Learning device, inference device, and wind speed measurement system

Abstract

To obtain a learning device that can estimate the wind speed inside a ventilation hood without installing a dedicated sensor for calculating wind speed. [Solution] The learning device 5 is a learning device 5 for learning the wind speed inside the ventilation hood 3, and includes a learning data acquisition unit that acquires learning data including sound pressure data inside the ventilation hood 3, combination data including the ventilation hood 3, a ventilation fan 1 that ventilates the room through the ventilation hood 3, and a ventilation fan attachment member 2 attached to the ventilation fan 1, data relating to a sound pressure data acquisition unit 4 that acquires sound pressure data, and the wind speed inside the ventilation hood 3, and a model generation unit that uses the learning data to generate a trained model for inferring the wind speed inside the ventilation hood 3 from the sound pressure data, combination data and data relating to the sound pressure data acquisition unit 4. As a result, the amount of air blown into the room can be managed using the wind speed inside the ventilation hood 3, which is the inference result.

Description

Technical Field

[0001] The present disclosure relates to a learning device, an inference device, and a wind speed measurement system for estimating the wind speed inside a ventilation hood.

Background Art

[0002] Conventionally, as a technique for estimating the wind speed, there is the technique disclosed in Patent Document 1. In Patent Document 1, the wind sound from a columnar sensor having a predetermined strobe number is collected, the collected wind sound is frequency-analyzed to obtain a wind sound spectrum, and then the wind speed is calculated based on the dominant frequency of the dominant peak in the wind sound spectrum and the strobe number.

[0003] Specifically, Patent Document 1 discloses that when the dominant frequency of the wind sound is f (Hz), D is the outer diameter (m) of the sensor, V is the wind speed (m / s), and the strobe number is St, the wind speed V is calculated by the formula V = fD / St.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] When installing ventilation fans in a building, a ventilation hood is often installed together with the fan to prevent rainwater from entering the room through the fan. When using a ventilation fan to ventilate the interior of a building, it is important to manage the airflow to ensure sufficient volume of air is supplied to the interior or exterior. To manage the airflow to the interior or exterior, it is necessary to accurately calculate the wind velocity inside the ventilation hood. Therefore, it is conceivable to apply the technology disclosed in Patent Document 1 to calculate the wind velocity inside the ventilation hood, but there may not be enough space to install a sensor with a predetermined strohull number inside the ventilation hood. Furthermore, even if it were possible to install the sensor inside the ventilation hood, the sensor itself may become a new source of noise.

[0006] Furthermore, while blades to prevent rain intrusion may be installed inside ventilation hoods, and shutters, louvers, etc., may be attached to ventilation fans, these blades and the like become components that generate wind noise (hereinafter referred to as "wind noise generating components"). Therefore, when obtaining a wind noise spectrum by collecting wind noise from a sensor, as in the technology disclosed in Patent Document 1, the wind noise generated from the wind noise generating components may become noise, and it may not be possible to obtain the wind noise spectrum necessary for calculating wind speed.

[0007] This disclosure has been made in view of the above, and aims to provide a learning device that can estimate the wind speed inside a ventilation hood without installing a dedicated sensor for calculating wind speed. [Means for solving the problem]

[0008] To solve the above-mentioned problems and achieve the objective, the learning device according to this disclosure is a learning device for learning the wind speed inside a ventilation hood, and comprises a learning data acquisition unit and a model generation unit. The learning data acquisition unit acquires learning data including sound pressure data inside the ventilation hood, combination data including the shape name information of ventilation components including the ventilation hood, a ventilation fan that ventilates the room through the ventilation hood, and ventilation fan attachment components attached to the ventilation fan, data relating to the sound pressure data acquisition unit that acquires the sound pressure data, and the wind speed inside the ventilation hood. The model generation unit uses the learning data to generate a trained model for inferring the wind speed inside the ventilation hood from the sound pressure data, combination data, and data relating to the sound pressure data acquisition unit. [Effects of the Invention]

[0009] The learning device described herein has the effect of being able to estimate the wind speed inside a ventilation hood without installing a dedicated sensor for calculating wind speed. [Brief explanation of the drawing]

[0010] [Figure 1] Schematic diagram showing the wind speed measurement system according to Embodiment 1 [Figure 2] Perspective view showing a shutter according to Embodiment 1 [Figure 3] Perspective view showing a louver according to Embodiment 1 [Figure 4] Cross-sectional view showing another example of the ventilation hood according to Embodiment 1. [Figure 5] This figure shows the configuration of the learning device for the wind speed measurement system according to Embodiment 1. [Figure 6] This figure shows the configuration of the neural network used in the learning device according to Embodiment 1. [Figure 7] Flowchart showing the processing procedure of the learning process by the learning device according to Embodiment 1 [Figure 8] This figure shows the configuration of the inference device for the wind speed measurement system according to Embodiment 1. [Figure 9]A flowchart illustrating the processing procedure of the inference process by the inference device according to Embodiment 1 and the output processing by the calculation unit. [Figure 10] This figure shows the hardware configuration of the learning device according to Embodiment 1. [Modes for carrying out the invention]

[0011] The learning device, inference device, and wind speed measurement system according to the embodiment will be described in detail below with reference to the drawings.

[0012] Embodiment 1. Figure 1 is a schematic diagram showing the wind speed measurement system 100 according to Embodiment 1. As shown in Figure 1, the wind speed measurement system 100 comprises a ventilation fan 1, a ventilation fan attachment 2, a ventilation hood 3, a sound pressure data acquisition unit 4, a learning device 5, an inference device 6, a calculation unit 7, and a terminal device 8. When describing the direction of each component of the wind speed measurement system 100 below, we will follow the right-handed XYZ coordinate system shown in Figure 1. The X, Y, and Z axes are three axes perpendicular to each other. Of each axis, the direction of the arrow is considered the + direction, and the direction opposite to the arrow is considered the - direction. The direction along the X axis (X-axis direction) and the direction along the Z axis (Z-axis direction) are directions included in the horizontal direction. The direction along the Y axis (Y-axis direction) coincides with the vertical direction. Furthermore, the + direction of the Y axis is considered upward, and the - direction of the Y axis is considered downward, and the Y-axis direction is referred to as the vertical direction.

