Microwave sensor performance detection system based on machine learning

Abstract

The invention discloses a microwave sensor performance detection system based on machine learning, and belongs to the technical field of microwave sensors. According to the system, by monitoring the temperature, the environment temperature and humidity and the vibration amplitude of the microwave sensor in real time, the moment when the operation deviation is large can be recognized in time, the stability of the operation state is judged through abnormal value calculation, and therefore the operation value is obtained; the system also identifies a baseline drift and calculates a drift mean speed by comparing the baseline value to an initial base value; analyzing a target peak and a miscellaneous peak in the spectrogram at the same time, judging the accuracy and interference degree of the spectrogram through a symmetric value and a miscellaneous peak interference value, and finally obtaining the symmetric value, the kurtosis and the miscellaneous peak interference value; the method can effectively distinguish the interference of the to-be-detected object and the miscellaneous peak; the system can calculate the confidence value of each detection result so as to quantify the reliability of the detection result and automatically judge the credibility of the detection result according to the confidence value.

Description

Technical Field

[0001] The present invention relates to the field of microwave sensor technology, and in particular to a microwave sensor data processing system and method based on machine learning, and belongs to the field of application of the combination of microwave sensor technology and machine learning. Background Art

[0002] Microwave sensor is a high-performance sensing device based on microwave resonance technology. It can identify and quantitatively analyze target substances by detecting the interaction between substances and microwave signals. This type of sensor has the advantages of high sensitivity, rapid response, and non-contact detection. It is widely used in environmental monitoring, medical diagnosis, industrial process control, food safety, and material characterization. Especially in the field of medical diagnosis, when using microwave sensors for biomolecule detection, it is necessary to have extremely high detection sensitivity and selectivity, so the performance requirements of the sensor are relatively high.

[0003] At present, the performance evaluation of microwave sensors is usually calibrated and verified using standard samples, but this method has the following limitations: First, microwave sensors are greatly affected by ambient temperature and humidity. This is because the sensitive area of ​​the microwave sensor is exposed to the surrounding medium, and the medium will change its dielectric constant due to changes in ambient temperature and humidity, thereby affecting its resonant frequency. Therefore, it is difficult to achieve real-time monitoring and evaluation of microwave sensor performance; second, there has been a lack of quantitative analysis of long-term performance changes of microwave sensors for a long time, and it is impossible to effectively predict and evaluate the performance attenuation trend of microwave sensors. These problems make it difficult to ensure the reliability of microwave sensor detection results in practical applications, and it is urgent to establish an intelligent data processing and performance evaluation system. Summary of the invention

[0004] Based on this, it is necessary to provide a microwave sensor data processing system and method based on machine learning to address the problems mentioned in the above background technology, and to realize intelligent data processing and adaptive performance evaluation by introducing machine learning algorithm technology.

[0005] The object of the present invention can be achieved by the following technical solutions: A microwave sensor performance detection system based on machine learning includes: a sensor hardware data acquisition module, a data fusion module and a performance detection module;

[0006] First, the sensor hardware data acquisition module communicates with the microwave sensor to collect and store operating information and spectrum information; the specific operating information includes sensor temperature T, vibration amplitude F, ambient temperature H and ambient humidity S; the spectrum information includes baseline and spectrum;

[0007] Secondly, the data fusion module monitors and analyzes the operation status and spectrum status of the microwave sensor based on the operation information of the microwave sensor during operation to obtain the performance parameters of the microwave sensor, where the performance parameters include the operation value PC and the spectrum value RG;

[0008] Finally, the performance detection module determines the confidence level of the output spectrum based on the received performance parameters to guide the output of the detection results, specifically:

[0009] Step 1: The technicians in this field first preliminarily set a confidence coefficient corresponding to each microwave sensor, and record it as δ; then start the microwave sensor and start collecting data, and perform formulaic calculation and analysis on the calculated operating value PC, spectrum value RG and confidence coefficient δ to obtain the confidence value U. The specific calculation formula is:

[0010]

[0011] Wherein α1 and α2 are respectively set weight constants, and their values ​​are set by technicians in this field according to actual needs; for example, α1 is set to 0.37, and α2 is set to 0.63;

[0012] Step 2: When the confidence value is greater than or equal to the set confidence threshold, the spectrum is output as an accurate result; when the confidence value is less than the set confidence threshold, the spectrum is recorded as a risk result and retested until the confidence value is greater than the set confidence threshold, and the spectrum is output as an accurate result; the number of repeated tests is accumulated. If the number of repeated tests is greater than the set number threshold, it is determined that the performance of the microwave sensor is relatively poor, and a sensor maintenance instruction is generated to the corresponding engineer;

[0013] Each time a test is performed, a historical test record is generated. Each specific test record includes the response time, recovery time and test result, where the test result is an accurate result or a risk result;

[0014] Step 3: Retrieve historical detection records, including the number of historical detections, the response time, the recovery time, and the detection results of each detection; perform cumulative analysis on the historical detection records to obtain the updated confidence coefficient δ, and update it to step 1.

