Earthquake precursor signal anomaly detection method, processor and observation equipment

Abstract

The invention provides an earthquake precursor signal anomaly detection method, a processor and observation equipment. The detection method comprises the following steps: acquiring carbon dioxide concentration observation data and electromagnetic disturbance observation data of a detection site; preprocessing and de-noising the carbon dioxide concentration observation data, and then judging whether the carbon dioxide gas concentration is abnormal or not to obtain the probability that the carbon dioxide gas concentration is abnormal; electromagnetic disturbance observation data is preprocessed and denoised, whether electromagnetic disturbance exceeds a threshold value or not is judged, a dynamic threshold value model is established, and whether the carbon dioxide concentration is abnormal or not serves as the basis for dynamic threshold value adjustment; and counting the number of times that the electromagnetic disturbance exceeds a threshold value in the effective time window, and calculating the probability of occurrence of the electromagnetic abnormal event. The method is applied to an earthquake precursor signal monitoring process, an electromagnetic disturbance abnormal threshold value does not need to be manually adjusted, and an earthquake electromagnetic pulse abnormal event can be automatically judged. And few micro-processing resources are occupied, and the method can be operated on embedded equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of earthquake precursor observation, and particularly to a method for detecting abnormal earthquake precursor signals, a processor, and an observation device. Background Art

[0002] Seismic electromagnetic disturbance observation is a means of seismic precursor electromagnetic observation. The observation objects are electric fields and magnetic fields, and the observation method is to obtain electromagnetic pulse abnormal events related to earthquake gestation and occurrence.

[0003] The sensitivity of seismic precursor electromagnetic disturbance observation is very high. The resolution of electric field observation is better than 2 mV / km, and the resolution of magnetic field observation is better than 0.01 nT. During observation, it is often affected by various interferences, including interferences caused by thunderstorms, (moving ironware) automobiles, (crowds) student activities, etc. At some observation points, subway operation interferences have also been found.

[0004] In the seismic industry standard "Electromagnetic Disturbance Observation in the Method of Seismic Geoelectric Observation", the judgment criteria for electromagnetic anomalies are given: the long-short time window ratio method is used as the judgment criterion for the electromagnetic event threshold. The long-short time window ratio method refers to the ratio (STA / LTA) of the absolute value average of long-time observation data to the absolute value average of short-time observation data. When the ratio is greater than a certain preset threshold, it is considered that an event occurs, and when the ratio is less than a certain preset value, it is considered that the event ends.

[0005] How to distinguish real earthquake precursor electromagnetic pulse anomalies from numerous electromagnetic interference pulses is a key problem that must be solved in electromagnetic disturbance observation. In the past, researchers' experience was relied on to determine the threshold, which was extremely difficult for grass-roots earthquake observers. Usually, (2 - 5) was taken. From the perspective of earthquake reflection effect, there is a problem of high false alarm rate in electromagnetic disturbance observation, which limits the popularization and application of this observation technology. Summary of the Invention

[0006] In order to solve the above technical problems, the present invention proposes a method for detecting abnormal earthquake precursor signals, a processor, and an observation device, using the abnormal determination result of carbon dioxide gas observation as the basis for adjusting the electromagnetic disturbance observation threshold, and dynamically adjusting the threshold with TinyML to solve the subjectivity problem of manually determining the threshold.

[0007] On the one hand, a method for detecting abnormal earthquake precursor signals includes the following steps:

[0008] Obtain the carbon dioxide concentration observation data and electromagnetic disturbance observation data of the detection location;

[0009] Preprocess and denoise the carbon dioxide concentration observation data, and then judge whether the carbon dioxide gas concentration is abnormal to obtain the probability of abnormal carbon dioxide concentration;

[0010] Preprocess and denoise the electromagnetic disturbance observation data, and determine whether the electromagnetic disturbance exceeds the threshold, including establishing a dynamic threshold model and using whether the carbon dioxide concentration is abnormal as the basis for adjusting the dynamic threshold;

[0011] Statistically count the number of times the electromagnetic disturbance exceeds the threshold within the effective time window, and calculate the probability of the occurrence of electromagnetic anomaly events.

