CN114609410B - Portable wind characteristic measuring equipment based on acoustic signals and intelligent algorithm - Google Patents
Portable wind characteristic measuring equipment based on acoustic signals and intelligent algorithm Download PDFInfo
- Publication number
- CN114609410B CN114609410B CN202210305160.1A CN202210305160A CN114609410B CN 114609410 B CN114609410 B CN 114609410B CN 202210305160 A CN202210305160 A CN 202210305160A CN 114609410 B CN114609410 B CN 114609410B
- Authority
- CN
- China
- Prior art keywords
- wind
- noise reduction
- shell
- acoustic signals
- wind speed
- 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
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01P—MEASURING LINEAR OR ANGULAR SPEED, ACCELERATION, DECELERATION, OR SHOCK; INDICATING PRESENCE, ABSENCE, OR DIRECTION, OF MOVEMENT
- G01P5/00—Measuring speed of fluids, e.g. of air stream; Measuring speed of bodies relative to fluids, e.g. of ship, of aircraft
- G01P5/24—Measuring speed of fluids, e.g. of air stream; Measuring speed of bodies relative to fluids, e.g. of ship, of aircraft by measuring the direct influence of the streaming fluid on the properties of a detecting acoustical wave
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01L—MEASURING FORCE, STRESS, TORQUE, WORK, MECHANICAL POWER, MECHANICAL EFFICIENCY, OR FLUID PRESSURE
- G01L11/00—Measuring steady or quasi-steady pressure of a fluid or a fluent solid material by means not provided for in group G01L7/00 or G01L9/00
- G01L11/04—Measuring steady or quasi-steady pressure of a fluid or a fluent solid material by means not provided for in group G01L7/00 or G01L9/00 by acoustic means
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Aviation & Aerospace Engineering (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention discloses portable wind characteristic measuring equipment and an intelligent algorithm based on acoustic signals, which comprise a condenser microphone, a processor and an ANC active noise reduction device, wherein the condenser microphone, the processor and the ANC active noise reduction device are arranged in a cuboid equipment shell; one side surface of the shell is provided with a wind signal collecting port, the other three side surfaces are provided with environment bottom noise collecting ports, and a condenser microphone is respectively arranged in the shell close to each collecting port; an ANC active noise reduction device is arranged at the central position in the shell, and a processor is arranged at the upper part of the ANC active noise reduction device; the four condenser microphones are connected in series with the ANC active noise reduction device and the processor. The measuring equipment of the invention takes a condenser microphone of the equipment as a sensor, has no external sensor, directly reads in multi-source acoustic signals in real time through a sound transmission device of the equipment and carries out noise reduction treatment, and analyzes and maps the multi-source acoustic signals by using an artificial intelligent deep learning technology by utilizing the nonlinear mapping relation between the signal intensity and the wind speed, thereby achieving the purpose of quickly and efficiently restoring the natural wind characteristic.
Description
Technical Field
The invention relates to the technical field of civil engineering measurement of near-earth wind characteristics, in particular to portable measuring equipment for restoring natural wind characteristics including average wind, pulsating wind and pulsating wind power spectrum based on acoustic signals and an embedded intelligent algorithm.
Background
Wind characteristic measurement is an important content in the wind resistance design and construction process of bridges, high-rise buildings and wind power facilities, and the in-situ wind characteristic of a structure is a basic parameter and basis for carrying out wind resistance design and checking calculation on the structure. Currently, the in-situ wind characteristics of civil structures are mostly acquired by special instruments, and common wind measuring instruments comprise cup anemometers, hot wire anemometers, digital anemometers, acoustic anemometers, propeller type anemometers and the like. Except for the measuring instrument discs, most of the existing instruments need to be additionally connected with matched equipment such as a sensor, a collecting box, a processor and the like, need special power supply, are large in size, high in manufacturing cost, difficult to install and move, need special personnel for operation, and are inconvenient to carry by high-altitude operation, offshore operation and field construction personnel. Some small equipment based on mechanical propeller devices have single measurement objects, most of the small equipment is average wind speed, and the small equipment cannot reduce the characteristics of near complex wind such as average wind, pulsating wind, turbulence intensity, pulsating wind power spectrum and the like in a field in real time. Therefore, it is necessary to provide an optimized measuring method and apparatus for field operation, which is convenient for measurement, reduces manpower, has a certain precision for measurement, and reduces cost.
The prior patent CN111781396a discloses a wind speed measuring method, which comprises the following steps: 1: firstly, preparing different corrugated pipe detectors and calibrating; 2: placing the pipe in air to be tested, allowing the air to pass along the pipeline, making a sound, and presenting a stable state; 3: sound data are collected through a microphone, and sound is collected through a sound collecting device and converted into an electric signal; 4: collecting sound data of a left sound channel and a right sound channel to obtain two groups of processing results; 5: processing sound data, transmitting the sound data to terminal processing equipment, and obtaining the distribution of corresponding sound frequencies; 6: determining a value fn of the resonance frequency; 7: and (3) comparing data, fitting the data according to a parameter comparison table, automatically obtaining the speed of the wind speed passing through the corrugated pipe according to the comparison table of the sound frequency fn and the wind speed, and finally measuring to obtain the speed U of the wind at the moment. The method preliminarily solves the problem that the existing wind speed measuring method has limitation, and breaks through the limitation of the wind speed measuring method to a certain extent. However, the method still requires indirect use of a sensitive "probe" (bellows) to indirectly measure the mean windThe speed and convenience are not improved; automatically from the sound frequency f by data fitting using conventional spectral analysis methods n And obtaining the corresponding wind speed by comparing with a wind speed U, and not considering the influence of environmental noise. Under low wind speed, the environmental background noise is large, the peak frequency obtained by the frequency spectrum analysis cannot reflect the relation with the wind speed, if the wind speed is suitable for each measuring environment, an infinite number of comparison tables are formed manually according to the method of an author for numerous times, so that the wind speed measuring device can be suitable for each measuring environment, and the working efficiency is possibly low.