[0013] In this embodiment, we assume that the ventilation fan 1 and ventilation hood 3 are installed on the wall 10a of the building 10. In Figure 1, the area to the right of the wall 10a is the interior of the building, and the area to the left of the wall 10a is the exterior. The wall 10a of the building 10 has an opening 10b that connects the interior and exterior of the building 10. The opening 10b penetrates the wall 10a in the direction of the wall thickness. In Figure 1, cross-sectional hatching is applied to the wall 10a for ease of understanding.

[0014] The ventilation fan 1 is a device that ventilates the interior of the building 10 through the ventilation hood 3. In this embodiment, the ventilation fan 1 is a dedicated air supply ventilation fan that supplies outdoor air into the room through the ventilation hood 3 to ventilate the room, but it may also be a dedicated exhaust ventilation fan that exhausts indoor air to the outside through the ventilation hood 3 to ventilate the room. The ventilation fan 1 has a frame 1a attached to the wall 10a, an electric motor 1b supported by the frame 1a, and blades 1c detachably attached to the rotating shaft (not shown) of the electric motor 1b. The frame 1a is attached to the wall surface of the wall 10a facing the interior of the room. As the blades 1c rotate with the rotation of the rotating shaft of the electric motor 1b, outdoor air is supplied into the room of the building 10 through the ventilation hood 3. The arrow A shown in FIG. 1 represents the direction in which the air flow generated by the rotation of the blades 1c flows during the operation of the ventilation fan 1.

[0015] The ventilation fan accessory member 2 is a member attached to the ventilation fan 1. The ventilation fan accessory members 2 include a filter 2a, a shutter 2b, a grille 2c, etc. The ventilation fan 1 is installed in the building 10 in combination with ventilation fan accessory members 2 such as a filter 2a, a shutter 2b, and a grille 2c according to the assumed usage environment. The uses of the filter 2a are various, such as dust prevention, insect prevention, and bird prevention. The type of the filter 2a is appropriately selected according to the use. In the example shown in FIG. 1, the filter 2a is installed in the opening 10b and between the ventilation hood 3 and the sound pressure data acquisition unit 4 and the ventilation fan 1.

[0016] Both the shutter 2b and the grille 2c are members that control the air flow. When it is desired to control the air flow, either one of the shutter 2b and the grille 2c is installed in the building 10. In the example shown in FIG. 1, the shutter 2b or the grille 2c is installed at a position closer to the interior of the room than the filter 2a within the opening 10b and between the ventilation hood 3 and the sound pressure data acquisition unit 4 and the ventilation fan 1.

[0017] FIG. 2 is a perspective view showing the shutter 2b according to Embodiment 1. As shown in FIG. 2, a plurality of blades 2d for controlling the air flow are installed in the shutter 2b. The plurality of blades 2d are installed at intervals in the vertical direction. The blades 2d of the shutter 2b are variable. The blades 2d of the shutter 2b open and close in accordance with the operation and stop of the ventilation fan 1. That is, when the ventilation fan 1 operates, each blade 2d opens to open the opening 10b, and the interior and exterior of the building 10 communicate with each other. On the other hand, when the ventilation fan 1 stops, each blade 2d closes to cover the opening 10b, and the interior and exterior of the building 10 do not communicate with each other.

[0018] FIG. 3 is a perspective view showing the gallery 2c according to Embodiment 1. As shown in FIG. 3, a plurality of blades 2e for controlling the air flow are also installed in the gallery 2c. The plurality of blades 2e are installed at intervals in the vertical direction. The blades 2e of the gallery 2c are fixed.

[0019] The ventilation hood 3 shown in FIG. 1 is a member for preventing rainwater from entering from the outside to the inside. The ventilation hood 3 is attached to the wall surface of the wall 10a facing the outside, and covers the opening 10b from above and in front of the opening 10b. In this specification, the ventilation fan 1, the ventilation fan accessory member 2, and the ventilation hood 3 are collectively referred to as "ventilation members". The wind speed measurement system 100 is provided with an air passage 9 that reaches the interior through the ventilation hood 3, the opening 10b, and the ventilation fan 1. FIG. 4 is a cross-sectional view showing another example of the ventilation hood 3 according to Embodiment 1. As shown in FIG. 4, a plurality of blades 3a may be installed in the ventilation hood 3. Each blade 3a has a role of preventing rainwater from entering from the outside to the inside. The plurality of blades 3a are installed at intervals in the X-axis direction.

[0020] The sound pressure data acquisition unit 4 shown in Figure 1 acquires sound pressure data inside the ventilation hood 3. The sound pressure data acquisition unit 4 is installed inside the ventilation hood 3. The sound pressure data acquisition unit 4 is, for example, a microphone. The sound pressure data acquired by the sound pressure data acquisition unit 4 is transmitted to the calculation unit 7.

[0021] The learning device 5 learns the wind speed inside the ventilation hood 3, which corresponds to the sound pressure data inside the ventilation hood 3, the combination data including the ventilation hood 3, the ventilation fan 1 that ventilates the room through the ventilation hood 3, and the ventilation fan attachment 2 attached to the ventilation fan 1, as well as data related to the sound pressure data acquisition unit 4. The learning device 5 also generates a trained model for inferring the wind speed inside the ventilation hood 3. Details of the learning device 5 will be described later.

[0022] The inference device 6 uses a trained model to infer the airflow velocity inside the ventilation hood 3. The airflow velocity inside the ventilation hood 3 inferred by the inference device 6 is transmitted to the calculation unit 7. Details of the inference device 6 will be described later.

[0023] The calculation unit 7 outputs the wind speed inside the ventilation hood 3, inferred by the inference device 6, to the terminal device 8. The calculation unit 7 may, if necessary, perform processing such as converting the sound pressure data received from the sound pressure data acquisition unit 4 into a frequency spectrum. This makes it possible to smoothly associate the sound pressure data inside the ventilation hood 3 with the wind speed inside the ventilation hood 3. In this specification, the sound pressure data received by the calculation unit 7 from the sound pressure data acquisition unit 4, and the sound pressure data processed by the calculation unit 7 from the sound pressure data acquisition unit 4, are collectively referred to as "sound pressure data inside the ventilation hood 3".

