An improved temperature compensation method based on extreme learning machine algorithm

Abstract

The application discloses an improved temperature compensation method based on an extreme learning machine algorithm, and comprises the following steps: through analysis of factors influencing temperature compensation precision, it is found that the precision of a pressure monitoring sensor is mainly affected by thermal drift, therefore, an improved temperature compensation method is proposed, a calculation model of a thermistor is established, and a big gap between the calculation model of the thermistor and actual use is reduced, so that actual compensation effect is deteriorated; output voltages at different temperatures are converted into values at a fixed room temperature, a calibration coefficient at the room temperature is used, a compensation voltage is converted into an equivalent correction pressure value, and fitting error between actual output voltage and the compensation voltage is minimized; an extreme learning machine algorithm is used to further optimize calculation to obtain parameter values, and linear error is compensated. Experiment proves that the low-cost piezoresistive pressure sensor with the compensation algorithm is feasible in realizing the environmental self-adapting function, and provides certain reference for long-term unattended observation and monitoring under water.

Description

Technical Field

[0001] This invention relates to the field of oil and gas resource exploration, and in particular to an improved temperature compensation method based on the extreme learning machine algorithm. Background Technology

[0002] With my country's economic development, land-based resources are increasingly unable to meet the growing demand for oil and natural gas, leading to a strategic shift in oil and gas exploration and development from land to sea. Pressure monitoring sensors are used to monitor underwater oil pressure and are fundamental instruments in marine development equipment.

[0003] In recent years, underwater oil pressure monitoring sensors have faced significant challenges in design and verification, complex manufacturing processes, and high precision requirements, with no domestically available substitutes. The industry is also hampered by foreign technological blockades, high prices due to market monopolies, long delivery cycles, and difficulties in ensuring timely delivery, severely restricting the localization and industrialization of oil and gas monitoring equipment in my country. Furthermore, resolving the contradiction between the need for stable and accurate measurement of pressure sensors over extended periods and the inability of circuits to guarantee accuracy under prolonged operating conditions is crucial. Improving the real-time performance and accuracy of signal detection, as well as the reliability of intelligent signal processing, has become a mainstream demand in the industry. Summary of the Invention

[0004] To overcome the aforementioned problems in the existing technology, this invention proposes an improved temperature compensation method based on the extreme learning machine algorithm.

[0005] The technical solution adopted by this invention to solve its technical problem is: an improved temperature compensation method based on the extreme learning machine algorithm, comprising the following steps:

[0006] S1: Analyze the temperature characteristics of thermistors under traditional compensation methods to determine the factors affecting the temperature compensation accuracy of pressure sensors;

[0007] S2: Improve the thermistor model under the traditional compensation method in S1 and construct a suitable calculation model for thermistors;

[0008] S3: The thermistor model obtained in S2 is used to calculate the output voltage at different temperatures and convert it to a value at a fixed room temperature.

[0009] S4: Use the calibration coefficients at room temperature obtained in S3 to convert the compensation voltage into an equivalent correction pressure value;

[0010] S5: The equivalent corrected pressure value obtained in S4 is further optimized and calculated using the extreme learning machine algorithm to obtain the parameter value and compensate for the linear error.

[0011] The improved temperature compensation method based on the extreme learning machine algorithm described above, wherein the thermistor calculation model obtained in S2 is as follows:

[0012] R 1×N =k(R) X0 A X ,R Y0 A Y )R Y1×N

[0013] Where k(R) X0 A X ,R Y0 A Y R is a scaling factor related to the characteristic parameters of the thermistor to be used and the reference thermistor. X0 A X The characteristic parameters of the thermistor to be used; R Y0 A Y For reference the thermistor characteristic parameters, R Y1×N For reference, R 1×N R Y1×N All are N-dimensional row vectors.