[0015] In some embodiments, the specific steps of cumulatively analyzing historical detection records are:

[0016] The response time and recovery time of each detection are weighted to obtain the delay value;

[0017] Compare and analyze the delay value of each detection with the set delay interval. When the delay value is greater than the maximum value in the set delay interval, a high delay is accumulated; when the delay value is in the set delay interval, a moderate delay is accumulated; when the delay value is less than the minimum value in the set delay interval, a slight delay is accumulated; the accumulated times of high delay, moderate delay and slight delay are accumulated respectively, and recorded as L1, L2 and L3 respectively; the delay value of each detection is averaged to obtain the delay average value, which is recorded as L4;

[0018] The cumulative number of high delays L1, the cumulative number of medium delays L2, the cumulative number of mild delays L3 and the delay mean L4 are calculated and analyzed by formula to obtain the response value Lg; the specific calculation formula is:

[0019]

[0020] Wherein g1, g2, and g3 are respectively set weight constants, and g1>g2>g3>1; the delay value detected each time is averaged to obtain the delay mean;

[0021] The number of times the test results in the test records are accurate results and risk results is counted, and they are recorded as accurate times Q1 and risk times Q2 respectively; the accurate times Q1, risk times Q2 and response value Lg are normalized and their values ​​are taken, and the values ​​are calculated and analyzed by formula to obtain the confidence coefficient δ. The specific calculation formula is:

[0022]

[0023] Among them, γ1 and γ2 are the set weight constants respectively.

[0024] In some embodiments, the specific steps of performing data monitoring and analysis on the operating status are:

[0025] 3-1: Retrieve the operation information of the microwave sensor at each collection time. The specific operation information includes sensor temperature, vibration amplitude, ambient temperature and ambient humidity, and record them as Ti, Fi, Hi and Si respectively, where i = 1, 2, 3 ... I, I is a positive integer, I represents the total number of collection times of the microwave sensor in the detection time period, and i represents the number of any collection time; set each type of microwave sensor to correspond to a working parameter, and the specific standard working parameters include standard working temperature range, standard ambient temperature range and standard ambient humidity range, and record them as [BT1, BT2], [BH1, BH2] and [BS1, BS2] respectively;

[0026] 3-2: Calculate and analyze the distance between the sensor temperature, ambient temperature and ambient humidity at each acquisition time and the standard working temperature range, standard ambient temperature range and standard ambient humidity range to obtain the working temperature deviation value, ambient temperature deviation value and ambient humidity deviation value at each acquisition time, and record them as PTi, PHi and PSi respectively;

[0027] 3-3: Normalize the working temperature deviation value PTi, vibration amplitude Fi, ambient temperature deviation value PHi and ambient humidity deviation value PSi at each collection time and take their values, and perform formula calculation and analysis on the values ​​to obtain the deviation value PZi at each collection time; the specific calculation formula is:

[0028] PZi=log 2 (β1×e PTi +β2×e PHi +β3×e PSi +1)+β4×Fi

[0029] Wherein, β1, β2, β3, and β4 are respectively set weight constants, and their values ​​are set by technicians in this field according to the actual situation of the detection; when the deviation value is greater than the set deviation threshold, the acquisition moment is recorded as the abnormal deviation moment; thus, several abnormal deviation moments can be obtained and sorted in chronological order;

[0030] 3-4: Calculate the interval between two adjacent abnormal deviation moments and record it as C1, and calculate the average of the deviation values ​​of the two adjacent abnormal deviation moments to obtain the average deviation and record it as C2; normalize the interval between two adjacent abnormal deviation moments C1 and the deviation average C2 and take their values, and perform formula calculation and analysis on the values ​​to obtain the abnormal value PC of the two adjacent abnormal deviation moments; the specific calculation formula is:

[0031]

[0032] Wherein, β5 and β6 are respectively set weight constants, and their values ​​are set by those skilled in the art according to the actual detection conditions. For example, the value of β5 is 1.355, and the value of β6 is 1.109.

[0033] In some embodiments, the specific process of distance calculation analysis is:

[0034] The sensor temperature at each acquisition moment is compared and analyzed with the standard working temperature interval to obtain the working temperature deviation value at each acquisition moment, which is recorded as PTi; the specific comparison and analysis method is: when the sensor temperature is in the standard working temperature interval [BT1, BT2], the working temperature deviation is assigned to zero; when the sensor temperature is not in the standard working temperature interval, PTi = min [BT1-Ti, Ti-BT2], Ti is the sensor temperature, i = 1, 2, 3 ... I, I is a positive integer, I represents the total number of acquisition moments, and i represents the number of any one of the acquisition moments;

[0035] Similarly, the ambient temperature and ambient humidity at each collection moment are respectively calculated and analyzed with respect to the standard ambient temperature range and the standard ambient humidity range to obtain the ambient temperature deviation value and the ambient humidity deviation value at each collection moment.

[0036] In some embodiments, the specific steps of performing data monitoring and analysis on the operating status are:

[0037] 5-1: Retrieve the spectrum information output by the microwave sensor, including the baseline and spectrum; perform baseline drift analysis based on the baseline to obtain the drift average speed;

[0038] 5-2: Identify the absorption peak corresponding to the target analyte in the spectrum and mark it as the target peak, and mark other absorption peaks as miscellaneous peaks; make the symmetry axis of the target peak, and perform symmetry analysis based on the symmetry axis of the target peak to obtain the symmetry value;

[0039] 5-3: Fit the target peak using data fitting, extract characteristic data from the fitting curve, obtain the function expression f(x) after the target peak is fitted, and calculate the kurtosis based on the fitted target peak;

[0040] 5-4: Identify the impurity peaks in the spectrum, analyze the interference of the impurity peaks on the target peak, and obtain the impurity peak interference value;

[0041] 5-5: Normalize the drift average speed R, symmetry value AM, kurtosis K and miscellaneous peak interference value G and take their values, and perform formula calculation and analysis on the values ​​to obtain the spectrum value RG. The specific calculation formula is:

[0042]

[0043] Among them, λ1, λ2, λ3, and λ4 are respectively set weight constants, and their values ​​are set by those skilled in the art according to actual needs.