[0024] If information on the airflow area within the ventilation hood 3 is provided to the calculation unit 7, the calculation unit 7 calculates the airflow within the ventilation hood 3 by multiplying the wind speed within the ventilation hood 3, inferred by the inference device 6, by the airflow area within the ventilation hood 3. The airflow area within the ventilation hood 3 is input to the calculation unit 7 from the terminal device 8 for each ventilation hood 3 used. The calculation unit 7 may be integrated with the sound pressure data acquisition unit 4, or it may be separate from the sound pressure data acquisition unit 4. If the calculation unit 7 and the sound pressure data acquisition unit 4 are separate, the calculation unit 7 and the sound pressure data acquisition unit 4 are connected to each other via wired or wireless communication, and the sound pressure data acquired by the sound pressure data acquisition unit 4 is transmitted to the calculation unit 7 via wired or wireless communication.

[0025] Terminal device 8 is, for example, a personal computer. Terminal device 8 is used for inputting combination data, inputting data related to sound pressure data acquisition unit 4, inputting airflow area within ventilation hood 3, notifying administrators of inference results by inference device 6, and notifying administrators of calculation results by calculation unit 7. Terminal device 8 and calculation unit 7 are electrically connected by wire or wireless.

[0026] Next, we will explain in detail the learning phase in which the learning device 5 performs machine learning to generate a trained model.

[0027] Figure 5 shows the configuration of the learning device 5 of the wind speed measurement system 100 according to Embodiment 1. The learning device 5 comprises a learning data acquisition unit 5a, a model generation unit 5b, and a learned model storage unit 5c. In Figure 5, an example is shown in which the learned model storage unit 5c is located outside the learning device 5, but the learned model storage unit 5c may be located inside the learning device 5.

[0028] The learning data acquisition unit 5a acquires sound pressure data from inside the ventilation hood 3, combination data including the model name information of the ventilation member, data related to the sound pressure data acquisition unit 4, and wind speed inside the ventilation hood 3 as learning data.

[0029] As described above, the sound pressure data inside the ventilation hood 3 is either the sound pressure data received by the calculation unit 7 from the sound pressure data acquisition unit 4, or the sound pressure data processed by the calculation unit 7 from the sound pressure data acquisition unit 4. Specifically, the sound pressure data inside the ventilation hood 3 includes the waveform of the sound pressure data inside the ventilation hood 3, the frequency spectrum of the sound pressure data inside the ventilation hood 3, etc. The sound pressure data inside the ventilation hood 3 is input from the calculation unit 7 to the learning data acquisition unit 5a.

[0030] The combined data is information that affects the relationship between the sound pressure data inside the ventilation hood 3 and the wind speed inside the ventilation hood 3. The combined data is input to the learning data acquisition unit 5a from the terminal device 8 by an administrator or the like. The combined data includes at least the shape name information of the ventilation member. The shape of the ventilation member affects the wind noise during ventilation, depending on its shape. In other words, the shape of the ventilation member affects the sound pressure data inside the ventilation hood 3, and is therefore information that should be considered when estimating the wind speed inside the ventilation hood 3. However, detailed data on the shape of the ventilation member is not necessarily required; if the shape name information of the ventilation member is available, it is possible to associate the shape of the ventilation member with the wind speed inside the ventilation hood 3. Note that the shape of the ventilation member includes both the shape and dimensions (size) of the ventilation member.

[0031] In addition to the shape information of the ventilation member, the combination data may further include one or more of the following: pressure loss information of the air passage 9 shown in Figure 1, position information of the filter 2a, position information of the blades 2d, 2e, and 3a shown in Figures 2 to 4, data showing the relationship between the airflow and static pressure of the ventilation fan 1, shape data of the ventilation hood 3, material data of the ventilation hood 3, and information on the presence or absence of sound-absorbing material in the ventilation hood 3.

[0032] The pressure loss information for the air passage 9 includes at least one of the following: information on the presence or absence of filter 2a, information on the presence or absence of blade 3a installed inside the ventilation hood 3, and information on the presence or absence of blade 2d of shutter 2b or blade 2e of louver 2c. The presence of filter 2a or blades 2d, 2e, or 3a inside the air passage 9 causes pressure loss in the air flowing through the air passage 9. In addition, the presence of blades 2d, 2e, or 3a inside the air passage 9 makes it easier for wind noise to occur. Therefore, by including at least one of the following information in the combined data—information on the presence or absence of filter 2a, information on the presence or absence of blade 3a installed inside the ventilation hood 3, and information on the presence or absence of blade 2d of shutter 2b or blade 2e of louver 2c—the accuracy of inferring the wind speed inside the ventilation hood 3 is improved.

[0033] The positional information of filter 2a indicates the relative position of filter 2a to sound pressure data acquisition unit 4 within the airflow path 9. For example, if filter 2a is installed between sound pressure data acquisition unit 4 and ventilation fan 1, the sound pressure data of the ventilation fan 1 itself is acquired by sound pressure data acquisition unit 4 in a state where it is slightly mitigated by filter 2a. Therefore, even with the same wind speed, the sound pressure data acquired by sound pressure data acquisition unit 4 changes depending on the position of filter 2a relative to sound pressure data acquisition unit 4. For this reason, including the positional information of filter 2a in the combined data improves the accuracy of inferring the wind speed inside the ventilation hood 3.

[0034] The positional information of blades 2d, 2e, and 3a indicates the relative position of the sound pressure data acquisition unit 4 and blades 2d, 2e, and 3a within the airflow path 9. For example, if blades 2d, 2e, and 3a are installed between the sound pressure data acquisition unit 4 and the ventilation fan 1, the sound pressure data of the ventilation fan 1 itself is acquired by the sound pressure data acquisition unit 4 in a state where it is slightly mitigated by the blades 2d, 2e, and 3a. Therefore, even at the same wind speed, the sound pressure data acquired by the sound pressure data acquisition unit 4 changes depending on the position of blades 2d, 2e, and 3a relative to the sound pressure data acquisition unit 4. For this reason, including the positional information of blades 2d, 2e, and 3a in the combined data improves the accuracy of wind speed inference within the ventilation hood 3.

