Temperature compensation method of ultrasonic sensor

Abstract

The invention discloses a temperature compensation method for an ultrasonic sensor, and the method comprises the following steps: carrying out the temperature characteristic test of the ultrasonic sensor, and obtaining the test data of an experimental sample and the actual test data; training of a neural network model is completed by utilizing test data of the experimental sample; adopting a statistical method to select and reject abnormal values of actual test data; and calculating temperature compensation of the ultrasonic sensor through a neural network model based on the processed actual test data. The GA-BP model used in the invention combines the advantages of the genetic algorithm and the back propagation neural network, not only fully utilizes the learning and expression ability of the neural network, but also realizes the global optimization of the training parameters of the neural network by means of the genetic algorithm, thereby improving the overall performance and prediction precision of the model.

Description

Technical Field

[0001] The invention belongs to the technical field of partial discharge ultrasonic monitoring of electric power equipment, and in particular relates to a temperature compensation method of an ultrasonic sensor. Background Art

[0002] Under the extremely cold conditions in Northeast my country, temperature has become a key factor in the increased failure rate of power equipment condition monitoring devices. According to the "extreme cold" standard defined by meteorology, that is, when the temperature is below -40°C, the physical, mechanical and electrical properties of the sensor equipment will change significantly. Under the continuous influence of the low temperature environment, the condition monitoring devices in the power system show the characteristics of reduced accuracy and stable operation, and frequently face the risk of functional abnormalities of electronic components and circuit boards due to low temperatures. At the same time, the materials used for sealing show signs of performance degradation, and the cable material becomes more fragile at low temperatures, increasing the possibility of breakage. In addition, in a low temperature environment, the battery charging and discharging performance of the monitoring device is severely degraded, usually only about 30% of that at normal temperature, which constitutes a major constraint on the stable operation of the transmission line condition monitoring device in low temperature areas.

[0003] At present, although comprehensive research has been conducted on the intelligence and communication transmission of online monitoring systems at home and abroad, specific research on the characteristics of online monitoring systems and fault resolution in low-temperature environments is relatively scarce, and no in-depth research on reliability improvement has been conducted on the above-mentioned low-temperature faults. The detection methods for partial discharge of equipment in the power grid mainly use ultrasonic method, geoelectric wave method, ultra-high frequency method, high-frequency current method and other technologies for detection. The present invention takes the partial discharge ultrasonic monitoring device as the main research object, and proposes a temperature compensation method specifically for the performance degradation problem of partial discharge ultrasonic sensors caused by changes in various parameters in low-temperature environments. Summary of the invention

[0004] The technical solution adopted by the present invention to solve the technical problem is: a temperature compensation method of an ultrasonic sensor, characterized in that the method comprises the following steps: S100, performing a temperature characteristic test on the ultrasonic sensor to obtain experimental sample test data and actual test data; S200, completing the training of the neural network model using the experimental sample test data; S300, using statistical methods to select and eliminate abnormal values ​​of actual test data; S400, based on the processed actual test data, calculate the temperature compensation of the ultrasonic sensor through the neural network model.

[0005] Preferably, the temperature characteristic test in step S100 includes: testing the main resonant frequency of the ultrasonic sensor at different temperatures to obtain its variation pattern with temperature.

[0006] Preferably, the temperature characteristic test in step S100 further includes: at different temperatures, based on the variation pattern of the main resonant frequency with temperature, testing the average output voltage amplitude of the ultrasonic sensor, obtaining the variation pattern of the average output voltage amplitude with temperature, so as to obtain experimental sample test data.

[0007] Preferably, in step S200, the neural network model is a back propagation neural network model optimized by a genetic algorithm.

[0008] Preferably, the training method of the back propagation neural network model optimized by the genetic algorithm comprises: Taking the experimental sample test data of temperature and ultrasonic sensors as input, a BP neural network model is constructed; The weights and threshold length of the BP neural network are optimized through genetic algorithm, and the training error of the BP neural network is used as the fitness value.