[0044] In some embodiments, the specific process of baseline drift analysis is:

[0045] Assume that there is an initial base value; compare the baseline value at each time point in the baseline with the initial base value. When the baseline value is greater than the initial base value, the baseline value is recorded as the drift value. Thus, several drift values ​​in the baseline can be obtained, and the interval length between two adjacent drift values ​​is calculated. The two adjacent drift values ​​are summed and then the sum is divided by the interval length to obtain the drift speed of the two adjacent drift values. The average of all drift speeds is calculated to obtain the drift average speed.

[0046] In some embodiments, the specific method for determining the symmetry axis of the target peak is as follows: take the vertex of the target peak, draw a vertical line along the vertex and intersect the horizontal axis, and the vertical line is the symmetry axis of the target peak.

[0047] In some embodiments, the specific process of performing symmetry analysis based on the symmetry axis of the target peak is:

[0048] Divide the target peak into a left half and a right half along the symmetry axis, overlap the left half and the right half along the symmetry axis, and calculate the intersection area and union area of ​​the left half and the right half in the overlapping state;

[0049] Substitute the intersection area M∩ and the union area M∪ into the given formula

[0050]

[0051] The symmetry value AM is calculated.

[0052] In some embodiments, the kurtosis is calculated based on the fitted target peak, and the specific calculation process is:

[0053] Retrieve the function expression f(x) after fitting the target peak, mark the coordinate of each data point in the target peak as xj, where j=1, 2, 3...J, J is a positive integer, J represents the total number of points constituting the target peak, and j represents the number of any one of the points; substitute all the points xj in the target peak into the set formula, and calculate the kurtosis K, the specific formula is as follows;

[0054]

[0055] In some embodiments, the interference of the miscellaneous peaks on the target peak is analyzed, and the specific process is as follows:

[0056] Calculate the distance between the clutter peak and the target peak. When the distance is greater than the maximum value in the set distance interval, a highly influential clutter peak is accumulated; when the distance is within the set distance interval, a moderately influential clutter peak is accumulated; when the distance is less than the minimum value in the set distance interval, a slightly influential clutter peak is accumulated; the accumulated numbers of highly influential clutter peaks, moderately influential clutter peaks and slightly influential clutter peaks are counted respectively, and recorded as Z1, Z2 and Z3 respectively; the distances corresponding to highly influential clutter peaks, moderately influential clutter peaks and slightly influential clutter peaks are calculated by averaging to obtain highly influential distances, moderately influential distances and slightly influential distances, and recorded as Z4, Z5 and Z6 respectively;

[0057] The cumulative number of highly influential impurity peaks Z1, the cumulative number of moderately influential impurity peaks Z2, the cumulative number of slightly influential impurity peaks Z3, the highly influential distance Z4, the moderately influential distance Z5 and the slightly influential distance Z6 are normalized and their values ​​are taken, and the values ​​are calculated and analyzed by formula to obtain the impurity peak interference value G; the specific calculation formula is:

[0058]

[0059] Among them, η1, η2, and η3 are set weight constants respectively, and η1>η2>η3>1.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] 1. Through real-time monitoring of sensor temperature, ambient temperature and humidity, and vibration amplitude, it is possible to promptly identify the moment when the operation deviation is large, and to determine the stability of the operation status by calculating the abnormal value to obtain the operation value; it is possible to accurately evaluate the operation status of the sensor and provide a reference basis for the accuracy of the test results;

[0062] 2. By comparing the baseline value with the initial base value, identifying the baseline drift and calculating the drift average speed, the stability of the spectrum is evaluated, and then the target peak and the miscellaneous peaks in the spectrum are analyzed. The symmetry value and the miscellaneous peak interference value are used to judge the accuracy and interference degree of the spectrum, and the symmetry value, kurtosis and miscellaneous peak interference value are obtained. It can effectively distinguish the interference between the target analyte and the miscellaneous peaks; the drift average speed, symmetry value, kurtosis and miscellaneous peak interference value are comprehensively analyzed to obtain the spectrum value, realize the accurate judgment of the spectrum state, and provide a reference basis for the accuracy of the test results;

[0063] 3. By sending the operating value and spectrum value as performance parameters to the performance detection module, a comprehensive evaluation of the operating status and spectrum status of the microwave sensor can be achieved. The multi-dimensional data fusion method can more comprehensively reflect the actual performance of the equipment and provide data support for subsequent performance detection and maintenance;

[0064] 4. By combining the operating value, spectrum value and confidence coefficient, the confidence value of each test result can be calculated to quantify the reliability of the test result, and the confidence level of the test result can be automatically determined accordingly. The results with low confidence levels can be retested to ensure the accuracy of the final output results. When the test results are at low confidence levels for multiple times, the system will accumulate the number of repeated tests and determine the sensor performance status based on the set number threshold. If the threshold is exceeded, the system will automatically generate a maintenance instruction to notify the engineer to perform maintenance to ensure the continued stable operation of the sensor.

[0065] In summary, the data fusion module and performance detection module provide a comprehensive evaluation of the operating status and spectrum status of the microwave sensor through real-time monitoring and historical data analysis, and ensure the accuracy of the detection results and the long-term stable operation of the sensor through confidence evaluation and automatic maintenance strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the conventional technology, the drawings required for use in the embodiments or the conventional technology descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0067] Figure 1 It is a principle block diagram of the present invention. DETAILED DESCRIPTION

[0068] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific embodiments disclosed below.