[0035] The data showing the relationship between the airflow and static pressure of the ventilation fan 1 is information that shows the relationship between the airflow and static pressure of the ventilation fan 1 when the rotational speed of the ventilation fan 1 is adjusted by a speed control device such as an inverter. Even if the ventilation fan 1 has the same model name, the sound pressure data acquired by the sound pressure data acquisition unit 4 will differ depending on whether the rotational speed of the ventilation fan 1 is adjusted by the speed control device or not. For this reason, when the rotational speed of the ventilation fan 1 is adjusted by the speed control device, including the data showing the relationship between the airflow and static pressure of the ventilation fan 1 in the combined data improves the accuracy of the inference of the wind speed inside the ventilation hood 3.

[0036] The shape data of the ventilation hood 3 is information indicating the shape and dimensions of the ventilation hood 3. The material data of the ventilation hood 3 is information indicating the material of the ventilation hood 3. If the shape and material of the ventilation hood 3 are different, the sound pressure data acquired by the sound pressure data acquisition unit 4 will change even at the same wind speed. Therefore, by including the shape data and material data of the ventilation hood 3 in the combined data, the accuracy of inferring the wind speed inside the ventilation hood 3 is improved.

[0037] The information regarding the presence or absence of sound-absorbing material in the ventilation hood 3 indicates whether or not sound-absorbing material is present on the inner wall surface of the ventilation hood 3. Sound-absorbing material is sometimes installed on the inner wall surface of the ventilation hood 3 to suppress noise, but sound-absorbing material also affects sound reflection and the sound pressure inside the air passage 9. Therefore, including the information regarding the presence or absence of sound-absorbing material in the ventilation hood 3 in the combined data improves the accuracy of inferring the air velocity inside the ventilation hood 3.

[0038] The data relating to the sound pressure data acquisition unit 4 is information that affects the relationship between the sound pressure data inside the ventilation hood 3 and the wind speed inside the ventilation hood 3. The data relating to the sound pressure data acquisition unit 4 is input to the learning data acquisition unit 5a from the terminal device 8 by an administrator or the like. The data relating to the sound pressure data acquisition unit 4 includes at least one of the following: the installation position of the sound pressure data acquisition unit 4 inside the ventilation hood 3 and the shape of the sound pressure data acquisition unit 4. The installation position of the sound pressure data acquisition unit 4 inside the ventilation hood 3 and the shape of the sound pressure data acquisition unit 4, like the ventilation members, affect the sound pressure data inside the ventilation hood 3, and therefore should be considered when estimating the wind speed inside the ventilation hood 3.

[0039] The installation position of the sound pressure data acquisition unit 4 within the ventilation hood 3 is represented, for example, by how far the sound pressure data acquisition unit 4 is from an arbitrary point within the ventilation hood 3 in the X, Y, and Z directions. In other words, the installation position of the sound pressure data acquisition unit 4 within the ventilation hood 3 is information indicating the relative position between the reference point within the ventilation hood 3 and the sound pressure data acquisition unit 4.

[0040] Similar to the ventilation member, detailed shape data of the sound pressure data acquisition unit 4 is not necessarily required. If the part number information of the sound pressure data acquisition unit 4 is available, it is possible to associate the shape of the sound pressure data acquisition unit 4 with the airflow velocity inside the ventilation hood 3. Note that the shape of the sound pressure data acquisition unit 4 includes both its shape and dimensions (size).

[0041] When active noise cancellation technology is applied to the ventilation component, the sound pressure data acquisition unit 4 may include a microphone and a component that performs active noise cancellation. Components that perform active noise cancellation include speakers and speaker-related components. Hereinafter, speakers and speaker-related components will be collectively referred to as "speaker components." The installation position and shape of the speaker components within the ventilation hood 3, like the ventilation component, affect the sound pressure data within the ventilation hood 3, and therefore should be considered when estimating the wind speed within the ventilation hood 3.

[0042] The installation position of the speaker component within the ventilation hood 3 is represented, for example, by how far the speaker component is from an arbitrary point within the ventilation hood 3 in the X, Y, and Z directions. In other words, the installation position of the speaker component within the ventilation hood 3 is information indicating the relative position of the speaker component to the reference point within the ventilation hood 3.

[0043] Similar to ventilation components, detailed data on the shape of the speaker component is not necessarily required. If the speaker component's model name is available, it is possible to associate the speaker component's shape with the airflow velocity inside the ventilation hood 3. Note that the term "speaker component shape" includes both the shape and dimensions (size) of the speaker component.

[0044] The wind speed inside the ventilation hood 3 is obtained, for example, by measuring the actual wind speed inside the ventilation hood 3 when the combination of sound pressure data inside the ventilation hood 3, combination data including the model name information of the ventilation member, and data related to the sound pressure data acquisition unit 4 is changed. The actual wind speed inside the ventilation hood 3 is measured using an anemometer or the like.

[0045] The model generation unit 5b shown in Figure 5 learns the wind speed inside the ventilation hood 3 that corresponds to the sound pressure data inside the ventilation hood 3, the combination data including the model name information of the ventilation member, and the data related to the sound pressure data acquisition unit 4, based on the training data generated based on the combination of sound pressure data inside the ventilation hood 3, combination data including the model name information of the ventilation member, data related to the sound pressure data acquisition unit 4, and the wind speed inside the ventilation hood 3, which are sent from the training data acquisition unit 5a. In other words, when the model generation unit 5b receives the sound pressure data inside the ventilation hood 3, the combination data including the model name information of the ventilation member, and data related to the sound pressure data acquisition unit 4 as input, it generates a trained model that outputs the wind speed inside the ventilation hood 3 that corresponds to the sound pressure data inside the ventilation hood 3, the combination data including the model name information of the ventilation member, and the data related to the sound pressure data acquisition unit 4. Here, the training data is data that associates the sound pressure data inside the ventilation hood 3, the combination data including the model name information of the ventilation member, and data related to the sound pressure data acquisition unit 4 with the wind speed inside the ventilation hood 3.

[0046] The model generation unit 5b can use known learning algorithms such as supervised learning to generate a trained model. Here, we will describe the case where the model generation unit 5b performs supervised learning using a neural network.

[0047] The model generation unit 5b learns the wind speed inside the ventilation hood 3, corresponding to the sound pressure data inside the ventilation hood 3, the combined data including the shape name information of the ventilation members, and the data related to the sound pressure data acquisition unit 4, using so-called supervised learning according to a neural network model. Here, supervised learning is a method in which a set of input and result (label) data is provided to the learning device 5, the features of these learning data are learned, and the result is inferred from the input.