[0012] Preferably, the electromagnetic disturbance is the ratio of long and short time windows, that is,

[0013]

[0014] where N is the length of the long time window data, M is the length of the short time window data, and x(i) is the observed data value.

[0015] Preferably, the preprocessing and denoising of the carbon dioxide concentration observation data specifically include filling in missing data, removing interfering data, and eliminating the influence of temperature.

[0016] Preferably, the judgment of whether the carbon dioxide gas concentration is abnormal specifically includes,

[0017] Taking h1 hours as the time window, calculate the average carbon dioxide concentration μ in the past D1 days D1d , and check μ D1d The difference between the maximum value and the minimum value. If it is not greater than 0.5μ D1d , then the carbon dioxide concentration is of the stable type;

[0018] Perform FFT on the 24-hour data in the frequency domain and calculate the amplitude: If the amplitude is not greater than 0.2μ 24 ,

[0019] then the daily periodic carbon dioxide concentration is normal; where μ 24 is the average carbon dioxide concentration within 24 hours;

[0020] If the carbon dioxide concentration exceeds 2 times the average value μ D1d in the past D1 days, it is marked as abnormal;

[0021] Mark the carbon dioxide concentration data that does not belong to any of the stable type, daily periodic type, and abnormal type as the uncertain type.

[0022] Preferably, the preprocessing of the electromagnetic disturbance observation data specifically includes splitting the sequence of the electromagnetic disturbance observation data according to the minute window, and then standardizing each electromagnetic disturbance observation data with the global average value and standard deviation of the entire sequence.

[0023] Preferably, denoising the electromagnetic disturbance observation data specifically involves using signal feature processing to dynamically identify interference features, which specifically includes that thunderstorm interference is manifested as high-frequency transient signals, vehicle vibration interference is manifested as vibrations at specific frequencies, and the subway is manifested as periodic interference; during the period when the interference signal exists, the electromagnetic disturbance is of an uncertain type.

[0024] Preferably, the dynamic threshold model refers to the threshold of the ratio of the long and short time windows, and the threshold is adjusted through gradient descent to minimize its loss function.

[0025] More preferably, using whether the carbon dioxide concentration is abnormal as the basis for dynamic threshold adjustment specifically means that

[0026] If in the current time window, the ratio of the long and short time windows of the electromagnetic disturbance STA / LTA satisfies where θ μ is the current threshold, then the type of electromagnetic disturbance is uncertain, and it enters the next time window;

[0027] If the ratio of the long and short time windows of the electromagnetic disturbance STA / LTA > θ μ , then the electromagnetic disturbance is abnormal. At this time, if the carbon dioxide concentration is normal, then the false positive rate FPR = 100%, and the false negative rate FNR = 0%, then the threshold θ μ is adjusted, and the adjustment amplitude is: where α is the weight coefficient and η is the learning rate, is the loss gradient;

[0028] If STA / LTA > θ μ , then the electromagnetic disturbance is abnormal. If the carbon dioxide concentration is abnormal, at this time, if the false positive rate FPR = 0% and the false negative rate FNR = 0%, the threshold θ μ remains unchanged;

[0029] If STA / LTA < 0.8θ μ , then the electromagnetic disturbance is normal. If the carbon dioxide concentration is normal, at this time, if the false positive rate FPR = 0% and the false negative rate FNR = 0%, the threshold remains unchanged;

[0030] If STA / LTA < 0.8θ μ , then the electromagnetic disturbance is normal. If the carbon dioxide is abnormal, at this time, if the false positive rate FPR = 0% and the false negative rate FNR = 100%, the threshold θ μ is adjusted, and the adjustment amplitude is: where α is the weight coefficient and η is the learning rate, is the loss gradient;

[0031] The false positive rate FPR represents the proportion of misjudging normal events as abnormal; the false negative rate FNR represents the proportion of misjudging abnormal events as normal.

