An array-type piezoelectric rain sensor

Abstract

This invention relates to the field of piezoelectric sensors, specifically to an array-type piezoelectric rain sensor, comprising: a signal acquisition module, a rubber buffer pad, a circuit board, and a mounting base; the signal acquisition module is mounted on the mounting base; the rubber buffer pad is positioned between the signal acquisition module and the mounting base; the circuit board is disposed within the mounting base and is electrically connected to the signal acquisition module; the signal acquisition module includes a housing, a piezoelectric transducer array, and a temperature measurement array; both the piezoelectric transducer array and the temperature measurement array are tightly connected to the inner surface of the housing; the circuit board includes a signal conditioning module and a data processing module, wherein the signal conditioning module is used to condition the piezoelectric signal, and the data processing module is used to optimize the conditioned piezoelectric signal and the ambient temperature signal to obtain the rainfall measurement result; this invention uses a data compensation module to compensate the calculated data based on the temperature of the measurement environment, thereby improving the accuracy of the measurement.

Description

Technical Field

[0001] This invention relates to the field of piezoelectric sensor technology, and more specifically to an array-type piezoelectric rain sensor. Background Technology

[0002] A piezoelectric rain sensor is a sensor that uses the piezoelectric effect to measure rainfall. It is widely used in smart cities, meteorological detection, and industrial control. The principle of piezoelectric rain measurement is to use the piezoelectric effect to convert the mechanical vibration caused by raindrop impact into a voltage signal output. Based on the magnitude and changes of the output voltage waveform, the weight of a single raindrop can be calculated using the impulse measurement principle, and thus the continuous rainfall can be obtained.

[0003] Currently, most piezoelectric rain gauges use a single piezoelectric transducer structure, where the transducer is located at the center of an arc-shaped stainless steel sheet, forming a rain-sensing surface to measure the impulse of rainfall. This structure suffers from significant edge effects because the amplitude of the same raindrop falling on different areas of the rain-sensing surface varies. Since the sensor itself cannot distinguish whether the measured impulse is due to differences in raindrop weight or impact area, it leads to inaccurate measurement of the true impulse of the rain, resulting in measurement errors. Furthermore, rain gauges with a single piezoelectric transducer structure also suffer from problems such as inability to distinguish overlapping rain and poor interference resistance.

[0004] For example, patent application number 202022374691.5, entitled "Piezoelectric Rain Sensor and Piezoelectric Rain Detection System," discloses a piezoelectric rain sensor, comprising: a housing with a first mounting opening; a detection cover covering the first mounting opening; a diaphragm opposite to the detection cover, with the periphery of the diaphragm connected to the inner wall of the housing to form a sound cavity spaced apart from the detection cover; and a piezoelectric element stacked on the side of the diaphragm facing away from the detection cover, improving the measurement accuracy of light rainfall. However, because the rainfall measurement results are affected by environmental factors, the accuracy of the measurement results of this device is low. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention proposes an array-type piezoelectric rain sensor. The device includes a signal acquisition module, a rubber buffer pad, a circuit board, and a mounting base. The signal acquisition module is mounted on the mounting base. The rubber buffer pad is positioned between the signal acquisition module and the mounting base to reduce the impact of environmental vibration on the piezoelectric array. The circuit board is housed within the mounting base and is electrically connected to the signal acquisition module.

[0006] The signal acquisition module includes a housing, a piezoelectric transducer array, and a temperature measurement array; both the piezoelectric transducer array and the temperature measurement array are tightly connected to the inner side of the housing, wherein the piezoelectric transducer array is used to convert vibration signals into piezoelectric signals, and the temperature measurement array is used to measure ambient temperature;

[0007] The circuit board includes a signal conditioning module and a data processing module. The signal conditioning module is used to condition the piezoelectric signal, and the data processing module is used to optimize the conditioned piezoelectric signal and the ambient temperature signal to obtain the rainfall measurement result.

[0008] Preferably, the outer surface of the housing is provided with a superhydrophobic layer to prevent rainwater and other pollutants from accumulating on the housing and thus affecting its vibration characteristics.

[0009] Preferably, the piezoelectric transducer array includes one main piezoelectric transducer and 3n slave piezoelectric transducers, wherein the main piezoelectric transducer is located at the center of the inner side of the arc-shaped stainless steel shell, and the three slave piezoelectric transducers form a slave piezoelectric transducer group, and all slave piezoelectric transducer groups are evenly distributed around the main piezoelectric transducer.

[0010] Furthermore, each piezoelectric transducer in the piezoelectric transducer group is equidistant from the other piezoelectric transducer.

[0011] Preferably, the signal conditioning module includes n signal sampling channels, each of which consists of an amplifier circuit, a filter circuit, and a notch filter circuit; the amplifier circuit is used to amplify the piezoelectric signal, the filter circuit is used to filter the amplified signal, and the notch filter circuit is used to eliminate the power frequency interference signal of the filtered voltage.

[0012] Preferably, the data processing module includes a data acquisition circuit, a storage module, a communication circuit, and a data compensation module. The data acquisition circuit consists of a CPLD, used to acquire the time and number of shock waves caused by raindrops reaching the piezoelectric transducer, calculate the position of the raindrops using a triangulation algorithm, and calculate the actual impulse of the raindrops by combining the real-time temperature from the sensor. The storage module is used to store the acquired information and compensation information. The communication circuit is used to communicate with the host computer. The data compensation module compensates for the actual impulse of the raindrops based on the acquired temperature.

[0013] Furthermore, the process of calculating the raindrop position using the triangulation algorithm includes: fixing one of the piezoelectric transducers in the piezoelectric transducer group and constructing a coordinate system with that piezoelectric transducer as the origin; obtaining the time t1, t2, and t3 when the raindrop falls onto the first pulse output by the three piezoelectric transducers and the propagation speed v of the shock wave in the curved stainless steel; measuring the distance m between two piezoelectric transducers; and calculating the position coordinates (x, y) of the raindrop falling on the sensing surface based on the distance between the raindrop and the piezoelectric transducer.

[0014] The beneficial effects of this invention are:

[0015] This invention improves detection accuracy by setting up a master piezoelectric transducer and a slave piezoelectric transducer, so that the amplitude caused by raindrops falling on the rain-sensing surface can be sensed by both the master and slave piezoelectric transducers. This invention also improves measurement accuracy by using a data compensation module to compensate the calculated data according to the temperature of the measurement environment. Attached Figure Description

[0016] Figure 1 This is a structural diagram of an array-type piezoelectric rain sensor according to an embodiment of the present invention;

[0017] Figure 2 This is a schematic diagram of the signal acquisition module structure according to an embodiment of the present invention;

[0018] Figure 3 This is a schematic diagram of the signal conditioning module circuit structure according to an embodiment of the present invention;

[0019] Figure 4 This is a schematic diagram of a piezoelectric array according to an embodiment of the present invention;

[0020] Figure 5 This is a coordinate diagram of the raindrop landing points according to an embodiment of the present invention;

[0021] Figure 6 This is a schematic diagram of the wavelet neural network structure according to an embodiment of the present invention;

[0022] Figure 7 This is a flowchart illustrating the compensation for the actual impulse of raindrops according to an embodiment of the present invention;

[0023] The components include: 1. Signal acquisition module; 11. Temperature sensor; 12. Slave piezoelectric transducer; 13. Main piezoelectric transducer; 14. Transducer leads; 15. Housing; 2. Rubber buffer pad; 3. Circuit board; 4. Mounting base. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] An array-type piezoelectric rain sensor, such as Figure 1As shown, the device includes: a signal acquisition module 1, a rubber buffer pad 2, a circuit board 3, and a mounting base 4; the signal acquisition module 1 is mounted on the mounting base; the rubber buffer pad 2 is positioned between the signal acquisition module 1 and the mounting base 4 to reduce the impact of environmental vibration on the piezoelectric array; the circuit board 3 is mounted inside the mounting base and is electrically connected to the signal acquisition module 1.

[0026] In this embodiment, the overall structure of the array-type piezoelectric rain sensor is circular, with an arc at the edge of the circle, which makes the oscillation signal generated when the raindrops fall more accurate.

[0027] A specific implementation of a signal acquisition module, such as Figure 2 As shown, the module includes a housing 15, a piezoelectric transducer array, and a temperature measurement array; both the piezoelectric transducer array and the temperature measurement array are tightly connected to the inner side of the housing, wherein the piezoelectric transducer array is used to convert vibration signals into piezoelectric signals, and the temperature measurement array is used to measure ambient temperature.

[0028] In this embodiment, the outer casing 15 is an arc-shaped stainless steel casing, which is located at the top of the sensor and is in direct contact with rainwater.

[0029] Preferably, the outer surface of the curved stainless steel shell is provided with a superhydrophobic layer to prevent rainwater and other pollutants from adhering to the shell surface and thus affecting its vibration characteristics.

[0030] In this embodiment, the piezoelectric transducer array consists of (3n+1) piezoelectric transducers, including one master piezoelectric transducer and 3n slave piezoelectric transducers, where n is a natural number ≥ 1. The piezoelectric transducer array is connected to the inner surface of the curved stainless steel shell using adhesive. The master piezoelectric transducer is located at the exact center of the inner surface of the curved stainless steel shell, and the slave piezoelectric transducers are grouped in sets of three, forming an equilateral triangle and evenly distributed around the master piezoelectric transducer.

[0031] The temperature measurement array consists of three temperature sensors, which are connected to the inner side of the curved stainless steel housing by adhesive to form an equilateral triangle.

[0032] In this embodiment, as Figure 3 As shown, the signal conditioning module has (3n+1) signal sampling channels, each mainly consisting of an amplifier circuit, a filter circuit, and a notch filter circuit. Specifically, the signal conditioning module includes n signal sampling channels, each consisting of an amplifier circuit, a filter circuit, and a notch filter circuit; the amplifier circuit is used to amplify the piezoelectric signal, the filter circuit is used to filter the amplified signal, and the notch filter circuit is used to eliminate power frequency interference signals in the filtered voltage.

[0033] Preferably, the amplifier circuit is designed with a low-noise, low-temperature-drift instrumentation amplifier powered by dual power supplies, and a feedforward amplifier circuit is introduced to amplify the common-mode signal at the input end, thereby improving the common-mode rejection ratio of the signal.

[0034] Preferably, the filter circuit is designed with a fourth-order Butterworth bandpass filter circuit based on the maximum resonant frequency of the piezoelectric transducer, which is used to filter out the power frequency and high frequency signals amplified by the preamplifier circuit and improve the signal-to-noise ratio.

[0035] Preferably, the notch filter circuit is a second-order Butterworth 50Hz notch filter with an attenuation of 40dB, used to filter out power frequency interference signals coupled to the back end through the differential signal, thereby further improving the signal-to-noise ratio.

[0036] In this embodiment, the data processing module mainly consists of a data acquisition circuit, a storage circuit, a communication circuit, and a data compensation module.

[0037] Preferably, the data acquisition circuit mainly consists of a CPLD, which collects the time and number of shock waves caused by raindrops reaching the piezoelectric transducer, calculates the position of the raindrops using a triangulation algorithm, and calculates the actual impulse of the raindrops by combining the real-time temperature of the sensor, thereby obtaining the rainfall and rain intensity.

[0038] Preferably, the storage circuit is mainly used to store the acquired information and the compensation information.

[0039] Preferably, the communication circuit mainly includes a 485 communication module for communicating with the host computer.

[0040] Preferably, the data processing algorithms mainly include positioning algorithms and compensation algorithms.

[0041] A key component of this invention is a rainfall measurement algorithm integrated with hardware, which mainly consists of a positioning algorithm and a compensation algorithm.

[0042] In this embodiment, the positioning algorithm includes: when the number of electronic array sensors is (3+1), that is, when n=1, there is 1 master piezoelectric sensor and 3 slave piezoelectric transducers, as shown in the schematic diagram below. Figure 4 As shown. A coordinate system is constructed with transducer 1 as the origin. The coordinates of any raindrop falling on the sensing surface are (x, y), and the straight-line distances from these coordinates to the three sensors are d1, d2, and d3. The following coordinate system can then be obtained, see... Figure 5 .

[0043] If three sensors form an equilateral triangle with a fixed length of m, then the coordinates of the three sensors are (0,0), (m,0), and (m,0). Then we can obtain the following equation:

[0044]

[0045]

[0046]

[0047] Let t1, t2, and t3 be the times from when the raindrop falls to when the three sensors output the first pulse, and let v be the propagation speed of the shock wave caused by the raindrop. Then the formula for the distance difference between each piezoelectric transducer and the origin is:

[0048] |d1-d2|=|t1-t2|v

[0049] |d1-d3|=|t1-t3|v

[0050] |d2-d3|=|t2-t3|v

[0051] Based on the propagation velocity v of the shock wave in the curved stainless steel and the pulse time difference Δt sampled by the CPLD, the values ​​of d1, d2, and d3 can be obtained by solving the distance formulas for the three sensors. Substituting the calculated distances into the distance difference formula, the position (x, y) of the raindrop in the coordinate system is obtained. This method can determine the planar position (x, y) of any raindrop on the sensing surface in the coordinate system. n y n When n > 1, averaging and redundancy algorithms can be used to improve the accuracy and anti-interference ability of measurement data.

[0052] In this embodiment, the compensation algorithm is as follows: By conducting impact tests on the sensor with raindrops at the same height but different temperatures, the piezoelectric intensity distribution matrix N of the raindrops falling on the sensing surface can be obtained. A neural network algorithm is then used to fit the piezoelectric intensity distribution matrix N data, with an error target of less than 10. -3 Training is performed on the target, and the final rainfall measurement value z = f(x) is obtained. n y n , t n ), where x n y n Let t be the relative coordinates of the raindrop on the sensing surface. n This refers to the sensor temperature.

[0053] In this embodiment, as Figure 7 As shown, the data compensation module compensates for the actual impulse of raindrops by including:

[0054] Step 1: Obtain sample data; this sample data consists of data generated by water droplets of the same mass falling on the rain-sensing surface of the rain sensor under the same conditions and the rain-sensing surface of the calibrated rain sensor.

[0055] The process of acquiring sample data includes: dropping a water droplet of mass m from the same height onto the rain-sensing surface, recording the peak voltage output of the main voltage transducer, and assuming that the temperature sensor outputs temperature t at this time, and the peak voltage generated by the water droplet is u. p The water droplet falls at coordinates (x, y). Under the same conditions, a water droplet is placed at the exact center of the sensing surface of a calibrated standard rain gauge. Let the peak voltage generated at this moment be u. r A dataset T was constructed to record the peak impact voltage and center impact voltage caused by water droplets of different temperatures and masses, as shown in the table below:

[0056] Table 1

[0057]

[0058] Where x n Let x be the x-coordinate of the raindrop on the coordinate system, and y be the y-coordinate. n Let t be the Y-coordinate of the raindrop in the coordinate system. n u represents the temperature of the sensor when a raindrop falls. pn This refers to the peak voltage generated when a raindrop falls.

[0059] Step 2: Normalize the sample data and use the K-fold cross-validation principle to divide the normalized data into 10 training subsets and corresponding 10 test subsets;

[0060] Step 3: Construct a wavelet neural network; set the first mean square error threshold and the second mean square error threshold;

[0061] In this embodiment, the structure of the wavelet neural network is as follows: Figure 6 As shown, the structure includes an input layer, a hidden layer, and an output layer, with 4 nodes in the input layer, L nodes in the hidden layer, and 1 node in the output layer, thus constructing a 3-layer wavelet neural network.

[0062] The first mean square error threshold is set to 0.0001, and the second mean square error threshold is set to 0.0001.

[0063] The formula for calculating hidden layer nodes is: a∈[1,10]. The specific value of a is adjusted according to the training results.

[0064] Let G be the input sample of the wavelet neural network, g n For the network input of the nth sample, x n1 x n2 x n3 x n4 x in dataset T n y n t n upn p n For the network output of the nth sample, r n The expected output for the nth sample is:

[0065]

[0066] The wavelet neural network model can be represented as:

[0067]

[0068] Where ψ(x) is the Morlet wavelet basis function, ω ij Let ω be the weight from the i-th node in the input layer to the j-th node in the hidden layer. j Let a be the weight from the j-th node in the hidden layer to the output layer. j b is the scaling factor of the wavelet function. j x is the shift factor of the wavelet function. in p is the nth parameter of the point in the i-th node of the input layer. n This is the output of the neural network.

[0069] Step 4: Randomly initialize the position and velocity vectors of N particles, and use the scaling factor 'a' of the wavelet neural network. j Translation factor b j Connection weight ω ij and ω j The vector formed represents the position p of the particle. z , using s b The vector represents the particle's velocity v. z ,Right now

[0070]

[0071] Where, p z and v zd Let s represent the position vector and velocity vector of the z-th particle, respectively. b This represents the b-th value in the velocity vector.

[0072] Step 5: Optimize the wavelet neural network parameters using the particle swarm optimization algorithm;

[0073] Step 51: Set the maximum iteration threshold;

[0074] Step 52: Calculate the fitness function value based on the initialization parameters. Its expression is:

[0075]

[0076] Where fitness is the fitness function, p n For the network output of the nth sample, rn Let M be the expected output of the nth sample, and M be the number of samples in the subset.

[0077] Step 53: Update the particle's velocity and position based on the fitness function value. The update formula is as follows:

[0078]

[0079] Where ω is the inertia weight, c1 and c2 are learning factors, and r1 and r2 are random numbers between [0, 1]. Let be the velocities of particle i during the k-th and k+1-th selection processes, respectively. Let p be the position of particle i during the k-th and k+1-th iterations, respectively. best For the optimal position of an individual, g best This is the optimal position for the group.

[0080] Step 54: Calculate the global optimum based on the updated velocity and position, and recalculate the fitness function value;

[0081] Step 55: Determine whether the current iteration count has reached the maximum iteration count or the fitness value has increased to 95%. If so, output the scaling factor, translation factor, and connection weights as the initial values ​​of the wavelet network parameters. Otherwise, return to step 53.

[0082] Step 6: Input the 10 training subsets into the optimized wavelet neural network to obtain the output results;

[0083] Step 7: Calculate the mean square error of the wavelet neural network based on the output results; its expression is:

[0084]

[0085] Where, p n For the network output of the nth sample, r n Let M be the expected output of the nth sample, and M be the number of samples in the subset.

[0086] Step 8: Compare the calculated mean squared error with the set first mean squared error threshold. If it is greater than the set first mean squared error threshold, proceed to step 9. Otherwise, output the trained scaling factor, translation factor, and connection weights.

[0087] Step 9: Adjust the connection layer weights and wavelet basis coefficients using gradient descent, and return to step 6;

[0088] The formula for adjusting the connection layer weights and wavelet basis coefficients using the gradient descent method is as follows:

[0089]

[0090] in, Let be the weights from the i-th node in the input layer to the j-th node in the hidden layer at iteration k+1. Let J be the weight from the j-th node of the hidden layer to the output layer at iteration k+1. The scaling factor of the wavelet function at iteration k+1 is given. E is the shift factor of the wavelet function at iteration k+1, η1 is the weight learning rate, and E n Let λ be the network prediction error, λ be the momentum factor, and η2 be the scale factor and learning rate.

[0091] Step 10: Validate the trained wavelet neural network using the test set corresponding to the training set to obtain the test mean squared error; compare the test mean squared error with the second mean squared error threshold. If it is greater than the second mean squared error threshold, return to step 4; otherwise, calculate the average mean squared error of all mean squared errors in the test subset.

[0092] Substitute test subsets 1 to 10 into the sub-models trained from the training subsets to calculate the mean squared error, and then calculate the average of all mean squared errors, i.e., calculate the generalization error E of the current sub-model. Gs The details are as follows:

[0093]

[0094] Among them, E Gs Let d represent the generalization error of the s-th sub-model, and d represent the test set number.

[0095] Step 11: Select the parameters and structure of the sub-model corresponding to the minimum generalization error as the final network structure and network parameters;

[0096] Step 12: Compensate for the actual impulse of raindrops using the final network structure and network parameters.

[0097] In this embodiment, the mounting base is provided with a power supply interface and a communication interface for supplying power and communicating with the device.

[0098] The above-described embodiments further illustrate the purpose, technical solution, and advantages of the present invention. It should be understood that the above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made to the present invention within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An arrayed piezoelectric rain sensor, characterized by, The utility model relates to a raindrop sensor, which comprises a signal acquisition module, a rubber buffer pad, a circuit board and a mounting base. The signal acquisition module is arranged on the mounting base. The rubber buffer pad is arranged between the signal acquisition module and the mounting base to reduce the influence of environmental vibration on the piezoelectric array. The signal acquisition module comprises a shell, a piezoelectric transducer array and a temperature measurement array. The piezoelectric transducer array and the temperature measurement array are closely connected to the inner side of the shell. The piezoelectric transducer array is used to convert vibration signals into piezoelectric signals, and the temperature measurement array is used to measure the ambient temperature. The piezoelectric transducer array comprises one main piezoelectric transducer and three n slave piezoelectric transducers.

2. The array type piezoelectric rain sensor according to claim 1, wherein The main piezoelectric transducer is arranged at the center of the top plane of the arc-shaped stainless steel shell.

3. The array type piezoelectric rain sensor according to claim 1, wherein The three slave piezoelectric transducers form a slave piezoelectric transducer group.

4. The array type piezoelectric rain sensor according to claim 1, wherein All the slave piezoelectric transducer groups are evenly distributed around the main piezoelectric transducer.

5. The array type piezoelectric rain sensor according to claim 1, wherein The circuit board comprises a signal conditioning module and a data processing module.