[0048] A neural network consists of an input layer made up of multiple neurons, a hidden layer (intermediate layer) also made up of multiple neurons, and an output layer also made up of multiple neurons. The hidden layer can be one or more layers.

[0049] Figure 6 shows the configuration of the neural network used by the learning device 5 according to Embodiment 1. For example, in a three-layer neural network as shown in Figure 6, when multiple data are input to the input layers X1 to X3, these values ​​are multiplied by weights w11 to w16 and input to the hidden layers Y1 to Y2. The result is then further multiplied by weights w21 to w26 and output from the output layers Z1 to Z3. This output result varies depending on the values ​​of the weights w11 to w16 and w21 to w26.

[0050] The neural network of this embodiment learns the wind speed inside the ventilation hood 3 corresponding to the sound pressure data inside the ventilation hood 3, the combination data including the model name information of the ventilation member, and the data related to the sound pressure data acquisition unit 4, through so-called supervised learning, according to the training data (dataset) generated based on the combination of sound pressure data inside the ventilation hood 3 acquired by the training data acquisition unit 5a, combination data including the model name information of the ventilation member, data related to the sound pressure data acquisition unit 4.

[0051] In other words, the neural network learns by inputting combined data including sound pressure data inside the ventilation hood 3 and shape information of ventilation components, as well as data related to the sound pressure data acquisition unit 4, into input layers X1 to X3, and adjusting weights w11 to w16 and w21 to w26 so that the results output from output layers Z1 to Z3 approach the wind speed inside the ventilation hood 3.

[0052] The model generation unit 5b generates and outputs a trained model by performing the training described above.

[0053] The trained model storage unit 5c stores the trained model output from the model generation unit 5b.

[0054] Next, the processing procedure of the learning device 5 will be explained using Figure 7. Figure 7 is a flowchart showing the processing procedure of the learning device 5 according to Embodiment 1.

[0055] The learning device 5 first acquires learning data (step S1). Specifically, the learning data acquisition unit 5a acquires sound pressure data inside the ventilation hood 3, combination data including the model name information of the ventilation members, data related to the sound pressure data acquisition unit 4, and wind speed inside the ventilation hood 3 as learning data, and associates these to create learning data. The learning data acquisition unit 5a may acquire the sound pressure data inside the ventilation hood 3, the combination data including the model name information of the ventilation members, the data related to the sound pressure data acquisition unit 4, and wind speed inside the ventilation hood 3 at the same time, or at different times. In other words, if the learning data acquisition unit 5a can associate the sound pressure data inside the ventilation hood 3, the combination data including the model name information of the ventilation members, and the data related to the sound pressure data acquisition unit 4 with the wind speed inside the ventilation hood 3, the sound pressure data inside the ventilation hood 3, the combination data including the model name information of the ventilation members, the data related to the sound pressure data acquisition unit 4, and the wind speed inside the ventilation hood 3 can be acquired at any timing.

[0056] The learning device 5 then performs a learning process (step S2). Specifically, the model generation unit 5b performs a learning process according to the learning data, which is a combination of sound pressure data inside the ventilation hood 3 acquired by the learning data acquisition unit 5a, combination data including the shape name information of the ventilation members, data related to the sound pressure data acquisition unit 4, and the wind speed inside the ventilation hood 3. The model generation unit 5b, for example, learns the wind speed inside the ventilation hood 3 corresponding to the sound pressure data inside the ventilation hood 3, the combination data including the shape name information of the ventilation members, and the data related to the sound pressure data acquisition unit 4, using so-called supervised learning according to the learning data, and generates a trained model.

[0057] The learning device 5 then stores the trained model in the trained model storage unit 5c (step S3). Specifically, the model generation unit 5b outputs the trained model generated in step S2, and the trained model storage unit 5c stores the trained model.

[0058] The learning device 5 learns the wind speed inside the ventilation hood 3 that corresponds to the sound pressure data inside the ventilation hood 3, the combination data including the model name information of the ventilation members, the data related to the sound pressure data acquisition unit 4, and the wind speed inside the ventilation hood 3 by repeating the above learning process while changing the combination of the learning data, which is the sound pressure data inside the ventilation hood 3, the combination data including the model name information of the ventilation members, and the data related to the sound pressure data acquisition unit 4.

[0059] Next, we will explain the application phase in which the inference device 6 uses the trained model generated by the learning device 5 to infer the wind speed inside the ventilation hood 3.

[0060] Figure 8 shows the configuration of the inference device 6 of the wind speed measurement system 100 according to Embodiment 1. The inference device 6 comprises an inference data acquisition unit 6a and an inference unit 6b. The inference unit 6b is connected to the trained model storage unit 5c. Figure 8 also shows the calculation unit 7. The calculation unit 7 is not a component of the inference device 6, so it is shown with a dashed line.

[0061] The inference data acquisition unit 6a acquires sound pressure data from within the ventilation hood 3, combined data including the model name information of the ventilation members, and data related to the sound pressure data acquisition unit 4 as inference data. The sound pressure data from within the ventilation hood 3 is input to the inference data acquisition unit 6a from the calculation unit 7. The combined data is input to the inference data acquisition unit 6a from the terminal device 8 by an administrator or the like. The data related to the sound pressure data acquisition unit 4 is input to the inference data acquisition unit 6a from the terminal device 8 by an administrator or the like. In addition to the model name information of the ventilation members, the combined data may further include one or more of the following: pressure loss information of the air passage 9 shown in Figure 1, position information of the filter 2a, position information of the blades 2d, 2e, 3a shown in Figures 2 to 4, data showing the relationship between the airflow and static pressure of the ventilation fan 1, shape data of the ventilation hood 3, material data of the ventilation hood 3, and information on the presence or absence of sound-absorbing material in the ventilation hood 3.

[0062] The inference unit 6b uses the trained model stored in the trained model storage unit 5c to infer the wind speed inside the ventilation hood 3, which corresponds to the sound pressure data inside the ventilation hood 3, the combination data including the shape name information of the ventilation members, and the data related to the sound pressure data acquisition unit 4. In other words, the inference unit 6b inputs the combination data including the sound pressure data inside the ventilation hood 3 and the shape name information of the ventilation members, acquired by the inference data acquisition unit 6a, and the data related to the sound pressure data acquisition unit 4 into the trained model stored in the trained model storage unit 5c and performs inference. As a result, the inference unit 6b obtains the wind speed inside the ventilation hood 3 as an inference result. The inference unit 6b outputs the wind speed inside the ventilation hood 3, which is the inference result, to the calculation unit 7.