[0014] The improved temperature compensation method based on the extreme learning machine algorithm described above, wherein step S3 specifically includes:

[0015] S31: According to the empirical formula of the Wheatstone bridge, the output voltage V under any pressure is... a It can be represented by a polynomial based on the room temperature condition:

[0016] V a =V a0 +(λ0+λ1V a0 (t-t0)+λ2V a0 (t-t0) 2

[0017] Among them, V a0 λ0 is the output voltage of the pressure sensor at standard room temperature t0, λ0 is the temperature correction coefficient at room temperature t0, t is the actual ambient temperature, and λ1 and λ2 are the linear and nonlinear temperature correction coefficients under arbitrary pressure, respectively.

[0018] S32: Based on equation S31, etc., establish an equivalent matrix to calculate the temperature correction coefficient:

[0019]

[0020] Where t1, t2, and t3 are three temperatures recorded during the test, and V a1 V a2 V a3 The output voltages at standard pressures are t1, t2, and t3, where t0 is the standard room temperature, and V. a0Let λ be the output voltage of the pressure sensor at standard room temperature t0, and λ0, λ1, and λ2 be the temperature correction coefficients at temperatures t1, t2, and t3.

[0021] The above-mentioned improved temperature compensation method based on the extreme learning machine algorithm, wherein the temperature compensation calculation method in S4 is specifically as follows: based on the derivation of the formula in S31, each output voltage V a The voltage can be corrected to standard room temperature t0 using the temperature correction factor in S32. The temperature compensation algorithm is as follows:

[0022]

[0023] Among them, V b V represents the compensated output voltage at different temperatures. a0 This is the compensated output voltage at standard room temperature t0, where t is the actual ambient temperature.

[0024] The improved temperature compensation method based on the extreme learning machine algorithm described above, wherein the equivalent correction pressure value obtained by S5 from S4 specifically includes:

[0025] S51: The input variables are the pressure and temperature digital signals corresponding to the two neurons in the input layer, and the output variable is the pressure network compensation, then the output z j And input x j Relationship between them:

[0026]

[0027] Where and α i β i These are the weight vectors connecting the input hidden layer and the hidden output layer, y i For the deviation of the i-th hidden neuron, Let be the number of hidden nodes, j be the number of samples, and f be the non-linear activation function;

[0028] S52: α i β i The output weight β of the i-th hidden layer unit is randomly generated within the range of 0 to 1. i The values ​​of i = 1, 2, ..., N will be obtained through matrix operations:

[0029] β=H + T

[0030] Where H is the output of the hidden layer node, H + It is the Moore-Penrose generalized inverse of matrix H, and T is the desired output.

[0031] The above-mentioned improved temperature compensation method based on the Extreme Learning Machine algorithm includes the following specific steps in S5 for temperature compensation using the Extreme Learning Machine algorithm:

[0032] S531: Normalize the sample data to the range of [-1,1] and measure within the pressure and temperature range of the sensor;

[0033] S532: Randomly divide the normalized sample data into training data and test data in a 2:1 ratio;

[0034] S533: Select the number of hidden nodes in an ordered manner from 1 to the number of training samples;

[0035] S534: Randomly initialize input weights and hidden layer biases, input training data, and calculate output weights;

[0036] S535: Calculate the output of the test data based on the weights and biases obtained from S534;

[0037] S536: Repeat S532 to S534 until satisfactory compensation accuracy is obtained;

[0038] S537: Write the calculated weights and biases into the microprocessor and verify the algorithm within the sensor's temperature and pressure range.

[0039] S538: Calculate the actual accuracy of the calibrated sensor.

[0040] The beneficial effects of this invention are as follows: It proposes a simple and low-cost temperature compensation algorithm. By employing online testing and proportional modeling of the thermistors, the significant discrepancy between the calculated model and actual usage is reduced, improving the accuracy of subsequent calculations. The output voltage at different temperatures is converted to a value at a fixed room temperature. Using a calibration coefficient at room temperature, the compensation voltage is converted into an equivalent correction pressure value, effectively correcting the experimental pressure value that drifts with temperature and successfully achieving independence from temperature changes. The extreme learning machine algorithm is used to optimize the calculation and obtain parameter values, thus providing higher accuracy. Furthermore, its low computational complexity and theoretically guided single-parameter settings make it more suitable for compensation. Experimental comparisons show that the pressure sensor exhibits small pressure errors during testing, demonstrating good similarity and coherence. The pressure sensor can still operate stably under conditions of electromagnetic interference, high pressure and high temperature, and strong vibration and shock, providing a certain reference for long-term unattended underwater observations. Attached Figure Description

[0041] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0042] Figure 1 This is a system architecture diagram of the present invention;

[0043] Figure 2 This invention relates to an online thermistor measurement network.

[0044] Figure 3 This invention compares the calculated and tested values ​​of the thermistor.

[0045] Figure 4 This is the temperature compensation voltage for the pressure sensor of the present invention;

[0046] Figure 5 This invention relates to the pressure sensor error under different standard pressure values.

[0047] Figure 6 This is the structure of the Extreme Learning Machine algorithm of the present invention;

[0048] Figure 7 The algorithm compensation error was configured at three pressure points at different temperatures for experimental calibration of this invention. Detailed Implementation

[0049] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] This embodiment discloses an improved temperature compensation method based on the Extreme Learning Machine algorithm, such as... Figure 1 As shown, the main components include five parts: determining the factors affecting the temperature compensation accuracy of the pressure sensor, constructing a calculation model for the thermistor, converting the output voltage at different temperatures to values ​​at a fixed room temperature, converting the compensation voltage to an equivalent corrected pressure value, and optimizing the parameter values ​​using an extreme learning machine algorithm. Specifically, the steps are as follows:

[0051] S1: Analyze the temperature characteristics of traditional thermistors to determine the factors affecting the temperature compensation accuracy of pressure sensors. In this embodiment, the reference pressure sensor model is Siemens WEPS-149. Other sensor models can be selected according to actual needs. In this embodiment, the temperature and pressure ranges are determined based on the range of the pressure sensor.

[0052] The specific steps for determining the factors affecting the temperature compensation accuracy of a pressure sensor include:

[0053] S11: Calculation model of thermistor under temperature influence in traditional compensation method:

[0054]

[0055] Where T is the absolute temperature, t is the actual ambient temperature, and R... a The resistance is the value at absolute temperature t, and m is the material constant. * These are advanced material constants;

[0056] S12: Analyzing the thermistor calculation model in S11, it can be seen that the resistance value of the thermistor changes with temperature due to the influence of the dissipation coefficient. In addition, in actual circuits, some components radiate heat, causing temperature gradients. Therefore, the thermistor modeling under traditional compensation has a large error compared with reality.

[0057] S2: Improve the thermistor model under the traditional compensation method in S1 and construct a suitable calculation model for thermistors;

[0058] The specific steps for constructing a suitable computational model for a thermistor include:

[0059] S21: Thermistor measurements should use circuit configurations and approximately equivalent circuit parameters, as well as a temperature testing range, that is, as close as possible to their actual operating conditions, rather than measurements under zero-power and narrow-temperature conditions. Figure 2 As shown. Among them, the resistors with arrows are thermistors (R4, R5, R7), and the rest are fixed resistors (R1, R2, R3, R6, R8);

[0060] S22: Considering the good consistency of thermistors in the same batch and the simplicity of engineering applications, a thermistor proportional modeling method can be adopted. The thermistor tested in S21 is used as the reference thermistor, and the remaining thermistors of the same type and similar resistance values ​​are modeled as follows:

[0061] R 1×N =k(R) X0 A X ,R Y0 A Y )R Y1×N

[0062] Where k(R) X0 A X ,R Y0 A Y R is a scaling factor related to the characteristic parameters of the thermistor to be used and the reference thermistor. X0 A X The characteristic parameters of the thermistor to be used; R Y0 A Y For reference the thermistor characteristic parameters, R Y1×N For reference, R 1×N R Y1×N All are N-dimensional row vectors.