[0069] like Figure 1 As shown, the microwave sensor performance detection system based on machine learning includes a sensor hardware data acquisition module, a data fusion module and a performance detection module;

[0070] Microwave sensors can support multi-mode resonance (such as dual-band or broadband response) according to their design. This design enables microwave sensors to respond to a variety of objects to be detected (including gases and liquids). It should be noted that the microwave sensor specifically described in the present invention is suitable for a single target detection sensor, that is, the microwave sensor described in the present invention only detects a single target object to be detected.

[0071] The sensor hardware data acquisition module communicates with the microwave sensor to collect and store operating information and spectrum information; the specific operating information includes sensor temperature, vibration amplitude, ambient temperature and ambient humidity; the spectrum information includes baseline and spectrum;

[0072] The data fusion module monitors and analyzes the operation status and spectrum status of the microwave sensor based on its operation information during operation to obtain the performance parameters of the microwave sensor, where the performance parameters include operation values ​​and spectrum values. The specific steps of the specific monitoring and analysis are as follows:

[0073] Run the analysis:

[0074] The operation information of the microwave sensor at each acquisition time is retrieved. The specific operation information includes sensor temperature, vibration amplitude, ambient temperature and ambient humidity, and they are recorded as Ti, Fi, Hi and Si respectively, where i=1, 2, 3...I, I is a positive integer, I represents the total number of acquisition times of the microwave sensor in the detection time period, and i represents the number of any acquisition time; it should be noted that the material of the microwave sensor is sensitive to the temperature and humidity of the environment, and is easily affected, resulting in deviations in its detection results. Similarly, the microwave sensor is affected by mechanical stress and vibrates, resulting in an unstable baseline during detection, thereby causing deviations in the detection results;

[0075] Each type of microwave sensor is set to correspond to a working parameter. The specific standard working parameters include the standard working temperature range, the standard ambient temperature range and the standard ambient humidity range, which are recorded as [BT1, BT2], [BH1, BH2] and [BS1, BS2] respectively. It should be noted that the standard working parameters are determined by the production materials and production processes of the microwave sensor, and are usually marked in the instructions for use of the microwave sensor to provide clear specifications for the normal operation and stability of the sensor.

[0076] The sensor temperature at each acquisition moment is compared and analyzed with the standard working temperature range to obtain the working temperature deviation value at each acquisition moment, which is recorded as PTi; the specific comparison and analysis method is: when the sensor temperature is in the standard working temperature range [BT1, BT2], the working temperature deviation is assigned to zero; when the sensor temperature is not in the standard working temperature range, PTi = min [BT1-Ti, Ti-BT2];

[0077] The ambient temperature at each acquisition moment is compared and analyzed with the standard ambient temperature interval to obtain the ambient temperature deviation value at each acquisition moment, which is recorded as PHi; the specific comparison and analysis method is: when the ambient temperature is in the standard ambient temperature interval [BH1, BH2], the ambient temperature deviation is assigned to zero; when the ambient temperature is not in the standard ambient temperature interval, PHi = min [BH1-Hi, Hi-BH2];

[0078] The ambient humidity at each acquisition time is compared and analyzed with the standard ambient humidity interval to obtain the ambient humidity deviation value at each acquisition time, recorded as PSi; the specific comparison and analysis method is: when the ambient humidity is in the standard ambient humidity interval [BS1, BS2], the ambient humidity deviation is assigned to zero; when the ambient humidity is not in the standard ambient humidity interval, PSi = min [BS1-Si, Si-BS2];

[0079] The working temperature deviation value PTi, vibration amplitude Fi, ambient temperature deviation value PHi and ambient humidity deviation value PSi at each acquisition time are normalized and their values ​​are taken, and the values ​​are calculated and analyzed by formula to obtain the deviation value PZi at each acquisition time; the specific calculation formula is:

[0080] PZi=log 2 (β1×e PTi +β2×e PHi +β3×e PSi +1)+β4×Fi

[0081] Wherein β1, β2, β3, and β4 are respectively set weight constants, and their values ​​are set by technicians in this field according to the actual detection situation; compare the deviation value with the set deviation threshold value. When the deviation value is greater than the set deviation threshold value, it means that the operation deviation of the microwave sensor at the acquisition moment is relatively large, and the acquisition moment is recorded as the abnormal deviation moment; thus, several abnormal deviation moments can be obtained, and they are sorted in chronological order; the interval length between two adjacent abnormal deviation moments is calculated and recorded as C1, and the deviation values ​​of the two adjacent abnormal deviation moments are averaged to obtain the deviation mean value recorded as C2; the interval length C1 and the deviation mean value C2 of the two adjacent abnormal deviation moments are normalized and their values ​​are taken, and the values ​​are formulaically calculated and analyzed to obtain the abnormal value PC of the two adjacent abnormal deviation moments; the specific calculation formula is:

[0082]

[0083] Wherein, β5 and β6 are respectively set weight constants, and their values ​​are set by technicians in this field according to the actual situation of the detection. Specifically, the value of β5 is 1.355, and the value of β6 is 1.109. It can be seen from the formula that when the interval length is shorter, it means that the abnormal frequency is higher, and the abnormal value is larger; when the deviation mean is larger, it means that the degree of deviation is higher, and the abnormal value is larger; thus, the abnormal values ​​of all two adjacent abnormal deviation moments can be obtained, and the average value is calculated to obtain the operating value PC; it is not difficult to obtain from the calculation and analysis process of the operating value that when the operating value is larger, it means that the abnormality during the operation of the microwave sensor is more serious;

[0084] Through real-time monitoring of sensor temperature, ambient temperature and humidity, and vibration amplitude, it is possible to promptly identify moments when operating deviations are large, and to determine the stability of the operating state by calculating abnormal values ​​to obtain operating values; this enables accurate assessment of the operating state of the sensor, providing a reference for the accuracy of the test results.