[0032] On the other hand, the present invention provides a processor configured to execute the earthquake precursor signal anomaly detection method described in any of the above technical solutions.

[0033] In a third aspect, the present invention provides an earthquake precursor observation device including the above-mentioned processor.

[0034] The beneficial effects of the present invention are as follows: When applied to the monitoring process of earthquake precursor signals, the present invention can automatically judge abnormal events of earthquake electromagnetic pulses without manual adjustment of the electromagnetic disturbance anomaly threshold. It occupies extremely little microprocessing resources and can run on embedded devices. For example, based on the disaster information collector's disaster situation verification and reporting device, by plugging in a carbon dioxide gas sensor, an air temperature sensor, an electric field and a magnetic field sensor, an intelligent electromagnetic disturbance observatory is formed to automatically judge abnormal events of earthquake precursor electromagnetic pulses. Detailed implementation manners

[0035] In order to enable those skilled in the art to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0036] It should be noted that the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those clearly listed steps or units, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products or devices.

[0037] This embodiment provides an earthquake precursor signal anomaly detection method, including the following steps:

[0038] S1. Obtain the carbon dioxide concentration observation data and electromagnetic disturbance observation data at the detection location. As a preferred implementation manner, the carbon dioxide concentration observation uses minute value sampling, the sampling rate of the electromagnetic disturbance observation data is 100 Hz, and the electromagnetic disturbance observation data is segmented by a minute window to obtain 6,000 samples / window.

[0039] The electromagnetic disturbance is the ratio of long and short time windows, that is

[0040]

[0041] Where N is the length of the long time window data, M is the length of the short time window data, and x(i) is the observed data value.

[0042] The ratio is compared with a preset threshold. When STA / LTA is greater than the preset threshold, the event is considered to have occurred. When the ratio is less than the preset threshold, the event is considered to have ended.

[0043] S2. Preprocessing and denoising the carbon dioxide concentration observation data, and then determining whether the carbon dioxide gas concentration is abnormal, to obtain a probability of abnormal carbon dioxide concentration.

[0044] In this embodiment, preprocessing and denoising the carbon dioxide concentration observation data includes filling in missing data, removing interfering data, and eliminating temperature effects.

[0045] Specifically, for the processing of missing data, linear interpolation can be used to fill in the missing data within each 30-minute window.

[0046] The observed data are analyzed in the time domain and frequency domain. If the data meet the pumping interference characteristics, the time period is marked as pumping interference and the data in this time period are eliminated.

[0047] For the influence of temperature, we first calculate the correlation coefficient between gas concentration and temperature. If the correlation coefficient exceeds the threshold, we then build a regression model of carbon dioxide concentration and temperature, and then determine whether to exclude it based on the data retention rules. The details are as follows:

[0048] (a) Calculate the correlation coefficient between gas concentration and temperature

[0049] Taking the window length of 7 days as an example, the Pearson correlation coefficient is calculated minute by minute:

[0050]

[0051] Among them: CO2 is the current carbon dioxide gas concentration, is the average carbon dioxide concentration over 7 days, T is the current temperature, μ T The temperature is the 7-day average.

[0052] If r≥0.54, a linear regression model CO2=a·T+b is established, where a and b are regression coefficients.

[0053] Compute the residuals:

[0054] e(t)=CO2(t)-(a·T(t)+b)

[0055] That is, the carbon dioxide concentration after excluding the influence of temperature.

[0056] (b) Data retention rule: If r < 0.54, directly use the original carbon dioxide concentration data without regression correction; when the correlation coefficient r does not exceed the threshold, retain the original data.