[0063] S23: Based on the online measurement and proportional modeling method of thermistors, connect the two groups of three thermistors to... Figure 2 Measurements are performed within the range of -40 to +85℃ in the measurement network;

[0064] S24: Use the three thermistors in group 1 as reference thermistors for their corresponding network locations. Model and calculate the three thermistors in group 2 according to the formula described in S22. The comparison results of the modeled calculated values, their measured values, and the values ​​calculated using the traditional method are shown in [link to documentation]. Figure 3 , Figure 3 In the figures (1), (2), and (3), the results of comparing the calculated values ​​of R4, R5, and R7 with their actual measured values ​​and the calculated values ​​of the traditional method are shown. It can be seen that the calculated values ​​of the improved method in this embodiment are basically the same as the measured values.

[0065] S3: The thermistor model obtained in S2 is used to calculate the output voltage at different temperatures and convert it to a value at a fixed room temperature.

[0066] By improving the thermistor model, the specific steps for converting the output voltage at different temperatures to a value at a fixed room temperature are as follows:

[0067] S31: According to the empirical formula of the Wheatstone bridge, the output voltage V under any pressure is... a It can be represented by a polynomial based on the room temperature condition:

[0068] V a =V a0 +(λ0+λ1V a0 (t-t0)+λ2V a0 (t-t0) 2

[0069] Among them, V a0 λ0 is the output voltage of the pressure sensor at standard room temperature t0, λ0 is the temperature correction coefficient at room temperature t0, t is the actual ambient temperature, and λ1 and λ2 are the linear and nonlinear temperature correction coefficients under arbitrary pressure, respectively.

[0070] S32: Based on equation S31, etc., establish an equivalent matrix to calculate the temperature correction coefficient:

[0071]

[0072] Where t1, t2, and t3 are three temperatures recorded during the test, and V a1 V a2 V a3 The output voltages at standard pressures are t1, t2, and t3, where t0 is the standard room temperature, and V. a0 Let λ0, λ1, and λ2 be the output voltage of the pressure sensor at standard room temperature t0, and λ0, λ1, and λ2 be the temperature correction coefficients at temperatures t1, t2, and t3, respectively.

[0073] S4: Use the calibration coefficients at room temperature obtained in S3 to convert the compensation voltage into an equivalent correction pressure value;

[0074] The specific steps for converting the compensation voltage into an equivalent correction pressure value using calibration coefficients at room temperature include:

[0075] S41: Based on the derivation of equation S31, each output voltage V a The voltage can be corrected to standard room temperature t0 using the temperature correction factor in S32. The temperature compensation algorithm is as follows:

[0076]

[0077] Among them, V b V represents the compensated output voltage at different temperatures. a0 The compensated output voltage is at standard room temperature t0, where t is the actual ambient temperature;

[0078] S42: See also Figure 4 Each compensated output voltage V under the same pressure b It is rarely affected by temperature. Its values ​​are stable and the accompanying temperature drift is largely eliminated. Furthermore, within a pressure range of 0–20 MPa, all output voltages of the sensor are easily distinguishable, achieving high voltage resolution.

[0079] S43: The compensation voltage V is calculated using a transfer polynomial. b The corresponding correction pressure value P is used to verify the error between the compensation pressure value and the standard pressure value:

[0080]

[0081] Where P is the correction pressure value, which is obtained by fitting the polynomial through a series of pressure calibration experiments at standard room temperature t0, and k0, k1, k2, and k3 are pressure calibration coefficients.

[0082] S44: See also Figure 4 The data in the data are used to convert all output compensation voltages into equivalent correction pressure values. Figure 5 The error between the calibrated pressure value P and the value of the internal reference pressure gauge is shown. It can be clearly seen that the error fluctuates around zero, with the maximum pressure error between -0.30% and 0.30% FS. The results show that the compensated sensor achieves an accuracy of 0.3% across the entire pressure range of 0–20 MPa. By applying this temperature compensation method, the experimental pressure value that drifted with temperature was corrected, successfully achieving independence from temperature changes.