[0063] Next, the processing procedure of the inference process by the inference device 6 and the output processing by the calculation unit 7 will be explained using Figure 9. Figure 9 is a flowchart showing the processing procedure of the inference process by the inference device 6 and the output processing by the calculation unit 7 according to Embodiment 1.

[0064] The inference device 6 first acquires inference data (step S11). Specifically, the inference data acquisition unit 6a acquires sound pressure data from inside the ventilation hood 3, which is used for inferring the wind speed inside the ventilation hood 3, combined data including the model name information of the ventilation members, and data related to the sound pressure data acquisition unit 4.

[0065] Next, the inference device 6 inputs the inference data acquired in step S11 into the trained model stored in the trained model storage unit 5c (step S12) to obtain an inference result. Specifically, the inference unit 6b acquires the trained model from the trained model storage unit 5c. The inference unit 6b inputs the sound pressure data inside the ventilation hood 3 acquired by the inference data acquisition unit 6a, combined data including the shape name information of the ventilation members, and data related to the sound pressure data acquisition unit 4 into the acquired trained model. Accordingly, the inference unit 6b acquires the wind speed inside the ventilation hood 3 output from the trained model. As a result, the wind speed inside the ventilation hood 3 can be inferred from the sound pressure data inside the ventilation hood 3 acquired by the inference data acquisition unit 6a, combined data including the shape name information of the ventilation members, and data related to the sound pressure data acquisition unit 4.

[0066] The inference device 6 then outputs the wind speed inside the ventilation hood 3, which is the inference result (step S13). Specifically, the inference unit 6b outputs the wind speed inside the ventilation hood 3 to the calculation unit 7.

[0067] Next, the calculation unit 7 outputs the wind speed inside the ventilation hood 3, which is the inference result output from the inference device 6, to the terminal device 8 (step S14). Also, if information on the air passage area inside the ventilation hood 3 has been provided to the calculation unit 7, the calculation unit 7 uses the wind speed inside the ventilation hood 3, which is the inference result output from the inference device 6, to calculate the airflow rate inside the ventilation hood 3, and outputs the calculated airflow rate inside the ventilation hood 3 to the terminal device 8 (step S14). Specifically, the calculation unit 7 calculates the airflow rate inside the ventilation hood 3 by multiplying the wind speed inside the ventilation hood 3 by the air passage area inside the ventilation hood 3.

[0068] In this embodiment, the inference device 6 is described as using a trained model generated by the learning device 5, which constitutes the wind speed measurement system 100, to infer the wind speed inside the ventilation hood 3. However, the invention is not limited to this. For example, the inference device 6 may acquire a trained model generated outside the wind speed measurement system 100 and use this trained model to infer the wind speed inside the ventilation hood 3.

[0069] Furthermore, at least one of the learning device 5 and the inference device 6 may be connected to the wind speed measurement system 100, for example, via a network. Also, at least one of the learning device 5 and the inference device 6 may be a separate device from the wind speed measurement system 100. Moreover, at least one of the learning device 5 and the inference device 6 may reside on a cloud server.

[0070] Furthermore, although this embodiment describes the case where supervised learning is applied to the learning algorithm used by the model generation unit 5b, the learning algorithm is not limited to supervised learning. In addition to supervised learning, reinforcement learning, unsupervised learning, or semi-supervised learning can also be applied as the learning algorithm.

[0071] Furthermore, the model generation unit 5b may learn the wind speed inside the ventilation hood 3 according to the training data generated for multiple wind speed measurement systems 100.

[0072] Furthermore, the model generation unit 5b may acquire training data from multiple wind speed measurement systems 100 used in the same area. Alternatively, the model generation unit 5b may learn the wind speed inside the ventilation hood 3 using training data collected from multiple wind speed measurement systems 100 operating independently in different areas.

[0073] Furthermore, it is possible to add or remove the wind speed measurement system 100, which collects training data, from the target midway through the process.

[0074] Furthermore, a learning device 5 that has learned the wind speed inside the ventilation hood 3 for one wind speed measurement system 100 may be applied to another wind speed measurement system 100. In addition, the learning device 5 may relearn the wind speed inside the ventilation hood 3 for the other wind speed measurement system 100 to update the learned model.

[0075] Furthermore, the learning algorithm used in the model generation unit 5b can also be deep learning, which learns to extract the features themselves. The model generation unit 5b may also perform machine learning according to other known methods, such as genetic programming, functional logic programming, or support vector machines.

[0076] Next, the effects of Embodiment 1 will be described.

[0077] In this embodiment, as shown in Figures 1, 5, and 8, the wind speed measurement system 100 includes a learning device 5, a learned model storage unit 5c, and an inference device 6. This allows the wind speed inside the ventilation hood 3 to be inferred using a learned model generated by learning the wind speed inside the ventilation hood 3 corresponding to the sound pressure data inside the ventilation hood 3, combined data including the shape name information of the ventilation members, and data related to the sound pressure data acquisition unit 4. Therefore, in this embodiment, the wind speed inside the ventilation hood 3 can be estimated without installing a dedicated sensor for calculating the wind speed inside the ventilation hood 3.

[0078] In this embodiment, the combination data may further include, in addition to the model name information of the ventilation member, one or more of the following: pressure loss information of the air passage 9 shown in Figure 1, position information of the filter 2a, position information of the blades 2d, 2e, and 3a shown in Figures 2 to 4, data showing the relationship between the airflow and static pressure of the ventilation fan 1, shape data of the ventilation hood 3, material data of the ventilation hood 3, and information on the presence or absence of sound-absorbing material in the ventilation hood 3. Doing so can further improve the accuracy of estimating the airflow velocity inside the ventilation hood 3.