Disclosure of Invention
The invention aims to provide portable wind characteristic measuring equipment based on acoustic signals and a corresponding intelligent algorithm aiming at the problems of large equipment volume and inconvenience in installation and use and the problem that environmental background noise influences measurement in the conventional wind characteristic measuring method. The measuring device has the advantages of good portability, high processing efficiency and good man-machine interaction, realizes real-time wind characteristic test, is not limited to wind speed test, and can accurately restore near-average wind, pulsating wind, turbulence characteristics and pulsating power spectrum characteristics.
The invention provides portable wind characteristic measuring equipment based on acoustic signals, which structurally comprises a cuboid shell, a condenser microphone, a processor and an ANC active noise reduction device, wherein the condenser microphone, the processor and the ANC active noise reduction device are arranged in the shell. One side of the shell is a main channel and is provided with a wind signal collecting port, and wind to be measured enters the measuring equipment through the collecting port. The other three sides of the shell are auxiliary channels and are provided with an environment bottom noise collecting port. A condenser microphone is disposed within the housing proximate the primary passageway and proximate the secondary passageway, respectively. An ANC active noise reduction device is arranged at the center position in the shell, and a processor is arranged above the ANC active noise reduction device. The four condenser microphones are connected with the ANC active noise reduction device and the processor in series. The inner space of the shell is divided into four condenser microphones and an ANC active noise reduction device in different areas by arranging a plurality of partition plate partitions. The bottom of the shell is provided with a battery accommodating cavity, and the battery supplies power for the working operation of the whole measuring equipment.
The capacitor microphone is used for receiving a vibration signal of wind to the diaphragm and converting the vibration signal into an electric signal.
The ANC active noise reduction device supports floating point operation for a 32-bit DSP, the dominant frequency operation capability reaches 200MHz, and the signal-to-noise ratio reaches 103dB. The multi-source acoustic signals are subjected to noise reduction processing, namely, various interference noises in the environment are removed while the main direction target acoustic signals are protected.
The processor adopts a newly released fourth generation Raspberry Pi 4B development board of Raspberry pie, is embedded with the algorithm provided by the invention, mainly analyzes the acoustic signals subjected to noise reduction processing, is used for analyzing and calculating the acoustic signals of wind and accurately restoring the wind characteristics, and can transmit the signals to a cloud for remote analysis.
Preferably, the shell has the dimensions of 150mm in length, 70mm in width and 15mm in thickness; the middle of the side surface of one long edge of the shell is provided with a wind signal collecting port, and the middle of the side surface of the other three edges is provided with an environment bottom noise collecting port. The shell and the partition board are both made of ABS engineering plastics. Low cost, excellent heat and weather resistance and good impact resistance.
The intelligent algorithm based on the portable wind characteristic measuring equipment comprises the following steps:
s1, acquiring acoustic signals of wind
Starting up the portable wind characteristic measuring equipment, horizontally placing the portable wind characteristic measuring equipment on a wind measuring platform, placing a main channel on the windward side, and respectively collecting the acoustic signals of the wind at different wind speeds; and the rest auxiliary channels receive vibration signals of wind to the microphone diaphragm, the environment bottom noise collecting port receives environment noise, and the ANC active noise reduction device is used for carrying out noise reduction treatment on the acoustic signals of the wind signal collecting port.
S2, reducing average wind
And performing signal processing on the acoustic signals subjected to noise reduction processing by adopting Python language, extracting a Mel cepstrum coefficient, taking the coefficient matrix as the input of a one-dimensional convolution neural network, taking the actual wind speed as the output, and constructing a wind speed prediction model by using the one-dimensional convolution neural network, wherein the prediction model is used for reducing the average wind.
Selecting 6 layers of Conv1D to extract features, adding a layer of Max boosting 1D behind the 3 rd layer and the 6 th layer of Conv1D to retain main features and reduce calculated amount, wherein each convolution layer uses a linear rectification function ReLU as an activation function; finally, outputting a wind speed predicted value through the two fully-connected layers, wherein the predicted value of the wind speed tends to a real value more and more under the approximation of an MSE loss function; the optimization algorithm of the model adopts Adam, the model adopts 20 epochs and the size of each batch is 64; in order to evaluate the accuracy of the network model training and testing process, the algorithm self-defines a measurement function:
R＝1-SE/(SR+SE)
wherein SR is regression sum of squares, SE is residual sum of squares, and the closer R is to 1, the more accurate the model prediction is; the network uses two convolution kernels, 110 × 3 and 1 × 3, the convolution kernel moving step is 1, the pooling receptive field size is 1 × 3, and the step is 1.