[0009] Preferably, the implementation process of the genetic algorithm part includes: Initialize first and randomly generate a set of initial solutions P g , g is the current evolutionary generation, set the maximum evolutionary generation G and make the current evolutionary generation g=0; Calculate the solution P according to the fitness function g Fitness evaluation of each solution in ; Apply selection operator, crossover operator and mutation operator to solution P g , after selecting crossover mutation, a new generation of solutions P is obtained g+1 , this process is repeated in a cycle; When the evolutionary generation reaches the maximum evolutionary number, that is, g=G, or the fitness of the optimal solution reaches the threshold, the genetic algorithm will terminate, and the optimal weights and thresholds obtained through multiple iterations will be output to the BP neural network; Then the neural network calculates the error and updates the weight threshold, and determines the termination condition, finally obtaining a fully trained GA-BP neural network model.

[0010] Preferably, the temperature compensation comprises: using an algorithm to compensate the ultrasonic sensor.

[0011] Preferably, the algorithm compensation includes: using the experimental sample test data to complete the training of the GA-BP neural network model, and then when the actual temperature and the measured value are input, firstly determine whether the value is an abnormal value through the Grubb test, if not, the true value can be automatically predicted through the optimal calculation of the model.

[0012] The technical advantages of the present invention are as follows: The GA-BP model used in the present invention combines the advantages of genetic algorithms and back propagation neural networks. It not only fully utilizes the learning and expression capabilities of neural networks, but also uses genetic algorithms to achieve global optimization of neural network training parameters, thereby improving the overall performance and prediction accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A flow chart of a temperature compensation method for an ultrasonic sensor in one embodiment of the present invention; FIG2 (a) shows the resonant frequency of the ultrasonic sensor at different temperatures in one embodiment of the present invention; FIG2( b ) is an average output voltage peak value of an ultrasonic sensor at different temperatures in one embodiment of the present invention; Figure 3 It is a GA-BP algorithm optimization flow chart in one embodiment of the present invention; Figure 4 This is a comparison chart of the predicted value and expected value of the GA-BP neural network in one embodiment of the present invention. DETAILED DESCRIPTION

[0014] The following will refer to the attached Figures 1 to 4 Specific embodiments of the present invention are described in detail. Although specific embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0015] It should be noted that certain words are used in the specification and claims to refer to specific components. Those skilled in the art should understand that technicians may use different nouns to refer to the same component. This specification and claims do not use the difference in nouns as a way to distinguish components, but use the difference in the functions of the components as the criterion for distinction. As mentioned throughout the specification and claims, "including" or "comprising" is an open term, so it should be interpreted as "including but not limited to". The subsequent description of the specification is a preferred embodiment of the present invention, but the description is based on the general principles of the specification and is not intended to limit the scope of the present invention. The scope of protection of the present invention shall be determined by the attached claims.

[0016] To facilitate understanding of the embodiments of the present invention, further explanation will be given below by taking specific embodiments as examples in conjunction with the accompanying drawings, and each of the accompanying drawings does not constitute a limitation on the embodiments of the present invention.

[0017] The present invention provides a temperature compensation method for an ultrasonic sensor, comprising the following steps: Conduct temperature characteristic test on ultrasonic sensor and obtain experimental sample test data; perform temperature compensation according to the temperature characteristic test results; specifically, adopt algorithm compensation method for ultrasonic sensor, use experimental sample test data to complete the training of GA-BP neural network model (back propagation neural network model optimized by genetic algorithm), and then when the actual temperature and measurement value are input, first use Grubb test (maximum standard residual test) to determine whether the value is an abnormal value, if not, the predicted value can be automatically calculated through model optimal calculation to complete sensor temperature compensation.

[0018] In one embodiment, see Figure 1 , a temperature compensation method for an ultrasonic sensor, comprising the following steps: S100, performing a temperature characteristic test on the ultrasonic sensor to obtain experimental sample test data and actual test data; S200, completing the training of the neural network model using the experimental sample test data; S300, using statistical methods to select and eliminate abnormal values ​​of actual test data; S400, based on the processed actual test data, calculate the temperature compensation of the ultrasonic sensor through the neural network model.

[0019] Furthermore, firstly, the main resonant frequency and average output voltage amplitude of the four selected ultrasonic sensors were tested in a wide temperature range (-60~30℃) with temperature as a single variable factor to obtain their variation patterns with temperature, see Figure 2 (a) to Figure 2 (b).