[0083] S5: The corrected pressure value obtained in S4 is further optimized and calculated using the extreme learning machine algorithm to obtain the parameter value and compensate for the linear error.

[0084] The specific steps for further optimizing the calculation of parameter values ​​and compensating for linearity errors using the Extreme Learning Machine algorithm are as follows:

[0085] S51: The input variables are the pressure and temperature digital signals corresponding to the two neurons in the input layer, and the output variable is the pressure network compensation, such as... Figure 6 As shown, the output z j And input x j Relationship between them:

[0086]

[0087] Where and α i β i These are the weight vectors connecting the input hidden layer and the hidden output layer, y i For the deviation of the i-th hidden neuron, Let be the number of hidden nodes, j be the number of samples, and f be the non-linear activation function;

[0088] S52: α i β i The output weight β of the i-th hidden layer unit is randomly generated within the range of 0 to 1. i The values ​​of i = 1, 2, ..., N will be obtained through matrix operations:

[0089] β=H + T

[0090] Where H is the output of the hidden layer node, H + It is the Moore-Ponous generalized inverse of matrix H, and T is the desired output;

[0091] S53: Specific steps for temperature compensation using the Extreme Learning Machine algorithm:

[0092] Step 1: Normalize the sample data to the range of [-1,1], measure within the pressure range of 0 to 20 MPa and the temperature range of -40 to +85℃;

[0093] Step 2: Randomly divide the normalized sample data (voltage, temperature, external pressure) into training data and test data in a 2:1 ratio;

[0094] Step 3: Select the number of hidden nodes in an ordered manner from 1 to the number of training samples;

[0095] Step 4: Randomly initialize the input weights and hidden layer biases, input the training data, and calculate the output weights;

[0096] Step 5: Calculate the output of the test data based on the weights and biases obtained in Step 4;

[0097] Step 6: Repeat steps 2-4 until satisfactory compensation accuracy is obtained;

[0098] Step 7: Write the calculated weights and biases into the microprocessor and verify the algorithm within the range of -40 to 85°C and the pressure range of 0 to 20 MPa.

[0099] Step 8: Calculate the actual accuracy of the calibrated sensor.

[0100] S54: Since the number of calibration points and temperature are limiting factors for calibration capability, to fully test the algorithm's calibration capability and select an appropriate number of calibration points to achieve high accuracy, 5 × 11 = 55 samples were designed as experimental data, calibrated using 5 temperature points (-40, -10, 20, 50, 85) °C and 11 pressure points (0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20) MPa. See [link to relevant documentation] Figure 7 ,from Figure 7 As can be seen in (a), the compensation error at the three pressure points was reduced to 0.13%FS through algorithm optimization. Figure 7 In section (b), under the same temperature testing conditions, different pressure values ​​revealed that the error fluctuation was small, around 0.1%, indicating that the temperature fluctuation for error compensation was relatively small. (See also...) Figure 5 The compensation error before optimization was approximately 0.3% FS. After algorithm compensation, it was reduced to approximately 0.13% FS. (See [link to algorithm]). Figure 7 .

[0101] The above embodiments are merely exemplary embodiments of the present invention and are not intended to limit the present invention. The scope of protection of the present invention is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to the present invention within its spirit and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of the present invention.

Claims

1. An improved temperature compensation method based on the Extreme Learning Machine algorithm, characterized in that, Includes the following steps: S1: Analyze the temperature characteristics of thermistors under traditional compensation methods to determine the factors affecting the temperature compensation accuracy of pressure sensors; S2: Improve the thermistor model under the traditional compensation method in S1 and construct a suitable calculation model for thermistors; S3: The thermistor model obtained in S2 is used to calculate the output voltage at different temperatures and convert it to a value at a fixed room temperature. S4: Use the calibration coefficients at room temperature obtained in S3 to convert the compensation voltage into an equivalent correction pressure value; S5: The equivalent corrected pressure value obtained in S4 is further optimized and calculated using the extreme learning machine algorithm to obtain the parameter value and compensate for the linearity error. S3 specifically includes: S31: According to the empirical formula of the Wheatstone bridge, the output voltage V under any pressure is... a It can be represented by a polynomial based on the room temperature condition: Among them, V a0 It is the output voltage of the pressure sensor at standard room temperature t0. 0 is the temperature correction factor at room temperature t0, where t is the actual ambient temperature. 1 and 2 represents the linear and nonlinear temperature correction coefficients under arbitrary pressure, respectively; S32: Based on equation S31, etc., establish an equivalent matrix to calculate the temperature correction coefficient: Where t1, t2, and t3 are three temperatures recorded during the test, and V a1 V a2 V a3 The output voltages at standard pressures are t1, t2, and t3, where t0 is the standard room temperature, and V. a0 This represents the output voltage of the pressure sensor at standard room temperature t0. 0、 1. 2 is the temperature correction factor at temperatures t1, t2, and t3; The S4 temperature compensation calculation method is specifically as follows: based on the derivation of the formula in S31, each output voltage V... a The voltage can be corrected to standard room temperature t0 using the temperature correction factor in S32. The temperature compensation algorithm is as follows: Among them, V b V represents the compensated output voltage at different temperatures. a0 This is the compensated output voltage at standard room temperature t0, where t is the actual ambient temperature.

2. The improved temperature compensation method based on the Extreme Learning Machine algorithm according to claim 1, characterized in that, The thermistor calculation model obtained in S2 is as follows: Where k(R) X0 A X ,R Y0 A Y R is a scaling factor related to the characteristic parameters of the thermistor to be used and the reference thermistor. X0 A X The characteristic parameters of the thermistor to be used; R Y0 A Y For reference the thermistor characteristic parameters, R Y1×N For reference, R 1×N R Y1×N All are N-dimensional row vectors.

3. The improved temperature compensation method based on the Extreme Learning Machine algorithm according to claim 1, characterized in that, The equivalent corrected pressure value obtained by S5 from S4 specifically includes: S51: The input variables are the pressure and temperature digital signals corresponding to the two neurons in the input layer, and the output variable is the pressure network compensation, then the output z j And input x j Relationship between them: Where and α i β i These are the weight vectors connecting the input hidden layer and the hidden output layer, y i Let be the bias of the i-th hidden neuron, ã be the number of hidden nodes, j be the number of samples, and f be the non-linear activation function; S52: α i β i The output weight β of the i-th hidden layer unit is randomly generated within the range of 0 to 1. i The values ​​of i = 1, 2, ..., N will be obtained through matrix operations: Where H is the output of the hidden layer node, H + It is the Moore-Penrose generalized inverse of matrix H, and T is the desired output.

4. The improved temperature compensation method based on the Extreme Learning Machine algorithm according to claim 1, characterized in that, The specific steps for temperature compensation using the Extreme Learning Machine algorithm in S5 include: S531: Normalize the sample data to the range of [-1,1] and measure within the pressure and temperature range of the sensor; S532: Randomly divide the normalized sample data into training data and test data in a 2:1 ratio; S533: Select the number of hidden nodes in an ordered manner from 1 to the number of training samples; S534: Randomly initialize input weights and hidden layer biases, input training data, and calculate output weights; S535: Calculate the output of the test data based on the weights and biases obtained from S534; S536: Repeat S532~S534 until satisfactory compensation accuracy is obtained; S537: Write the calculated weights and biases into the microprocessor and verify the algorithm within the sensor's temperature and pressure range. S538: Calculate the actual accuracy of the calibrated sensor.

Images