The input is firstly mapped by a 110 multiplied by 39 one-dimensional convolution kernel, 1 90-dimensional characteristic vector is obtained after passing through a ReLu activation function, then 16 88-dimensional characteristic vectors are obtained after passing through 16 one-dimensional convolution kernels with the length of 3 and the ReLu activation function, then 16 28-dimensional characteristic vectors are obtained through 16 one-dimensional convolution kernels with the length of 3 and the ReLu activation function, 16 2-dimensional characteristic vectors are obtained after continuously passing through 3 convolution kernels and 1 pooling, and finally a predicted value is output through full connection layer activation by a Linear function, so that the purpose of predicting the wind speed is achieved.
S3, reducing the pulsating wind speed and the wind power spectrum
And processing the acoustic signals after noise reduction into data with the same sampling frequency as that of the cobra three-dimensional pulsating wind speed measuring instrument, building an LSTM neural network model by the acoustic signals and the pulsating wind speed collected by the cobra three-dimensional pulsating wind speed measuring instrument, and establishing a mapping relation between the acoustic signals and the pulsating wind speed so as to achieve the purpose of reducing the pulsating wind speed and the wind power spectrum by the acoustic signals.
The method comprises the steps of rapidly building a model by using a keras deep learning framework, establishing a Sequential model, adding an LSTM layer into the model, setting Dropout to be 0.15, adding a Dense layer to polymerize the dimensionality to be 1, using ReLU as an activation function for the activation function, setting a loss function to be Mean Square Error (MSE), adopting Adam as an optimization algorithm, adopting 50 epochs for the model, and setting the size of each batch to be 100.
Compared with the prior art, the invention has the advantages that:
the invention provides real-time wind characteristic testing equipment and an embedded algorithm which are good in portability, high in processing efficiency and good in man-machine interaction, and the invention is not limited to wind speed testing, but is an innovative invention capable of accurately restoring the characteristics of nearly average wind, pulsating wind, turbulence and pulsating power spectrum.
The measuring equipment is mainly provided with a microphone, a raspberry development board, an ANC active noise reduction device, a receiving channel and a power supply battery, except for a condenser microphone of the equipment, the equipment has no redundant external sensor, multi-source acoustic signals are directly read in real time through a sound transmission device of the equipment and subjected to noise reduction treatment, and then the multi-source acoustic signals are analyzed and mapped by using an artificial intelligent deep learning technology by utilizing the nonlinear mapping relation between the signal intensity and the wind speed, so that the purposes of man-machine interaction, convenience, rapidness and high-efficiency natural wind characteristic restoration are achieved. Meanwhile, the equipment is easy to carry, the measurement data can be shared to the cloud, the measurement and analysis are carried out synchronously, the manpower input can be reduced, and the dangerous operation can be reduced.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention.
Drawings
Fig. 1 is a perspective view of a portable wind characteristics measuring device based on acoustic signals according to the present invention.
Fig. 2 is a schematic structural diagram of the portable wind characteristic measuring device based on the acoustic signal.
FIG. 3 is a flow chart of extraction of Mel cepstral coefficients from the wind acoustic signals acquired by the apparatus of the present invention.
Fig. 4 and 1 are schematic diagrams of the DCNN prediction.
Fig. 5 shows the structure of LSTM prediction.
FIG. 6 shows a structure of LSTM.
FIG. 7 is a test layout diagram of the measuring equipment and the wind speed measuring instrument in the embodiment.
FIG. 8 is a diagram of training effect of the training set.
FIG. 9 is a test set prediction effect diagram.
FIG. 10, 1DCNN graph of the training and prediction results.
Fig. 11 is a diagram showing the effect of the pulsating wind speed prediction.
FIG. 12 is a wind speed spectrum prediction effect diagram.
Fig. 13, loss values of the model after 50 iterative training.
Detailed Description
The preferred embodiments of the present invention will be described in conjunction with the accompanying drawings, and it will be understood that they are described herein for the purpose of illustration and explanation and not limitation.
As shown in fig. 1 and 2, the structure of the portable wind characteristics measuring device based on acoustic signals provided by the present invention comprises four condenser microphones 1, a processor 2, and an ANC active noise reduction apparatus 3, wherein the condenser microphones 1, the processor 2, and the ANC active noise reduction apparatus 3 are disposed inside a rectangular parallelepiped housing 4. One side of the shell is provided with a wind signal collecting port 7 as a main channel. The other three sides of the shell are provided with an environment bottom noise collecting port 8 as a secondary channel. A condenser microphone is disposed within the housing proximate the primary passageway and proximate the secondary passageway, respectively. The wind to be measured enters the measuring device from the wind signal collecting port 7 and is received by the opposite capacitor microphone. The background noise enters from the ambient background noise collection port 8 and is received by the opposing condenser microphone. An ANC active noise reduction device is arranged at the center in the shell, and a processor 2 is arranged above the ANC active noise reduction device. Four condenser microphones 1 are connected in series with an ANC active noise reduction device 3 and a processor 3. The space in the shell is divided into four condenser microphones and an ANC active noise reduction device in different areas by arranging a plurality of partition plates 6. The shell bottom sets up the battery and holds the chamber, holds intracavity installation lithium ion battery 5, can dismantle, and is chargeable, and battery 5 supplies power for whole measuring equipment's work operation.
