
| CN108132458B (en) | Indoor distance measuring method, device, equipment and storage medium | |
| US20210329478A1 (en) | Radio wave strength estimation device, position estimation system, and radio wave strength estimation method | |
| CN113685923A (en) | Dust-free workshop environment regulation and control method and system for mask production | |
| JP7553768B2 (en) | Equipment location identification system | |
| Rarig et al. | Risk/dispersion index method | |
| US20120203471A1 (en) | Generation source estimation apparatus and method of diffusion material | |
| WO2023145515A1 (en) | Information processing system, information processing method, and information processing program |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |