A method for calculating the arrival time of sound in acoustic tomography based on ensemble learning

Abstract

The present invention provides a method for calculating the arrival time of sound based on ensemble learning, including using a plurality of algorithms to respectively calculate the arrival time of sound and obtain a plurality of calculation results; comparing the calculation results with the true results and calculating the errors of the results of each algorithm; calculating the reliability of each algorithm result according to the errors of the results of each algorithm, and then fusing them; using the new data to obtain the various sound arrival time results through algorithm calculation, and then fusing them into the final result through ensemble learning, so as to improve the accuracy and stability of the algorithm. The method for calculating the arrival time of sound based on ensemble learning provided by the present invention integrates a plurality of basic calculation methods, so that its results are more stable than each of the basic algorithms, and the algorithm can also integrate more basic algorithms in the future to further improve the overall accuracy, and further improve the accuracy of measuring the flow rate of rivers.

Description

Technical Field

[0001] The invention relates to the field of river acoustic tomography flow detection, and in particular to an acoustic tomography sound arrival time calculation method based on ensemble learning. Background Art

[0002] River acoustic tomography flow measurement technology is a measurement technology that uses sound velocity difference to measure river flow velocity. It has the characteristics of easy installation, little influence of river surface conditions, and convenient measurement. At present, river acoustic tomography flow measurement technology is still in the development stage. Sound signals are affected by factors such as basin topography, temperature, and the equipment itself, and there will be signal weakening and signal loss, which will cause great interference to the measurement of sound arrival time, and ultimately affect the calculation of flow velocity and flow results. Various existing single measurement algorithms have large errors in fluid measurement sound signals.

[0003] Therefore, there is an urgent need in this field for a technical solution that can improve the measurement results of river flow and flow velocity. Summary of the invention

[0004] In order to improve the accuracy of measuring the sound arrival time in the process of acoustic tomography flow measurement, the present invention proposes an acoustic tomography sound arrival time calculation method based on ensemble learning, which uses ensemble learning combined with multiple weak learners to obtain a result with minimal error and higher accuracy.

[0005] The method of the present invention comprises:

[0006] Step 1: Use multiple algorithms to calculate the sound arrival time and obtain several calculation results;

[0007] Step 2: Compare the calculated results with the true results to calculate the errors of the results of each algorithm;

[0008] Step 3: Calculate the reliability of each algorithm result based on the error of each algorithm result and then merge them;

[0009] Step 4: The new data is calculated through the algorithm to obtain the arrival time results of each sound, and then integrated into the final result through integrated learning to improve the accuracy and stability of the algorithm.

[0010] Optionally, the algorithm in step one includes an edge detection algorithm, a peak detection algorithm and a prediction model algorithm.

[0011] Optionally, the error of the algorithm results is calculated as follows: the mean square error is used to describe the result error. Assuming that the sound arrival time calculated by an algorithm at the i-th moment is The actual arrival time is t i , a total of n groups of data are calculated, then the mean square error of the algorithm is:

[0012] Optionally, the step three specifically includes:

[0013] The step three specifically includes: using a weighted average algorithm, assuming that there are n algorithm results in total, and the error of each result is e i , then the weight corresponding to the i-th result is:

[0014] The calculation result t of the weighted average algorithm is:

[0015] Optionally, step three further includes: using a Kalman filter algorithm, assuming that the sound arrival time result obtained by the prediction model algorithm is t pred , error Q = e pred , the result obtained by the edge detection algorithm is t obs1 , the error is e obs1 , the result obtained by the peak detection algorithm is t obs2 , the error is e obs2 , the initial fusion error is P, then the five calculation formulas of Kalman filtering are:

[0016]

[0017]

[0018]

[0019]

[0020]

[0021] The optimal sound arrival time t and the updated fusion error P are obtained, and the updated fusion error is used in the next Kalman filter algorithm calculation.

[0022] Compared with the prior art, the present invention has the following advantages:

[0023] The sound arrival time calculation method based on ensemble learning provided by the present invention integrates a variety of basic algorithms, so that its results are more stable than any of the basic algorithms, and the algorithm of the present invention can also integrate more basic algorithms to further improve the overall accuracy, and further improve the accuracy of measuring river flow rate. Facing the complex river environment of different regions, seasonal temperatures and water flow sizes, it is more advanced and scientific than the existing single algorithm. Because the algorithm of the present invention uses mean square error and weighted average algorithm to obtain weight values ​​respectively, and finally calculates the reliability of each algorithm result according to the error of each algorithm result, reduces the error and then integrates it to obtain the optimal measurement of sound arrival time. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0025] Figure 1 A schematic diagram of a method for calculating the arrival time of sound in acoustic tomography based on ensemble learning provided in an embodiment of the present invention.

[0026] Figure 2 A schematic diagram of the integrated learning training process provided by an embodiment of the present invention.

[0027] Figure 3 A schematic diagram of the integrated learning calling process provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] The object of the present invention is to provide a method capable of improving the accuracy of river flow velocity detection.

[0030] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] Embodiment 1:

[0032] This embodiment provides a method for calculating the arrival time of sound in acoustic tomography based on ensemble learning, including:

[0033] Step 1: Use multiple algorithms to calculate the sound arrival time and obtain several calculation results;

[0034] Step 2: Compare the calculated results with the true results to calculate the errors of the results of each algorithm;

[0035] Step 3: Calculate the reliability of each algorithm result based on the error of each algorithm result, and then perform fusion analysis and calculation;

[0036] Step 4: The new data is calculated through the algorithm to obtain the arrival time results of each sound, and then integrated into the final result through integrated learning to improve the accuracy and stability of the algorithm.

[0037] Furthermore, the multiple algorithms include an edge detection algorithm, a peak detection algorithm and a prediction model algorithm, which collect 10 consecutive groups of sound signal data and calculate 10×3 sound arrival time results through three basic algorithms.

[0038] Serial number Peak Time Boundary Time Prediction time Real time 1 1.741965 1.74195693 1.74196096 1.74196096 2 1.74195 1.74196193 1.74196096 1.74195538 3 1.74195682 1.74196836 1.74195538 1.74196323 4 1.74195761 1.74196693 1.74196323 1.74196219 5 1.74196287 1.74197193 1.74196219 1.74196778 6 1.74196462 1.74197693 1.74196778 1.74197114 7 1.74196654 1.74198193 1.74197114 1.74197472 8 1.74197169 1.74197943 1.74197472 1.74197562 9 1.74197 1.74198193 1.74197562 1.74197601 10 1.74197 1.74198193 1.74197601 1.74197596

[0039] The error of the algorithm results is calculated as follows: The mean square error is used to describe the result error. Assuming that the sound arrival time calculated by an algorithm at the i-th moment is The actual arrival time is t i , a total of n groups of data are calculated, then the mean square error of the algorithm is:

[0040] Peak Detection Results Edge detection results Prediction model results Mean square error <![CDATA[2.932×10 -11 ]]> <![CDATA[2.434×10 -11 ]]> <![CDATA[2.988×10 -11 ]]>

[0041] The step three specifically includes: using a weighted average algorithm, assuming that there are n algorithm results in total, and the error of each result is e i , then the weight corresponding to the i-th result is:

[0042] Peak Detection Results Edge detection results Prediction model results Weighted Weight 0.314 0.378 0.308

[0043] The calculation result t of the weighted average algorithm is:

[0044]

[0045]

[0046] According to the mean square error formula (1), the mean square error between the weighted average result and the true result is 2.820×10 -12 , which proves that the results are better than those of the three basic algorithms.

[0047] Furthermore, the step three also includes: using the Kalman filter algorithm, assuming that the sound arrival time result obtained by the prediction model algorithm is t pred , error Q = e pred , the result obtained by the edge detection algorithm is t obs1 , the error is e obs1 , the result obtained by the peak detection algorithm is t obs2 , the error is e obs2 , the initial fusion error is P, then the five calculation formulas of Kalman filtering are:

[0048]

[0049]

[0050]

[0051]

[0052]

[0053] The optimal sound arrival time t and the updated fusion error P are obtained, and the updated fusion error is used in the next Kalman filter algorithm calculation.

[0054] serial number Kalman filter Real results Fusion Error serial number Kalman filter Real results Fusion Error 1 1.74196059 1.74196096 <![CDATA[1.321×10 -11 ]]> 6 1.74197652 1.74197596 <![CDATA[1.313×10 -11 ]]> 2 1.74196309 1.74196277 <![CDATA[1.313×10 -11 ]]> 7 1.74197652 1.74197597 <![CDATA[1.313×10 -11 ]]> 3 1.74196776 1.74196778 <![CDATA[1.313×10 -11 ]]> 8 1.74198704 1.74198708 <![CDATA[1.313×10 -11 ]]> 4 1.74197486 1.74197524 <![CDATA[1.313×10 -11 ]]> 9 1.74199644 1.74199643 <![CDATA[1.313×10 -11 ]]> 5 1.74197650 1.74197605 <![CDATA[1.313×10 -11 ]]> 10 1.74199845 1.74199824 <![CDATA[1.313×10 -11 ]]>

[0055] According to the mean square error formula (1), the mean square error between the Kalman filter result and the true result is 1.259×10 -13 , which proves that the results are better than those of the three basic algorithms.

[0056] like Figure 2 As shown, the integrated learning training process in the present invention is as follows:

[0057] Use a variety of basic algorithms to calculate the acoustic tomography data and obtain n sound arrival time results. Compare the actual sound arrival time with the result to calculate the result error, and then calculate the proportion of each error in the overall algorithm to obtain the weight of the basic algorithm result and complete the training of ensemble learning.

[0058] like Figure 3 As shown, the integrated learning calling process in the present invention is as follows:

[0059] A variety of basic algorithms are used to calculate the acoustic tomography data to obtain n sound arrival time results. All the results are fused according to the weights to obtain the optimal sound arrival time result. The method of the present invention minimizes the influence of factors such as basin topography, temperature, and the equipment itself on the sound signal, and improves the accuracy of measuring the sound arrival time during the acoustic tomography flow measurement process.

[0060] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0061] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for calculating sound arrival time based on ensemble learning, characterized in that: The following steps are involved: Step 1: using multiple algorithms to calculate the sound arrival time respectively and obtain several calculation results; the algorithms include edge detection algorithm, peak detection algorithm and prediction model algorithm; Step 2: Compare the calculated results with the true results to calculate the errors of the results of each algorithm; Step 3: Calculate the reliability of each algorithm result based on the error of each algorithm result and then merge them; Step 4: The new data is calculated by the algorithm to obtain the arrival time of each sound, and then integrated into the final result through ensemble learning; The step three also includes: using the Kalman filter algorithm, assuming that the sound arrival time result obtained by the prediction model algorithm is t pred , error Q = e pred , the result obtained by the edge detection algorithm is t obs1 , the error is e obs1 , the result obtained by the peak detection algorithm is t obs2 , the error is e obs2 , the initial fusion error is P, then the five calculation formulas of Kalman filtering are: The optimal sound arrival time t and the updated fusion error P are obtained, and the updated fusion error is used in the next Kalman filter algorithm calculation.

2. The method for calculating the sound arrival time by ensemble learning according to claim 1, characterized in that: The error of the algorithm results is calculated as follows: The mean square error is used to describe the result error. Assuming that the sound arrival time calculated by an algorithm at the i-th moment is The actual arrival time is t i , a total of n groups of data are calculated, then the mean square error of the algorithm is:

3. The method for calculating the sound arrival time by ensemble learning according to claim 1, characterized in that: The step three specifically includes: using a weighted average algorithm, assuming that there are n algorithm results in total, and the error of each result is e i , then the weight corresponding to the i-th result is: The calculation result t by the weighted average algorithm is:

Images