[0079] As shown in Figure 1, when a ventilation fan 1 and ventilation hood 3 are installed on the wall 10a of a building 10 and blow air from the outside to the inside of the building 10 to ventilate the interior, it is important to ensure a sufficient amount of airflow to be blown into the interior. One way to manage the amount of airflow to be blown into the interior is to manage the rotation speed of the blades 1c, for example, but the rotation speed of the blades 1c and the airflow do not necessarily correspond. Therefore, it is desirable to calculate and directly manage the amount of airflow to be blown into the interior. In this embodiment, the calculation unit 7 uses the wind speed inside the ventilation hood 3, which is the inference result output from the inference device 6, to calculate the airflow inside the ventilation hood 3 by multiplying the wind speed inside the ventilation hood 3 by the air passage area inside the ventilation hood 3. This makes it possible to directly manage the amount of airflow to be blown into the interior. The above also applies when ventilation is performed by blowing air from the inside to the outside of the building 10. That is, even when ventilation is performed by blowing air from the inside to the outside of the building 10, it is important to ensure a sufficient amount of airflow to be blown outside. In this embodiment, as described above, the airflow within the ventilation hood 3 can be calculated, and therefore the airflow to be sent outside can also be directly controlled.

[0080] Factors that affect the amount of air blown into a room include, for example, the pressure loss in the air passage 9 before and after the ventilation fan 1. When the pressure loss in the air passage 9 increases, the amount of air blown into the room decreases, but the rotation speed of the blades 1c does not necessarily change significantly at that time. In other words, even if only the change in the rotation speed of the blades 1c is measured, if the change in the rotation speed of the blades 1c is small, it may be difficult to determine the change in the amount of air blown into the room. In this embodiment, by including at least one of the following as pressure loss information for the air passage 9—information on the presence or absence of the filter 2a, information on the presence or absence of the blades 3a installed in the ventilation hood 3, and information on the presence or absence of the blades 2d of the shutter 2b or the blades 2e of the louver 2c—in the combined data, the accuracy of inferring the wind speed inside the ventilation hood 3 is improved, making it easier to determine the change in the amount of air blown into the room. The above also applies when ventilation is performed by blowing air from the inside to the outside of the building 10. In other words, even when ventilating the interior of building 10 by blowing air from inside to outside, if the pressure loss in the air passage 9 increases, the amount of air blown outside will decrease. Also, even if only the change in the rotation speed of the blade 1c is measured, if the change in the rotation speed of the blade 1c is small, it may be difficult to determine the change in the amount of air blown outside. In this embodiment, by including the pressure loss information of the air passage 9 described above in the combined data, the accuracy of inferring the wind speed inside the ventilation hood 3 is improved, making it easier to determine the change in the amount of air blown outside.

[0081] Next, we will describe the hardware configuration of the learning device 5 and the inference device 6 using Figure 10. Since the hardware configurations of the learning device 5 and the inference device 6 are the same, we will describe the hardware configuration of the learning device 5 here.

[0082] Figure 10 shows the hardware configuration of the learning device 5 according to Embodiment 1. The learning device 5 can be realized by a processor 200, memory 300, input device 400, and output device 500. An example of the processor 200 is a CPU (Central Processing Unit, also called a microprocessor, microcomputer, or DSP (Digital Signal Processor)) or a system LSI (Large Scale Integration). An example of the memory 300 is a semiconductor memory such as RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable Read Only Memory), or EEPROM (Registered Trademark) (Electrically Erasable Programmable Read Only Memory).

[0083] The learning device 5 is realized by the processor 200 reading and executing a computer-executable learning program stored in memory 300 for performing the operations of the learning device 5. The learning program, which is the program for performing the operations of the learning device 5, can also be described as a program that causes the computer to execute the procedures or methods of the learning device 5. In the case of the inference device 6, it is realized by the processor 200 reading and executing a computer-executable inference program stored in memory 300 for performing the operations of the inference device 6.

[0084] The learning program executed by the learning device 5 has a modular configuration that includes a learning data acquisition unit 5a and a model generation unit 5b, which are loaded into the main memory and generated in the main memory.

[0085] The input device 400 receives sound pressure data from inside the ventilation hood 3, combined data including the model name information of the ventilation components, data related to the sound pressure data acquisition unit 4, and the wind speed inside the ventilation hood 3, and sends them to the processor 200. The memory 300 is used as temporary memory when the processor 200 performs various processes. The memory 300 stores trained models, etc. The output device 500 outputs the trained models to the inference device 6. In the case of the inference device 6, the output device 500 outputs the inference results to an external device.

[0086] The learning program may be provided as a computer program product, stored on a computer-readable storage medium, in an installable or executable file format. Alternatively, the learning program may be provided to the learning device 5 via a network such as the Internet. Furthermore, some functions of the learning device 5 may be implemented using dedicated hardware, such as dedicated circuits, while others are implemented using software or firmware.

[0087] The configurations shown in the above embodiments are merely examples, and can be combined with other known technologies. It is also possible to omit or modify parts of the configuration without departing from the gist of the invention.

[0088] The various aspects of this disclosure are summarized below as an appendix.