The condenser microphone 1 is used for receiving a vibration signal of wind to the diaphragm and converting the vibration signal into an electric signal.
The ANC active noise reduction device 2 supports floating point operation for a 32-bit DSP, the main frequency operation capacity reaches 200MHz, and the signal-to-noise ratio reaches 103dB. The ANC active noise reduction technology is adopted, which is a technology for generating a signal which is opposite to noise and is generated by environmental noise collected by a secondary channel and playing back the signal by using a main channel microphone device to offset the noise, and an algorithm is used for carrying out superposition offset on the reverse signal which is 180 degrees different from the phase of a noise signal and an original signal. The multi-source acoustic signals are subjected to noise reduction processing, namely various interference noises in the environment are removed while the main direction target acoustic signals are protected.
The processor 3 adopts a Raspberry Pi 4B development board which is a latest release fourth generation product of a Raspberry group, transplants a Python program and related library files to the Raspberry group development board, mainly analyzes acoustic signals subjected to noise reduction processing, analyzes and calculates the acoustic signals of wind, accurately restores wind characteristics, and can transmit the signals to a cloud for remote analysis.
The preferred dimensions of the housing 4 are 150mm long, 70mm wide and 15mm thick. The middle of the side surface of one long edge of the shell is provided with a wind signal collecting port 7, and the middle of the side surface of the other three edges is provided with an environment bottom noise collecting port 8. The shell and the partition plate are both made of ABS engineering plastics, so that the shell and the partition plate are low in cost, excellent in heat resistance and weather resistance and good in impact resistance.
The algorithm for measuring by adopting the portable wind characteristic measuring equipment of the invention is as follows:
Firstly, the equipment is started to enable the equipment to be in a standby working state, the equipment is horizontally placed on a wind measuring platform, a main channel is placed on the windward side, a data acquisition switch is started, and acoustic signals of the windward at different wind speeds are respectively acquired. The main channel of the equipment receives a vibration signal and environmental background noise of wind to a vibrating diaphragm of the acoustic generator, the auxiliary channel mainly collects the environmental background noise, inverse white noise is generated through a DSP chip in the ANC active noise reduction device, and an acoustic signal of the wind is collected in an active noise reduction mode. In this embodiment, each wind speed acquires 20s time series data, and taking the 20s data as an example, when the sampling rate is 48kHz, 20s acquires 20 × 48000 data, so that machine learning is sufficiently ensured to have better generalization capability.
The method comprises the steps of performing signal processing on acoustic signals of wind after noise reduction processing by adopting Python language, enabling each section of acoustic signals to be composed of time sequence data recorded for 20s under each wind speed condition, dividing the sequences into 10 groups, namely taking each 2s of acoustic signals as one group, extracting Mel cepstrum coefficients (MFCCs) from the signals, performing training and testing on the wind speed through a one-dimensional convolutional neural network (1 DCNN), and optimizing the defects of the existing wind measuring technology by adopting a machine learning technology. Fig. 3 shows a flow chart of extracting mel-frequency cepstrum coefficients from an acoustic signal of wind obtained by the device of the present invention.
Where MFCCs is a feature widely used in automatic speech and speaker recognition, fig. 3 shows a specific process for extracting the coefficient, which mainly includes pre-emphasis, framing, and windowing of an input signal, then using Fast Fourier Transform (FFT) to convert the pre-processed signal from time domain to frequency domain, and then using mel filtering, logarithm taking, and discrete cosine transform to obtain mel cepstral coefficients. The mel-frequency cepstrum coefficient is a 39-dimensional matrix, and rows and columns respectively represent mel-frequency features and frame numbers, and the matrix is as follows:
where m =39,n =199.
Referring to the 1DCNN prediction structure diagram of fig. 4, a wind speed prediction model is constructed using a one-dimensional convolutional neural network (1 DCNN) with the mel feature matrix as an input of the one-dimensional convolutional neural network and the actual wind speed vector as an output. After multiple parameter adjustment, the algorithm selects 6 layers of Conv1D to extract features, a layer of Max boosting 1D (maximum pooling) is added behind the 3 rd layer and the 6 th layer of Conv1D respectively to keep main features, the calculated amount is reduced, and each convolution layer uses a linear rectification function (ReLU) as an activation function. And finally, outputting a wind speed predicted value through the two fully-connected layers, wherein the predicted value of the wind speed tends to a real value more and more under the approximation of an MSE loss function (mean square error function). The optimization algorithm for the model employed Adam, the model employed 20 epochs and each batch was 64 in size. In order to evaluate the accuracy of the network model training and testing process, the algorithm self-defines a measurement function:
R＝1-SE/(SR+SE)
wherein SR is regression sum of squares, SE is residual sum of squares, and the closer R is to 1, the more accurate the model prediction is. The network uses two convolution kernels, 110 × 3 and 1 × 3, the convolution kernel moving step is 1, the pooling receptive field size is 1 × 3, and the step is 1.