[0085] Spectral analysis:

[0086] The baseline and spectrum output by the microwave sensor are retrieved. It should be noted that the microwave sensor needs to preheat and stabilize the baseline before detecting the target object to be tested, and then the target object to be tested will be detected. After the detection, the spectrum is output, so each detection will have a corresponding baseline and spectrum; an initial base value is set (usually the baseline value when the sensor starts working); the baseline value at each time point in the baseline is compared with the initial base value. When the baseline value is greater than the initial base value, it indicates that there is a baseline drift, and the baseline value is recorded as the drift value; thus, several drift values ​​in the baseline can be obtained, the interval between two adjacent drift values ​​is calculated, and the two adjacent drift values ​​are summed and calculated, and then the sum is divided by the interval to obtain the drift speed of the two adjacent drift values, and all drift speeds are averaged and the average drift speed is recorded as R; it should be noted that the larger the drift value and the shorter the interval between the drift values, the more serious the baseline drift is, and the larger the average drift speed is, and the greater the impact on the accuracy of the spectrum;

[0087] The absorption peak corresponding to the target analyte in the identification spectrum is marked as the target peak, and the other absorption peaks are recorded as miscellaneous peaks. It should be noted that each analyte molecule will absorb or reflect electromagnetic waves (such as microwaves, infrared rays, etc.) at a specific frequency or wavelength after combining with the sensitive material on the microwave sensor. These absorption peaks are caused by the interaction between the analyte molecules and the electromagnetic waves, and belong to the characteristic absorption spectrum of the gas; by analyzing the peak position in the spectrum, the characteristic absorption peak of the gas can be identified, thereby identifying the target peak in the spectrum;

[0088] Draw the symmetry axis of the target peak in the spectrum. The specific method is as follows: take the vertex of the target peak (the vertex refers to the highest point of the absorption peak on the spectrum, that is, the point where the signal intensity of the target peak is the largest in the spectrum), draw a vertical line along the vertex to intersect with the horizontal axis, and the vertical line is the symmetry axis of the target peak; it should be noted that, ideally, the target peak should be symmetrical along the symmetry axis; divide the target peak into a left half and a right half along the symmetry axis, overlap the left half and the right half along the symmetry axis, calculate the intersection area of ​​the left half and the right half in the overlapping state, record it as M∩, and the union area as M ∪ ; The intersection area M ∩ Substitute the area of ​​the union M∪ into the formula

[0089]

[0090] The symmetry value AM is obtained by calculation. It can be seen from the calculation formula that the greater the difference between the intersection area and the union area, the greater the asymmetry of the target peak, the greater the impact on the accuracy of the spectrum, and the smaller the symmetry value;

[0091] The target peak is fitted by data fitting, and characteristic data (specific characteristic parameters include peak position, peak width and amplitude, etc.) are extracted from the fitting curve. The model usually used is Gaussian distribution. The specific Gaussian distribution model is:

[0092]

[0093] Where A is the amplitude (indicating the peak height), μ is the peak position (i.e. the position of the symmetry axis), and σ is the standard deviation (indicating the width of the peak, the larger the value, the wider the peak, and the smaller the value, the sharper the peak);

[0094] Thus, the function expression f(x) after fitting of the target peak can be obtained, and the horizontal coordinate of each data point in the target peak is marked as xj (the horizontal coordinate in the ordinary language usually refers to the frequency or wavelength value), where j = 1, 2, 3...J, J is a positive integer, J represents the total number of points that make up the target peak, and j represents the number of any one of the points; substitute all the points xj in the target peak into the set formula to calculate the kurtosis K, and the specific calculation formula is:

[0095]

[0096] It should be noted that for an ideal Gaussian peak, its kurtosis is 3;

[0097] Calculate the distance between the miscellaneous peak and the target peak. It should be noted that the smaller the distance between the miscellaneous peak and the target peak, the less clear the separation between the miscellaneous peak and the target peak, and the greater the impact of the miscellaneous peak on the target peak. Compare and analyze the distance with the set distance interval. When the distance is greater than the maximum value in the set distance interval, a highly influential miscellaneous peak is accumulated. When the distance is within the set distance interval, a moderately influential miscellaneous peak is accumulated. When the distance is less than the minimum value in the set distance interval, a slightly influential miscellaneous peak is accumulated. The accumulated numbers of highly influential miscellaneous peaks, moderately influential miscellaneous peaks and slightly influential miscellaneous peaks are counted respectively. The values ​​are recorded as Z1, Z2 and Z3 respectively; the distances corresponding to the highly influential impurity peaks, the moderately influential impurity peaks and the slightly influential impurity peaks are respectively calculated by averaging to obtain the highly influential distance, the moderately influential distance and the slightly influential distance, which are recorded as Z4, Z5 and Z6 respectively; the cumulative number of highly influential impurity peaks Z1, the cumulative number of moderately influential impurity peaks Z2, the cumulative number of slightly influential impurity peaks Z3, the highly influential distance Z4, the moderately influential distance Z5 and the slightly influential distance Z6 are normalized and their values ​​are taken, and the values ​​are analyzed by formula calculation to obtain the impurity peak interference value G; the specific calculation formula is:

[0098]

[0099] Among them, η1, η2, and η3 are respectively the set weight constants, and η1>η2>η3>1. It can be seen from the formula that when the distance between the impurity peak and the target peak is smaller, it means that the interference to the target peak is greater, and the greater the impurity peak interference value, the greater the possibility of inaccuracy of the target peak;

[0100] The drift average speed R, symmetry value AM, kurtosis K and miscellaneous peak interference value G are normalized and their values ​​are taken, and the values ​​are calculated and analyzed by formula to obtain the spectrum value RG. The specific calculation formula is:

[0101]

[0102] Among them, λ1, λ2, λ3, and λ4 are respectively set weight constants, and their values ​​are set by the personnel in this field according to actual needs; it can be seen from the formula that when the drift average speed is larger, the spectrum value is smaller; when the symmetry value is larger, the spectrum value is larger; when the kurtosis K>3, it means that the peak shape is sharp and conforms to the ideal Gaussian distribution, then the spectrum value is larger; otherwise, it means that the peak shape is flat or blunt, then the spectrum value is smaller; when the miscellaneous peak interference value is larger, the spectrum value is smaller;

[0103] The operating values ​​and the spectrum values ​​are recorded as performance parameters and sent to a performance detection module;

[0104] By comparing the baseline value with the initial base value, identifying the baseline drift and calculating the drift average speed, the stability of the spectrum is evaluated, and then the target peak and the miscellaneous peak in the spectrum are analyzed. The symmetry value and the miscellaneous peak interference value are used to judge the accuracy and interference degree of the spectrum, and the symmetry value, kurtosis and miscellaneous peak interference value are obtained. It can effectively distinguish the interference between the target analyte and the miscellaneous peak; the drift average speed, symmetry value, kurtosis and miscellaneous peak interference value are comprehensively analyzed to obtain the spectrum value, realize the accurate judgment of the spectrum state, and provide a reference basis for the accuracy of the test results; by sending the operating value and the spectrum value as performance parameters to the performance detection module, the comprehensive evaluation of the microwave sensor operating state and the spectrum state is realized. The multi-dimensional data fusion method can more comprehensively reflect the actual performance of the equipment and provide data support for subsequent performance detection and maintenance.

[0105] The performance detection module determines the confidence level of the detection results of the microwave sensor based on the received performance parameters, and outputs the confidence level of the detection results accordingly; determines the performance of the microwave sensor based on the confidence level of the detection results output by the microwave sensor, and executes corresponding strategies to ensure the stable detection capability of the microwave sensor; specifically:

[0106] Step 1: Set a confidence coefficient corresponding to each microwave sensor and record it as δ; perform formulaic calculation and analysis on the operating value PC, the spectrum value RG and the confidence coefficient δ to obtain the confidence value U. The specific calculation formula is:

[0107]

[0108] Wherein α1 and α2 are respectively set weight constants, and their values ​​are set by technicians in this field according to actual needs; specifically, the value of α1 is 0.37, and the value of α2 is 0.63;

[0109] Step 2: Compare and analyze the confidence value with the set confidence threshold. When the confidence value is greater than or equal to the set confidence threshold, it means that the confidence level of the spectrum output by this detection is high, and the spectrum is output as an accurate result; when the confidence value is less than the set confidence threshold, it means that the confidence level of the spectrum output by this detection is relatively low, and there is a risk of comparison error, then the spectrum is recorded as a risk result, and re-detected until the confidence value is greater than the set confidence threshold, and the spectrum is output as an accurate result; accumulate the number of repeated detections, if the number of repeated detections is greater than the set number threshold (specifically set by personnel in this field according to actual needs, for example, engineers will set the number threshold to 5 times), the performance of the microwave sensor is relatively poor, and a sensor maintenance instruction is generated to the corresponding engineer;

[0110] Each time a test is performed, a historical test record is generated. Each specific test record includes response time (the time taken to start detecting the output spectrum. The longer the response time, the aging of the material or the decrease in the diffusion efficiency of the analyte, and the performance of the microwave sensor is reduced), recovery time (the time from the removal of the target analyte to the signal recovery to the baseline, indicating that the adsorption material has a hysteresis phenomenon or the removal efficiency is reduced, and the performance of the microwave sensor is reduced) and test results (accurate results or risk results);

[0111] Step 3: Retrieve historical detection records, including the number of historical detections, the response time, the recovery time and the detection results of each detection; perform weighted calculation on the response time and the recovery time of each detection to obtain the delay value; compare and analyze the delay value of each detection with the set delay interval. When the delay value is greater than the maximum value in the set delay interval, a high delay is accumulated; when the delay value is in the set delay interval, a moderate delay is accumulated; when the delay value is less than the minimum value in the set delay interval, a mild delay is accumulated; the cumulative number of high delays, moderate delays and mild delays are accumulated respectively, and recorded as L1, L2 and L3 respectively; the delay value of each detection is averaged to obtain the delay average recorded as L4; the cumulative number of high delays L1, the cumulative number of moderate delays L2, the cumulative number of mild delays L3 and the delay average L4 are formulaically calculated and analyzed to obtain the response value Lg; the specific calculation formula is:

[0112]

[0113] Where g1, g2, and g3 are respectively set weight constants, and g1>g2>g3>1; when the delay value of each detection is larger, the corresponding response value is smaller;

[0114] The number of times the test results in the test records are accurate results and risk results is counted, and they are recorded as accurate times Q1 and risk times Q2 respectively; the accurate times Q1, risk times Q2 and response value Lg are normalized and their values ​​are taken, and the values ​​are calculated and analyzed by formula to obtain the confidence coefficient δ. The specific calculation formula is:

[0115]

[0116] Where γ1 and γ2 are the set weight constants respectively, and the confidence coefficient is updated to step 1. By dynamically updating the confidence coefficient, the confidence evaluation strategy of the detection result is continuously optimized;

[0117] By combining the operating value PC, the spectrum value RG and the confidence coefficient δ, the confidence value U of each test result can be calculated, so as to quantify and evaluate the reliability of the test results, and automatically judge the credibility of the test results accordingly. For test results with low confidence, the system will automatically trigger re-testing to ensure the accuracy of the final output result; when the test results are in a low confidence state for many times, the system will accumulate the number of repeated tests and evaluate the performance of the sensor based on the set number threshold; if the number of repeated tests exceeds the threshold, the system will automatically generate a maintenance instruction to notify the engineer to perform maintenance to ensure the continuous and stable operation of the sensor.

[0118] It should be noted that the above formulas are based on a large amount of collected data, obtained through software numerical simulation, and select formula forms that are highly close to the actual values. The coefficients in the formulas are set by technical personnel in this field according to specific application scenarios and actual conditions.

[0119] The various technical features of the above-described embodiments can be freely combined as needed. To simplify the description, not all possible combinations of technical features are described in detail; however, as long as there is no contradiction in the combination of these technical features, they should be regarded as covered by the scope of the present specification.

[0120] The above embodiments only show several implementations of the present invention, and the descriptions thereof are relatively specific and detailed, but do not limit the scope of patent protection of the present invention. It should be pointed out that, for those of ordinary skill in the art, various deformations and improvements can be made to the present invention without departing from the core concept of the present invention, and these deformations and improvements all belong to the scope of protection of the present invention. Therefore, the scope of patent protection of the present invention shall be subject to the attached claims.

Claims

1. A microwave sensor performance detection system based on machine learning, including a sensor hardware data acquisition module, a data fusion module and a performance detection module; the characteristics are: The sensor hardware data acquisition module is connected to the microwave sensor to collect the operation information and spectrum data of the microwave sensor in real time; it is responsible for obtaining the temperature, vibration amplitude, ambient temperature and humidity operating parameters of the sensor, and recording the relevant spectrum information, including baseline and spectrum data; The data fusion module monitors and analyzes the operation status and spectrum status of the microwave sensor based on the operation information of the microwave sensor during operation, and obtains the performance parameters of the microwave sensor, wherein the performance parameters include the operation value and the spectrum value; The performance detection module determines the confidence level of the output spectrum based on the received performance parameters to guide the output of the detection results, specifically: Step 1: Set a confidence coefficient δ for each microwave sensor, and calculate and analyze the operating value PC, the spectrum value RG and the confidence coefficient δ to obtain the confidence value U; the confidence value indicates the quantitative index of the reliability of the detection result, the operating value represents the comprehensive parameter of the sensor operating state, and the spectrum value represents the comprehensive evaluation quality of the spectrum quality; Step 2: When the confidence value is greater than or equal to the set confidence threshold, the spectrum is output as an accurate result; when the confidence value is less than the set confidence threshold, the spectrum is recorded as a risk result and retested until the confidence value is greater than the set confidence threshold, and the spectrum is output as an accurate result; the number of repeated tests is accumulated. If the number of repeated tests is greater than the set number threshold, the performance of the microwave sensor is relatively poor, and a sensor maintenance instruction is generated to the corresponding engineer; After each detection, the data fusion module will generate a historical detection record. Each specific detection record includes the response time, recovery time and detection result, where the detection result is an accurate result or a risk result; Step 3: Retrieve historical detection records, including the number of historical detections, the response time, recovery time, and detection results of each detection; perform cumulative analysis on the historical detection records to obtain the confidence coefficient δ, and update it to step 1.

2. The microwave sensor performance detection system based on machine learning according to claim 1 is characterized in that: The specific steps for cumulative analysis of historical detection records are as follows: The response time and recovery time of each detection are weighted to obtain the delay value; Compare and analyze the delay value of each detection with the set delay interval. When the delay value is greater than the maximum value in the set delay interval, a high delay is accumulated; when the delay value is in the set delay interval, a moderate delay is accumulated; when the delay value is less than the minimum value in the set delay interval, a mild delay is accumulated; the cumulative number of high delays, moderate delays and mild delays is accumulated respectively, and the delay value of each detection is averaged to obtain the delay average; The cumulative number of high delays L1, the cumulative number of medium delays L2, the cumulative number of mild delays L3 and the delay mean L4 are calculated and analyzed by formula to obtain the response value Lg; The number of times the test results in the test records are accurate results and risky results is counted, and they are recorded as accurate number Q1 and risky number Q2 respectively; the accurate number Q1, risky number Q2 and response value Lg are normalized and their values ​​are taken, and the confidence coefficient δ is obtained by formulating and analyzing the values.

3. The microwave sensor performance detection system based on machine learning according to claim 1 is characterized in that: The specific steps for data monitoring and analysis of the operating status are as follows: 3-1: Retrieve the operating information of the microwave sensor at each acquisition time, including sensor temperature, vibration amplitude, ambient temperature and ambient humidity; set a corresponding operating parameter for each type of microwave sensor, including the standard operating temperature range [BT1, BT2], the standard ambient temperature range [BH1, BH2] and the standard ambient humidity range [BS1, BS2]; 3-2: Calculate and analyze the distance between the sensor temperature, ambient temperature and ambient humidity at each acquisition time and the standard working temperature range, standard ambient temperature range and standard ambient humidity range to obtain the working temperature deviation value PTi, ambient temperature deviation value PHi and ambient humidity deviation value PSi at each acquisition time; 3-3: Normalize the working temperature deviation value PTi, vibration amplitude Fi, ambient temperature deviation value PHi and ambient humidity deviation value PSi at each collection moment and take their values, calculate and analyze the values ​​in a formula to obtain the deviation value PZi at each collection moment. The specific formula is as follows; when the deviation value PZi is greater than the set deviation threshold PZ0, the collection moment is recorded as an abnormal deviation moment; thus, several abnormal deviation moments can be obtained and sorted in chronological order; 3-4: Calculate the interval time C1 between two adjacent abnormal deviation moments, and calculate the mean of the deviation values ​​between two adjacent abnormal deviation moments to obtain the deviation mean C2; normalize the interval time and the deviation mean between two adjacent abnormal deviation moments and take their numerical values, perform formulaic calculation and analysis on the numerical values ​​to obtain the abnormal value PCi of two adjacent abnormal deviation moments; thus, the abnormal values ​​PC of all two adjacent abnormal deviation moments can be obtained, and the average is calculated to obtain the operating value PC.

4. The microwave sensor performance detection system based on machine learning according to claim 3 is characterized in that: The specific process of distance calculation analysis is as follows: The sensor temperature at each acquisition time is compared with the standard operating temperature range to obtain the operating temperature deviation value at each acquisition time, which is recorded as PT i ; The specific comparative analysis method is: when the sensor temperature is in the standard operating temperature range [BT1, BT2], the operating temperature deviation is assigned to zero; when the sensor temperature is not in the standard operating temperature range, PT i =min[BT1-T i , T i -BT2], T i is the sensor temperature, i=1,2,3……I, I is a positive integer, I represents the total number of acquisition moments, and i represents the number of any acquisition moment; Similarly, the ambient temperature and ambient humidity at each collection moment are respectively calculated and analyzed with respect to the standard ambient temperature range and the standard ambient humidity range to obtain the ambient temperature deviation value and the ambient humidity deviation value at each collection moment.

5. According to the microwave sensor performance detection system based on machine learning in claim 1, the specific steps of performing data monitoring and analysis on the sensor operation status are: 5-1: Retrieve the spectrum information output by the microwave sensor, including the baseline and spectrum; perform baseline drift analysis based on the baseline to obtain the drift average speed R; 5-2: Identify the absorption peak corresponding to the target substance in the spectrum and mark it as the target peak, and mark other absorption peaks as miscellaneous peaks; make the symmetry axis of the target peak, and perform symmetry analysis based on the symmetry axis of the target peak to obtain the symmetry value; 5-3: Use the data fitting function of the computer to fit the target peak, and extract the characteristic data from the fitting curve to obtain the function expression f(x) after the target peak is fitted. Calculate the kurtosis based on the fitted target peak. The specific analysis formula is shown below; 5-4: Identify the impurity peaks in the spectrum, analyze the interference of the impurity peaks on the target peak, and obtain the impurity peak interference value; 5-5: Normalize the drift average speed, symmetry value, kurtosis and noise peak interference value and take their numerical values, and perform formula calculation and analysis on the numerical values ​​to obtain the spectrum value.

6. The microwave sensor performance detection system based on machine learning according to claim 5 is characterized in that: The specific process of baseline drift analysis is as follows: Assume that there is an initial base value; compare the baseline value at each time point in the baseline with the initial base value. When the baseline value is greater than the initial base value, the baseline value is recorded as the drift value. Thus, several drift values ​​in the baseline can be obtained, and the interval length between two adjacent drift values ​​is calculated. The two adjacent drift values ​​are summed and then the sum is divided by the interval length to obtain the drift speed of the two adjacent drift values. The average of all drift speeds is calculated to obtain the drift average speed.

7. The microwave sensor performance detection system based on machine learning according to claim 5 is characterized in that: The specific method of determining the symmetry axis of the target peak is as follows: take the vertex of the target peak, draw a vertical line along the vertex to intersect with the horizontal axis, and the vertical line is the symmetry axis of the target peak.

8. The microwave sensor performance detection system based on machine learning according to claim 7 is characterized in that: The specific process of symmetric analysis based on the symmetry axis of the target peak is: Divide the target peak into a left half and a right half along the symmetry axis, overlap the left half and the right half along the symmetry axis, and calculate the intersection area and union area of ​​the left half and the right half in the overlapping state; The intersection area and the union area are numerically calculated and analyzed to obtain the symmetry value.

9. The microwave sensor performance detection system based on machine learning according to claim 5, characterized in that: The kurtosis is calculated based on the fitted target peak. The specific calculation process is: Retrieve the function expression f(x) after fitting the target peak, and mark the coordinate of each data point in the target peak as x j , where j = 1, 2, 3...J, J is a positive integer, J represents the total number of points that make up the target peak, and j represents the number of any point among them; all points x in the target peak are j Substitute the set formula into the following formula: The kurtosis K is calculated.

10. The microwave sensor performance detection system based on machine learning according to claim 5, characterized in that: Analyze the interference of miscellaneous peaks on the target peak. The specific process is as follows: Calculate the distance between the clutter peak and the target peak. When the distance is greater than the maximum value in the set distance interval, a highly influential clutter peak is accumulated; when the distance is within the set distance interval, a moderately influential clutter peak is accumulated; when the distance is less than the minimum value in the set distance interval, a slightly influential clutter peak is accumulated; the accumulated numbers of highly influential clutter peaks, moderately influential clutter peaks and slightly influential clutter peaks are counted respectively, and the distances corresponding to the highly influential clutter peaks, moderately influential clutter peaks and slightly influential clutter peaks are averaged to obtain the highly influential distance, moderately influential distance and slightly influential distance respectively; The cumulative number of highly influential impurity peaks, the cumulative number of moderately influential impurity peaks, the cumulative number of slightly influential impurity peaks, the highly influential distance, the moderately influential distance and the slightly influential distance are normalized and their values ​​are taken, and the values ​​are calculated and analyzed by formula to obtain the impurity peak interference value.

Images