[0020] It can be seen intuitively from Figure 2 (a) that the main resonant frequency of the four groups of ultrasonic sensors increases as the temperature decreases, and the relationship with the temperature change is approximately linear. The reason is that the temperature decrease changes the elastic constant of the piezoelectric ceramic, which leads to an increase in the main resonant frequency of the ultrasonic sensor. When the temperature is between -50℃ and 20℃, the main resonant frequency increases by 1.06% for every 10℃ drop, which is a large drift for the filter amplifier circuit after the ultrasonic sensor. The effective signal is likely to be filtered and cause measurement failure. In addition, the main resonant frequency is the center frequency of the sensor. The effect of temperature on the center frequency will cause the detection bandwidth of the ultrasonic sensor to shift. According to this change law, it is found that the center frequency is 40kHz at 20℃, but when it is -50℃, the center frequency becomes 43kHz. The range of the center frequency in the experimental temperature range is 40-43kHz. Therefore, when obtaining the change law of the average output voltage with temperature, the center frequency band of the standard ultrasonic signal source used should also be within the above range.

[0021] As can be seen from Figure 2 (b), the voltage peaks of the four groups of ultrasonic sensors all decrease as the temperature decreases. Generally, the main focus is on the decrease in sensitivity of ultrasonic sensors when they are close to the Curie temperature at high temperatures, but the output of ultrasonic sensors will also decrease under the influence of low temperatures. It can be seen that between -10℃ and 20℃, the sensitivity of ultrasonic sensors is relatively stable and does not change much with temperature changes; when the temperature is below -20℃, the output of ultrasonic sensors decreases rapidly with the decrease in temperature. Taking AE1 as an example, when it decreases from -20℃ to -50℃, the voltage peak decreases from 4.3mV to 3.7mV, a decrease of 13.95%. If combined with the influence of the post-amplification circuit, this will be a large error for the measurement results. It can be seen that temperature drift compensation for ultrasonic sensors is very necessary.

[0022] In another embodiment, the temperature compensation of the ultrasonic sensor is divided into two steps: the first step is to use statistical methods to select and eliminate abnormal values. Assuming that the distribution of the data is normal distribution, through Grubb test, for each data in the received data set (i.e., the experimental sample data of the ultrasonic sensor), the z-score is defined as: ,in is the mean of the data, and s is the standard deviation. If formula (1) is satisfied, then It is an abnormal point and can be removed.

[0023] (1) Where N is the number of samples in the data set, The significant level The value of the t-distribution under .

[0024] The second step is to calculate the temperature compensation of the processed ultrasonic sensor data and temperature data through the back propagation (BP) neural network model optimized by the genetic algorithm (GA) (see Figure 3 ). First, the experimental sample data of temperature and ultrasonic sensors are used as input to construct the BP neural network model; then the weights and threshold length of the BP neural network are optimized through the genetic algorithm (GA), and the training error of the BP neural network is used as the fitness value. The implementation process of the genetic algorithm (GA) part is as follows: first initialize and randomly generate a set of initial solutions P g , g is the current evolutionary generation, set the maximum evolutionary generation G and make the current evolutionary generation g=0; calculate the solution P according to the fitness function g The fitness evaluation of each solution in P; the selection operator, crossover operator, and mutation operator act on the solution P g , after selecting crossover mutation, a new generation of solutions P is obtained g+1 , this process is repeated; the termination condition of the GA algorithm is that any of the following conditions are met: ① When the evolutionary generation reaches the maximum evolutionary number, that is, g=G; ② The fitness of the optimal solution reaches the threshold. After the cycle optimization process is completed, the optimal weights and thresholds are obtained through multiple iterations of optimization and output to the BP neural network. Then the neural network calculates the error and updates the weight threshold, and finally obtains a fully trained GA-BP comprehensive model.

[0025] Based on the actual situation, a BP neural network with one input layer, one hidden layer and one output layer is established. The sample data input to the entire BP network is the characteristic temperature point and the output voltage value of the ultrasonic sensor corresponding to the temperature point, so the input layer nodes are 2; the number of nodes in the hidden layer needs to match the number of samples. Based on the actual data volume, the number of nodes in the hidden layer is set to 5; the output result of this model is the standard output voltage of the ultrasonic sensor, so the output layer node is 1. The structure of the entire BP neural network is a 2-5-1 structure. After the BP neural network is constructed, the initial weights and thresholds are optimally searched through GA. Set the solution scale to 50, the maximum evolutionary generation G to 100, the crossover probability to 0.8, the mutation probability to 0.01, and the generation gap GGAP to 0.93. The minimum error err=0.2922mV.