The activation function used by the model is a ReLU function:
ReLU(x)＝max{0，x}
linear function:
where O is an n-dimensional vector, O i Is the ith element of the vector O.
Referring to fig. 4, firstly, input is mapped through a 110 × 39 one-dimensional convolution kernel, 1 90-dimensional feature vector is obtained through a ReLu activation function, then 16 88-dimensional feature vectors are obtained through 16 one-dimensional convolution kernels with the length of 3 through a ReLu activation function, 16 28-dimensional feature vectors are obtained through 16 one-dimensional convolution kernels with the length of 3 through a maximum pooling, 16 2-dimensional feature vectors are obtained through 3 times of convolution and 1 time of pooling, and finally a predicted value is output through a full connection layer through Linear function activation, so that the purpose of predicting the wind speed is achieved.
The test result shows that the average wind speed can be predicted once every 2s by the equipment, the error is less than 3%, the engineering requirement is met, and the average wind can be accurately restored by adopting the one-dimensional convolution neural network prediction model.
The pulsating wind is caused by the irregularity of the wind, the intensity of the pulsating wind changes randomly along with the time, the intensity of an acoustic signal also changes randomly along with the time, and as the sampling frequency of the device is 48kHz and the sampling frequency of the cobra three-dimensional pulsating wind speed measuring instrument is 2kHz, the acoustic signal collected by the device needs to be processed into data with the same sampling frequency as that of the cobra three-dimensional pulsating wind speed measuring instrument through Python language, namely two groups of same-dimension vectors of 1 multiplied by 20 multiplied by 2000. An LSTM neural network model is built by the 20s acoustic signals acquired by the device and the 20s pulsating wind speed acquired by the cobra three-dimensional pulsating wind speed measuring instrument, and a mapping relation between the acoustic signals and the pulsating wind speed is built, so that the aim of restoring the pulsating wind speed and the wind power spectrum by the acoustic signals is fulfilled.
In the embodiment, a keras deep learning framework is used for quickly building the model, a Sequential model is built, an LSTM layer is added into the model, dropout is set to be 0.15, a Dense layer is added to converge the dimensionality of the model to be 1, a ReLU is used as an activation function of the activation function, a loss function is set to be Mean Square Error (MSE), adam is adopted as an optimization algorithm, 50 epochs are adopted in the model, and the size of each batch is 100.
The network prediction structure is shown in fig. 5, the model includes 4 LSTM and 1 fully-connected layer, wherein the 1 st layer inputs 1 × 40000 time series data, passes through the LSTM layer with 50 neurons in the hidden layer, outputs 50 × 40000 time series data, the second layer inputs 50 × 40000 time series data, passes through the LSTM layer with 100 neurons in the hidden layer, outputs 100 × 40000 time series data, sequentially outputs 300 × 40000 time series data through two layers of LSTM, and finally introduces Dropout, i.e. the number of neurons are inactivated randomly, the number of inactivated neurons accounts for 15% of the total number, finally the fully-connected layer outputs a pulsating wind speed vector after being activated by the ReLU function, and the output size is consistent with the input size, and X (t) is recorded.
The LSTM structure is shown in fig. 6, and the LSTM realizes information protection and control through three basic structures, namely an input gate, a forgetting gate and an output gate. The input gate determines how much new information is added to the cell state, and updates the information through the Sigmoid function and the tanh function, respectively.
Wherein, sigmoid function:
tan h function:
the forgetting gate, which will read the last cell h, decides what information to discard in the cell state t-1 And the cell x t A value between 0 and 1 is output.
The output gate determines the output value, processes it by tanh to get a value between-1-1, and multiplies it with the Sigmoid output.
The pulsation wind speed X (t) is predicted, and the autocorrelation function R (tau) of the random response is obtained, namely:
obtaining an autocorrelation function of the random response, and performing Fourier transform (FFT) on the autocorrelation function, namely:
namely, the pulsating wind power spectrum can be restored by the acoustic signal.
The portable measuring equipment and the embedded artificial intelligence algorithm for restoring the wind characteristic based on the acoustic signal are essentially the combination of artificial intelligence and the traditional technology, a microphone is used for reading a multi-source acoustic signal, and the artificial intelligence algorithm is used for mapping the nonlinear relation between the acoustic signal and the wind characteristic. The algorithms provided by the invention are respectively a one-dimensional convolution neural network (1 DCNN) for reducing the average wind speed and a long-short term memory neural network (LSTM) for reducing the pulsating wind speed and the wind power spectrum. All data in the embodiment are from a second test section of a single-backflow serial double-test-section industrial wind tunnel (XNJD-1) of southwest traffic university, the section of the test section is a rectangle with the width of 2.4m multiplied by 2.0m, the maximum incoming flow wind speed is 45m/s, and the minimum incoming flow wind speed is 0.5m/s; in this embodiment, in order to establish a mapping relationship between wind characteristics and acoustic signals, a Cobra three-dimensional pulsating wind velocity measuring instrument (Cobra Probe) and a pitot tube are disposed beside the apparatus of the present invention to record a pulsating component of wind and an average wind velocity, and the experimental arrangement of the measuring apparatus and the wind velocity measuring instrument of this embodiment is as shown in fig. 7 below.