6. The array type piezoelectric rain sensor according to claim 1, wherein The signal conditioning module is used to condition the piezoelectric signals, and the data processing module is used to optimize the conditioned piezoelectric signals and the ambient temperature signals to obtain the rain measurement results. The data processing module comprises a data acquisition circuit, a storage module, a communication circuit and a data compensation module. The data acquisition circuit is composed of a CPLD and is used to acquire the time when the shock wave caused by the raindrop reaches the piezoelectric transducer and the number of shock waves. The real impulse of the raindrop is calculated by a triangular positioning algorithm and the real-time temperature of the sensor. The storage module is used to store the collected information and compensation information. The communication circuit is used to communicate with the upper computer. The real impulse of the raindrop is compensated by the data compensation module according to the collected temperature. The shell is an arc-shaped stainless steel shell. The distance between each slave piezoelectric transducer in the slave piezoelectric transducer group and another slave piezoelectric transducer is equal. The signal conditioning module comprises n signal sampling channels. Each signal sampling channel is composed of an amplification circuit, a filter circuit and a notch circuit. The amplification circuit is used to amplify the piezoelectric signals. The filter circuit is used to filter the amplified signals. The notch circuit is used to eliminate the power frequency interference signals of the filtered voltage. The process of calculating the raindrop position by the triangular positioning algorithm comprises the following steps. A slave piezoelectric transducer in the slave piezoelectric transducer group is fixed, and a coordinate system is constructed with the piezoelectric transducer as the coordinate origin. The first pulse time t1, t2 and t3 output by the three slave piezoelectric transducers when the raindrop falls and the propagation speed v of the shock wave in the arc-shaped stainless steel are obtained. The distance m between two slave piezoelectric transducers is measured. The position coordinates (x, y) of the raindrop falling on the sensing surface are calculated according to the distance between the raindrop and the slave piezoelectric transducer. The data compensation module compensates the real impulse of the raindrop, which comprises the following steps. Step 1: Obtain sample data. The sample data is the data generated when water droplets of the same mass fall on the rain sensing surface of the rain sensor and the calibrated rain sensor under the same conditions. Step 2: Normalize the sample data, and divide the normalized data into 10 training subsets and corresponding 10 test subsets by using K-fold cross-validation principle; Step 3: Construct a wavelet neural network; set a first mean square error threshold and a second mean square error threshold; Step 4: Randomly initialize the position and velocity vectors of N particles, and use the scaling factor 'a' of the wavelet neural network. j Translation factor b j Connection weights and The vector formed represents the position of the particle. ,use The constituent vector represents the velocity of the particle. ; Step 5: Optimize the wavelet neural network parameters by using a particle swarm optimization algorithm; Step 6: Input the 10 training subsets into the optimized wavelet neural network respectively to obtain output results; Step 7: Calculate the mean square error of the wavelet neural network according to the output results; Step 8: Compare the calculated mean square error with the set first mean square error threshold; if greater than the set first mean square error threshold, execute Step 9, otherwise, output the trained scaling factor, translation factor and connection weight; Step 9: Adjust the connection layer weight and wavelet base coefficient by using gradient descent method, and return to Step 6; Step 10: Verify the trained wavelet neural network by using the test set corresponding to the training set to obtain a test mean square error; compare the test mean square error with the second mean square error threshold; if greater than the second mean square error threshold, return to Step 4, otherwise, calculate the average value of all mean square errors in the test subsets; Step 11: Select the parameters and structure of the sub-model corresponding to the minimum generalization error as the final network structure and network parameters; Step 12: Compensate the real impulse of raindrops by using the final network structure and network parameters.

7. The arrayed piezoelectric rain sensor according to claim 6, wherein, The expression of the wavelet neural network model is: ； wherein, is a wavelet basis function, is a weight from the i-th node of the input layer to the j-th node of the hidden layer, is a weight from the j-th node of the hidden layer to the output layer, is a scaling factor of the wavelet function, is a translation factor of the wavelet function, is the n-th parameter of the i-th node of the input layer, is the output of the neural network.

8. The array type piezoelectric rain sensor according to claim 6, wherein The formula for adjusting the connection layer weight and wavelet base coefficient by using gradient descent method is: ； wherein, is the weight of the input layer i-th node to the hidden layer j-th node at the k+1 iteration, is the weight of the hidden layer j-th node to the output layer at the k+1 iteration, is the scaling factor of the wavelet function at the k+1 iteration, is the translation factor of the wavelet function at the k+1 iteration, is the weight learning rate, is the network prediction error, is the momentum factor, is the scale factor learning rate.

Images