[0026] The GA-BP model combines the advantages of genetic algorithms and back-propagation neural networks. It not only makes full use of the learning and expression capabilities of neural networks, but also uses genetic algorithms to achieve global optimization of neural network training parameters, thereby improving the overall performance and prediction accuracy of the model.

[0027] In another embodiment, see Figure 4, after the temperature compensation correction of the GA-BP model, the output prediction value is basically close to the true value (expected value). It should be noted that the true value exists objectively and is not changed by the measurement system; the predicted value is the value after the measurement data is compensated by the algorithm; the true value is equal to the expected value. The purpose of the present invention is to expect the predicted value to be as close to the true value as possible and eliminate the influence of temperature factors on data accuracy. The present invention effectively reduces the output error of the ultrasonic sensor caused by temperature, and the absolute value of the error is effectively controlled within 0.5mV. Before compensation, the maximum drop of the ultrasonic sensor caused by low temperature was 22.10%; after compensation, the sensor output only changed by 4.54% compared with the standard value. The GA-BP neural network model achieves the best temperature compensation effect within the preset parameter range, ensuring the accuracy and stability of the model.

[0028] The above general description of the invention involved in this application and the description of its specific implementation methods should not be understood as limiting the technical solutions of the invention. Based on the disclosure of this application, those skilled in the art can, without violating the constituent elements of the invention involved, add, reduce or combine the disclosed technical features in the above general description or / and specific implementation methods (including examples) to form other technical solutions within the scope of protection of this application.

Claims

1. A temperature compensation method for an ultrasonic sensor, characterized in that: The method comprises the following steps: S100, performing a temperature characteristic test on the ultrasonic sensor to obtain experimental sample test data and actual test data; S200, completing the training of the neural network model using the experimental sample test data; S300, using statistical methods to select and eliminate abnormal values ​​of actual test data; S400, based on the processed actual test data, calculate the temperature compensation of the ultrasonic sensor through the neural network model.

2. The method according to claim 1, characterized in that Preferably, the temperature characteristic test in step S100 includes: testing the main resonant frequency of the ultrasonic sensor at different temperatures to obtain a variation pattern of the main resonant frequency with temperature.

3. The method according to claim 2, characterized in that The temperature characteristic test in step S100 also includes: at different temperatures, based on the variation of the main resonant frequency with temperature, testing the average output voltage amplitude of the ultrasonic sensor, obtaining the variation of the average output voltage amplitude with temperature, so as to obtain experimental sample test data.

4. The method according to claim 1, characterized in that: In step S200, the neural network model is a back propagation neural network model optimized by a genetic algorithm.

5. The method according to claim 4, characterized in that The training method of the back propagation neural network model optimized by the genetic algorithm includes: Taking the experimental sample test data of temperature and ultrasonic sensors as input, a BP neural network model is constructed; The weights and threshold length of the BP neural network are optimized through genetic algorithm, and the training error of the BP neural network is used as the fitness value.

6. The method according to claim 5, characterized in that The implementation process of the genetic algorithm part includes: Initialize first and randomly generate a set of initial solutions P g , g is the current evolutionary generation, set the maximum evolutionary generation G and make the current evolutionary generation g=0; Calculate the solution P according to the fitness function g Fitness evaluation of each solution in ; Apply selection operator, crossover operator and mutation operator to solution P g , after selecting crossover mutation, a new generation of solutions P is obtained g+1 , this process is repeated in a cycle; When the evolutionary generation reaches the maximum evolutionary number, that is, g=G, or the fitness of the optimal solution reaches the threshold, the genetic algorithm will terminate, and the optimal weights and thresholds obtained through multiple iterations will be output to the BP neural network; Then the neural network calculates the error and updates the weight threshold, and determines the termination condition, finally obtaining a fully trained GA-BP neural network model.

7. The method according to claim 1, characterized in that In step S400, the temperature compensation includes: using an algorithm to compensate the ultrasonic sensor.

Images