Through tests, the accuracy rate of reducing the average wind speed by using 1DCNN can reach more than 95%, the accuracy rate of reducing the pulsating wind speed by using LSTM can reach more than 90%, the test effect can accurately reduce the natural wind characteristic, and the robustness is good. Fig. 8 and 9 show the training effect and the testing effect of the 1DCNN reduction of the average wind speed, respectively, and fig. 10 shows the model loss value and the testing accuracy of the 1DCNN after 20 times of iterative training. Fig. 11 and 12 show the test effect of the LSTM on reducing the pulsating wind speed and the wind power spectrum, respectively, and fig. 13 shows the loss value of the model after 50 times of iterative training, it can be seen that the predicted loss decreases rapidly and finally converges to a very small value, the error value is within the acceptable range, and the prediction effect is relatively ideal.
In conclusion, the device for restoring the characteristics of the near-earth wind is simple, does not need an additional sensing device or other connectors, only depends on a capacitance microphone which is arranged on the device, the microphone collects multisource sound signals at a sampling frequency of 48kHz, collects vibration signals of wind on a vibrating diaphragm of the microphone in real time, converts the vibration signals into electrical signals which are easy to recognize by a machine, then carries out noise reduction processing on acoustic signals of a main channel (namely a wind signal collecting port) through an ANC active noise reduction device, and finally carries out real-time analysis on the acoustic signals after the noise reduction processing through a Raspberry Pi 4B development board embedded with a Python program, so that wind characteristic indexes including average wind speed, pulsating wind speed and wind power spectrum are quickly and accurately restored. The invention is obviously different from the prior art, and the prior wind measuring technology adopts a traditional method and achieves the wind measuring purpose by means of an external mechanical sensor. The equipment of the invention utilizes the characteristics of high robustness of the acoustic signal, good sensitivity of the sensor and the like, and uses an artificial intelligence algorithm embedded into the equipment to establish the implicit mapping relation between the acoustic signal and the pulsating wind signal. The wind measuring instrument has the advantages of high precision, good portability, clear algorithm structure and strong operability, and can be suitable for wind measuring situations of different occasions.
Although the present invention has been described with reference to a preferred embodiment, it should be understood that various changes, substitutions and alterations can be made herein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (7)
1. An intelligent algorithm of a portable wind characteristic measuring device is characterized by comprising the following steps:
s1, acquiring acoustic signals of wind
The structure of the portable wind characteristic measuring equipment comprises a cuboid shell, a condenser microphone, a processor and an ANC active noise reduction device, wherein the condenser microphone, the processor and the ANC active noise reduction device are arranged in the shell; one side surface of the shell is a main channel and is provided with a wind signal collecting port, the other three side surfaces of the shell are auxiliary channels and are provided with environment bottom noise collecting ports, and a condenser microphone is respectively arranged in the shell and close to the air inlet and the three noise inlets; an ANC active noise reduction device is arranged at the center position in the shell, and a processor is arranged above the ANC active noise reduction device; the four condenser microphones are connected with the ANC active noise reduction device and the processor in series; the inner space of the shell divides the four condenser microphones and the ANC active noise reduction device into different areas by arranging a plurality of partition plates; the bottom of the shell is provided with a battery accommodating cavity, and the battery supplies power for the working operation of the whole measuring equipment;
starting the portable wind characteristic measuring equipment, horizontally placing the portable wind characteristic measuring equipment on a wind measuring platform, placing a wind signal collecting port on the windward side, and respectively collecting the wind acoustic signals under different wind speeds; the wind signal collecting port receives a vibration signal of wind to a microphone diaphragm, the environment bottom noise collecting port receives environment noise, and an ANC active noise reduction device is used for carrying out noise reduction processing on an acoustic signal of the wind signal collecting port;
s2, reducing average wind
Carrying out signal processing on the acoustic signals subjected to noise reduction processing by adopting Python language, extracting Mel cepstrum coefficients, taking the coefficient matrix as the input of a one-dimensional convolution neural network, taking the actual wind speed as the output, and constructing a wind speed prediction model by using the one-dimensional convolution neural network, wherein the prediction model is used for reducing average wind;
s3, reducing the pulsating wind speed and the wind power spectrum
And processing the acoustic signals after noise reduction into data with the same sampling frequency as that of the cobra three-dimensional pulsating wind speed measuring instrument, building an LSTM neural network model by the acoustic signals and the pulsating wind speed collected by the cobra three-dimensional pulsating wind speed measuring instrument, and establishing a mapping relation between the acoustic signals and the pulsating wind speed so as to achieve the purpose of reducing the pulsating wind speed and the wind power spectrum by the acoustic signals.
2. The intelligent algorithm according to claim 1, wherein the step S2 is specifically as follows:
selecting 6 layers of Conv1D for extracting features, adding a layer of Max scaling 1D behind the 3 rd layer and the 6 th layer of Conv1D respectively to retain main features, reducing calculated amount, and using a linear rectification function ReLU as an activation function for each convolution layer; finally, outputting a wind speed predicted value through the two fully-connected layers, wherein the predicted value of the wind speed tends to a real value more and more under the approximation of an MSE loss function; the optimization algorithm of the model employs Adam, the model employs 20 epochs and each batch has a size of 64; in order to evaluate the accuracy of the network model training and testing process, the algorithm self-defines a measurement function:
R＝1-SE/(SR+SE)
wherein SR is regression sum of squares, SE is residual sum of squares, and the more R approaches to 1, the more accurate model prediction is represented; the network uses two convolution kernels of 110 multiplied by 3 and 1 multiplied by 3, the step length of the convolution kernel is 1, the size of the pooling receptive field is 1 multiplied by 3, and the step length is 1;
firstly, input is mapped by a 110 x 39 one-dimensional convolution kernel, and 1 90-dimensional characteristic vector is obtained after the input passes through a ReLu activation function; then, 16 one-dimensional convolution kernels with the length of 3 are used for obtaining 16 88-dimensional characteristic vectors after the 16 one-dimensional convolution kernels pass through a ReLu activation function; then 16 one-dimensional convolution kernels with the length of 3 pass through a ReLu activation function, and 16 28-dimensional feature vectors are obtained through maximum pooling; and obtaining 16 2-dimensional characteristic vectors after continuously performing convolution for 3 times and pooling for 1 time, and finally activating and outputting a predicted value by a Linear function through a full connection layer to achieve the aim of predicting the wind speed.
3. The intelligent algorithm according to claim 1, wherein in step S3, a keras deep learning framework is used to build the model quickly, a Sequential model is built, an LSTM layer is added thereto, dropout is set to 0.15, a sense layer is added to aggregate its dimensions to 1, a ReLU is used as an activation function, a loss function is set to mean square error MSE, adam is used as an optimization algorithm, 50 epochs are used for the model, and the size of each batch is 100.
4. The intelligent algorithm of claim 1, wherein the ANC active noise reduction device has the performance of 32-bit DSP supporting floating point operation, a dominant frequency operation capability of 200MHz, and a signal-to-noise ratio of 103dB.
5. The intelligent algorithm of claim 1, wherein the processor employs a Raspberry Pi 4B development board, a fourth generation latest release product, to analyze and compute the acoustic signature of the wind and accurately recover the wind characteristics.
6. The intelligent algorithm of claim 1, wherein the shell dimensions are 150mm long, 70mm wide, 15mm thick; the middle of one long side of the shell is provided with a wind signal collecting port, and the middle of the other three sides is provided with an environment bottom noise collecting port.
7. The intelligent algorithm of claim 6, wherein the housing and the diaphragm are both made of ABS engineering plastic.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210305160.1A CN114609410B (en) | 2022-03-25 | 2022-03-25 | Portable wind characteristic measuring equipment based on acoustic signals and intelligent algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210305160.1A CN114609410B (en) | 2022-03-25 | 2022-03-25 | Portable wind characteristic measuring equipment based on acoustic signals and intelligent algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114609410A CN114609410A (en) | 2022-06-10 |
| CN114609410B true CN114609410B (en) | 2022-11-18 |
Family
ID=81866086
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210305160.1A Active CN114609410B (en) | 2022-03-25 | 2022-03-25 | Portable wind characteristic measuring equipment based on acoustic signals and intelligent algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114609410B (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CA2458427A1 (en) * | 2003-02-21 | 2004-08-21 | Harman Becker Automotive Systems - Wavemakers, Inc. | System for suppressing wind noise |
| CN102625201A (en) * | 2012-03-07 | 2012-08-01 | 深圳市福智软件技术有限公司 | Wind noise reducing device for microphone |
| CN103399173A (en) * | 2013-08-08 | 2013-11-20 | 中国科学院上海微系统与信息技术研究所 | Wind speed and wind direction evaluating system and method |
| CN106645792A (en) * | 2016-10-18 | 2017-05-10 | 南京信息工程大学 | Supersonic wave wind-speed and wind-direction measuring apparatus and measuring method |
| CN107292446A (en) * | 2017-07-03 | 2017-10-24 | 西南交通大学 | A kind of mixing wind speed forecasting method based on consideration component relevance wavelet decomposition |
| CN110987066A (en) * | 2019-11-26 | 2020-04-10 | 青岛科技大学 | Ocean wind speed and direction measuring method and system capable of achieving automatic correction |
| CN111781396A (en) * | 2020-06-23 | 2020-10-16 | 晋中学院 | Wind speed measuring method |
| EP3734296A1 (en) * | 2019-05-03 | 2020-11-04 | FRAUNHOFER-GESELLSCHAFT zur Förderung der angewandten Forschung e.V. | A method and an apparatus for characterizing an airflow |
| DE102020114146A1 (en) * | 2019-06-13 | 2020-12-17 | Apple Inc. | SPEAKER IMAGE OF A MICROPHONE FOR WIND DETECTION |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7343793B2 (en) * | 2006-05-03 | 2008-03-18 | The Boeing Company | Acoustic profiler for wind, temperature, and turbulence |
| US20150377667A1 (en) * | 2014-06-30 | 2015-12-31 | Saudi Arabian Oil Company | Virtual multiphase flow metering and sand detection |
| GB2547284B (en) * | 2016-02-15 | 2019-11-06 | Ft Tech Uk Ltd | Acoustic resonator sensor for determining temperature |
| US10504537B2 (en) * | 2018-02-02 | 2019-12-10 | Cirrus Logic, Inc. | Wind noise measurement |
| CN112326210A (en) * | 2019-07-17 | 2021-02-05 | 华北电力大学（保定） | Large motor fault diagnosis method combining sound vibration signals with 1D-CNN |
| WO2021034300A1 (en) * | 2019-08-16 | 2021-02-25 | Bp Exploration Operating Company Limited | Das data processing to characterize fluid flow |
| US11462204B2 (en) * | 2020-09-08 | 2022-10-04 | Siemens Gamesa Renewable Energy A/S | Wind turbine and method for noise reduction for a wind turbine |
-
2022
- 2022-03-25 CN CN202210305160.1A patent/CN114609410B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CA2458427A1 (en) * | 2003-02-21 | 2004-08-21 | Harman Becker Automotive Systems - Wavemakers, Inc. | System for suppressing wind noise |
| CN102625201A (en) * | 2012-03-07 | 2012-08-01 | 深圳市福智软件技术有限公司 | Wind noise reducing device for microphone |
| CN103399173A (en) * | 2013-08-08 | 2013-11-20 | 中国科学院上海微系统与信息技术研究所 | Wind speed and wind direction evaluating system and method |
| CN106645792A (en) * | 2016-10-18 | 2017-05-10 | 南京信息工程大学 | Supersonic wave wind-speed and wind-direction measuring apparatus and measuring method |
| CN107292446A (en) * | 2017-07-03 | 2017-10-24 | 西南交通大学 | A kind of mixing wind speed forecasting method based on consideration component relevance wavelet decomposition |
| EP3734296A1 (en) * | 2019-05-03 | 2020-11-04 | FRAUNHOFER-GESELLSCHAFT zur Förderung der angewandten Forschung e.V. | A method and an apparatus for characterizing an airflow |
| DE102020114146A1 (en) * | 2019-06-13 | 2020-12-17 | Apple Inc. | SPEAKER IMAGE OF A MICROPHONE FOR WIND DETECTION |
| CN110987066A (en) * | 2019-11-26 | 2020-04-10 | 青岛科技大学 | Ocean wind speed and direction measuring method and system capable of achieving automatic correction |
| CN111781396A (en) * | 2020-06-23 | 2020-10-16 | 晋中学院 | Wind speed measuring method |
Non-Patent Citations (2)
| Title |
|---|
| Artificial neural network based wake model for power prediction of wind farm;遆子龙 等;《Renewale Energy》;20211231;618-631 * |
| 基于变量选择深度信念神经网络的风速预测;李大中 等;《华北电力大学学报(自然科学版)》;20211231;第48卷(第1期);62-68 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114609410A (en) | 2022-06-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107644650B (en) | Improved sound source positioning method based on progressive serial orthogonalization blind source separation algorithm and implementation system thereof | |
| Zhang et al. | Fault identification based on PD ultrasonic signal using RNN, DNN and CNN | |
| Yangzhou et al. | A deep neural network approach to acoustic source localization in a shallow water tank experiment | |
| CN113763986B (en) | Abnormal sound detection method for air conditioner indoor unit based on sound classification model | |
| CN108802687A (en) | The more sound localization methods of distributed microphone array in reverberation room | |
| Foy et al. | Mean absorption estimation from room impulse responses using virtually supervised learning | |
| CN112052712B (en) | Power equipment state monitoring and fault identification method and system | |
| Adalbjörnsson et al. | Sparse localization of harmonic audio sources | |
| CN110515034B (en) | Acoustic signal azimuth angle measurement system and method | |
| CN114609410B (en) | Portable wind characteristic measuring equipment based on acoustic signals and intelligent algorithm | |
| CN110049423A (en) | A kind of method and system using broad sense cross-correlation and energy spectrum detection microphone | |
| CN113705418A (en) | Infrasound signal identification method, system and equipment based on MFCC and HMM | |
| CN117275510A (en) | Small sample underwater sound target identification method and system based on multi-gradient flow network | |
| CN117116293A (en) | Machine equipment fault diagnosis system in complex sound field environment | |
| CN112992182B (en) | Vehicle wind noise level testing system and testing method thereof | |
| CN111968671B (en) | Low-altitude sound target comprehensive identification method and device based on multidimensional feature space | |
| Zea et al. | Learning the finite size effect for in-situ absorption measurement | |
| Mirabilii et al. | Data-driven wind speed estimation using multiple microphones | |
| Gao et al. | Room geometry blind inference based on the localization of real sound source and first order reflections | |
| Yu et al. | Research and Application of Intelligent Diagnosis Method of Mechanical Fault Based on Transformer Vibration and Noise and BP Neural Network | |
| Liu et al. | Gas Leak Fault Detection based on Improved MobileNetV3 | |
| Zhang et al. | Sound event localization and classification using WASN in Outdoor Environment | |
| Yang et al. | Research on Voiceprint recognition method of buried drainage pipe based on MFCC and GMM-HMM | |
| Cao et al. | Acoustics recognition of excavation equipment based on MF-PLPCC features and RELM | |
| CN117268796B (en) | Vehicle fault acoustic event detection method |
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 |