[0089] (Note 1) A learning device for learning the wind speed inside a ventilation hood, A learning data acquisition unit that acquires learning data including sound pressure data inside the ventilation hood, model name information of ventilation components including the ventilation hood, a ventilation fan that ventilates the room through the ventilation hood, and ventilation fan attachment components attached to the ventilation fan, data relating to a sound pressure data acquisition unit that acquires the sound pressure data, and wind speed inside the ventilation hood, A model generation unit generates a trained model for inferring the wind speed inside the ventilation hood from the sound pressure data, the combination data, and the data relating to the sound pressure data acquisition unit, using the aforementioned training data. A learning device characterized by being equipped with the following features. (Note 2) An inference device for inferring the wind speed inside a ventilation hood, An inference data acquisition unit acquires inference data including sound pressure data within the ventilation hood, combination data including part number information of ventilation components including the ventilation hood, a ventilation fan that ventilates the room through the ventilation hood, and ventilation fan accessory components attached to the ventilation fan, and data relating to the sound pressure data acquisition unit that acquires the sound pressure data. An inference unit that uses a trained model to infer the wind speed inside the ventilation hood from the sound pressure data, the combination data, and the data related to the sound pressure data acquisition unit, An inference device characterized by comprising: (Note 3) The sound pressure data acquisition unit includes a microphone, The learning device according to Appendix 1, characterized in that the data relating to the sound pressure data acquisition unit includes at least one of the following: the model name information of the microphone and the installation position of the microphone within the ventilation hood. (Note 4) The sound pressure data acquisition unit includes a microphone, The inference device according to Appendix 2, characterized in that the data relating to the sound pressure data acquisition unit includes at least one of the following: the model name information of the microphone and the installation position of the microphone within the ventilation hood. (Note 5) The learning device according to Appendix 1 or 3, characterized in that the aforementioned combination data includes pressure loss information of the airflow path. (Note 6) The learning device according to any one of appendices 1, 3, or 5, characterized in that the combination data includes data showing the relationship between the airflow and static pressure of the ventilation fan. (Note 7) The learning device according to any one of appendices 1, 3, 5, or 6, characterized in that the combination data includes at least one of the following: information on the presence or absence of a filter which is an accessory component of the ventilation fan; information on the presence or absence of a blade which is installed inside the ventilation hood; and information on the presence or absence of a blade which is a shutter or louver which is an accessory component of the ventilation fan. (Note 8) The learning device according to Appendix 7, characterized in that the combination data includes the position information of the filter. (Note 9) The learning device according to appendix 7 or 8, characterized in that the combination data includes position information of the blade. (Note 10) The learning device according to any one of appendices 1, 3, 5 to 9, characterized in that the combination data includes at least one of the shape data of the ventilation hood and the material data of the ventilation hood. (Note 11) The learning device according to any one of appendices 1, 3, 5 to 10, characterized in that the combination data includes information on the presence or absence of sound-absorbing material in the ventilation hood. (Note 12) A ventilation member including a ventilation hood, a ventilation fan that ventilates the room through the ventilation hood, and a ventilation fan accessory component attached to the ventilation fan, A sound pressure data acquisition unit that acquires sound pressure data inside the ventilation hood, A learning device described in any one of the appendices 1, 3, 5 through 10, The inference device described in Appendix 2 or 4, A calculation unit that outputs the wind speed inside the ventilation hood, inferred by the inference device, to a terminal device, A wind speed measurement system characterized by having the following features. (Note 13) The wind speed measurement system according to Appendix 12, characterized in that the calculation unit calculates the airflow rate inside the ventilation hood by multiplying the wind speed inside the ventilation hood by the airflow area inside the ventilation hood. [Explanation of Symbols]

[0090] 1 Ventilation fan, 1a Frame, 1b Motor, 1c Blades, 2 Ventilation fan accessories, 2a Filter, 2b Shutter, 2c Louver, 2d, 2e, 3a Blades, 3 Ventilation hood, 4 Sound pressure data acquisition unit, 5 Learning device, 5a Learning data acquisition unit, 5b Model generation unit, 5c Learned model storage unit, 6 Inference device, 6a Inference data acquisition unit, 6b Inference unit, 7 Calculation unit, 8 Terminal device, 9 Air passage, 10 Building, 10a Wall, 10b Opening, 100 Wind speed measurement system, 200 Processor, 300 Memory, 400 Input device, 500 Output device.

Claims

1. A learning device for learning the wind speed inside a ventilation hood, A learning data acquisition unit that acquires learning data including sound pressure data inside the ventilation hood, model name information of ventilation components including the ventilation hood, a ventilation fan that ventilates the room through the ventilation hood, and ventilation fan attachment components attached to the ventilation fan, data relating to a sound pressure data acquisition unit that acquires the sound pressure data, and wind speed inside the ventilation hood, A model generation unit generates a trained model for inferring the wind speed inside the ventilation hood from the sound pressure data, the combination data, and the data relating to the sound pressure data acquisition unit, using the aforementioned training data. A learning device characterized by being equipped with the following features.

2. An inference device for inferring the wind speed inside a ventilation hood, An inference data acquisition unit acquires inference data including sound pressure data within the ventilation hood, combination data including part number information of ventilation components including the ventilation hood, a ventilation fan that ventilates the room through the ventilation hood, and ventilation fan accessory components attached to the ventilation fan, and data relating to the sound pressure data acquisition unit that acquires the sound pressure data. An inference unit that uses a trained model to infer the wind speed inside the ventilation hood from the sound pressure data, the combination data, and the data related to the sound pressure data acquisition unit, An inference device characterized by comprising:

3. The sound pressure data acquisition unit includes a microphone, The learning device according to claim 1, characterized in that the data relating to the sound pressure data acquisition unit includes at least one of the following: the model name information of the microphone and the installation position of the microphone within the ventilation hood.

4. The sound pressure data acquisition unit includes a microphone, The inference device according to claim 2, characterized in that the data relating to the sound pressure data acquisition unit includes at least one of the following: the model name information of the microphone and the installation position of the microphone within the ventilation hood.

5. The learning device according to claim 1, characterized in that the aforementioned combination data includes pressure loss information of the airflow path.

6. The learning device according to claim 1, characterized in that the combination data includes data showing the relationship between the airflow and static pressure of the ventilation fan.

7. The learning device according to claim 1, characterized in that the combination data includes at least one of the following: information on the presence or absence of a filter which is an accessory component of the ventilation fan; information on the presence or absence of a blade which is installed inside the ventilation hood; and information on the presence or absence of a blade which is a shutter or louver which is an accessory component of the ventilation fan.

8. The learning device according to claim 7, characterized in that the combination data includes the position information of the filter.

9. The learning device according to claim 7, characterized in that the combination data includes position information of the blade.

10. The learning device according to claim 1, characterized in that the combination data includes at least one of the shape data of the ventilation hood and the material data of the ventilation hood.

11. The learning device according to claim 1, characterized in that the combination data includes information on the presence or absence of sound-absorbing material in the ventilation hood.

12. A ventilation member including a ventilation hood, a ventilation fan that ventilates the room through the ventilation hood, and a ventilation fan accessory component attached to the ventilation fan, A sound pressure data acquisition unit that acquires sound pressure data inside the ventilation hood, The learning device according to claim 1, The inference device according to claim 2, A calculation unit that outputs the wind speed inside the ventilation hood, inferred by the inference device, to a terminal device, A wind speed measurement system characterized by having the following features.

13. The wind speed measurement system according to claim 12, characterized in that the calculation unit calculates the air volume inside the ventilation hood by multiplying the wind speed inside the ventilation hood by the air passage area inside the ventilation hood.

Images