[0057] Furthermore, determining whether the carbon dioxide gas concentration is abnormal specifically includes:

[0058] Taking h1 hours as the time window, preferably 24 hours, calculate the average carbon dioxide concentration μ in the past D1 days D1d , where D1 can be optionally 7 days, and check μ D1d The difference between the maximum value and the minimum value. If it is not greater than 0.5μ D1d , then the carbon dioxide concentration is stable;

[0059] Perform FFT on the 24-hour data in the frequency domain and calculate the amplitude: If the amplitude is not greater than 0.2μ 24 , then the daily periodic carbon dioxide concentration is normal; where μ 24 is the average carbon dioxide concentration within 24 hours;

[0060] If the carbon dioxide concentration exceeds 2 times the average value μ of the past D1 days D1d , then it is marked as abnormal;

[0061] Mark the carbon dioxide concentration data that does not belong to any of the stable type, daily periodic type, and abnormal type as uncertain type.

[0062] S3. Preprocess and denoise the electromagnetic disturbance observation data, and determine whether the electromagnetic disturbance exceeds the threshold, including establishing a dynamic threshold model and using whether the carbon dioxide concentration is abnormal as the basis for dynamic threshold adjustment.

[0063] Preprocessing the electromagnetic disturbance observation data specifically includes: dividing the sequence of electromagnetic disturbance observation data according to a minute window, and then standardizing each electromagnetic disturbance observation data with the global average value and standard deviation of the entire sequence.

[0064] Specifically, the sampling rate of the electromagnetic disturbance observation data is preferably 100Hz, obtaining 6000 samples / window, and the standardized data where x(t) is the current observed value, μ global , σ global are the global mean and standard deviation of the entire sequence respectively.

[0065] The denoising of the electromagnetic disturbance observation data specifically involves using signal feature processing to dynamically identify interference features. Specifically, thunderstorm interference is manifested as high-frequency transient signals, vehicle vibration interference is manifested as vibrations at specific frequencies, and the subway is manifested as periodic interference. During the presence of the interference signal, the electromagnetic disturbance is of an uncertain type. Typical interference signals and determination methods are shown in Table 1.

[0066] Table 1

[0067]

[0068] The dynamic threshold model refers to the threshold of the ratio of long and short time windows, and the threshold is adjusted by gradient descent to minimize its loss function.

[0069] Ratio of long and short time windows The threshold is such that when STA / LTA is greater than the preset threshold, an event is considered to occur, and when it is less than the preset threshold, the event is considered to end. In the prior art, the preset threshold is determined by the experience of researchers, which is extremely difficult for grass-roots earthquake observers. Usually, it is taken as (1.5 - 5), which limits the popularization and application of this observation technology.

[0070] In this embodiment, a dynamic threshold adjustment algorithm based on TinyML is adopted. This is a machine learning method that allows the model to be gradually updated when new data is received without having to be retrained from scratch. Its core goal is to continuously integrate new knowledge while retaining existing knowledge, and it is applicable to scenarios where data streams arrive continuously and computing resources are limited.

[0071] The present invention uses an online gradient descent strategy to dynamically adjust the threshold of the long and short time window ratio method, and the window is set to 24 hours. Assuming the initial threshold is a certain value, based on the data of the past 24 hours, the loss function is calculated, and then the threshold is adjusted by gradient descent to minimize the loss. The observation of whether carbon dioxide is abnormal is used as the basis for adjusting the electromagnetic disturbance observation threshold. For example, if there are many electromagnetic anomaly events in the recent 24 hours, but the carbon dioxide sensor does not detect an anomaly, the threshold may need to be increased to reduce false alarms. We first give the initial judgment threshold of the long and short time window ratio method: take (1.5 - 5). Then, based on the micro machine learning TinyML, the threshold is dynamically determined.

[0072] The loss function L(θ) is the weighted error of minimizing the false positive rate FPR(θ) and the false negative rate FNR(θ)

[0073] L(θ) = α·FPR(θ) + (1 - α)FNR(θ)

[0074] Where α is the weight coefficient, and to emphasize reducing false positives, α takes the value of 0.6. The values of FPR(θ) and FNR(θ) are both 0 or 1.

[0075] Regarding the loss gradient, according to the calculation formula of the loss function L(θ), the calculation formula of the loss gradient is obtained:

[0076]

[0077] When the false alarm rate or the miss rate is 0, the corresponding term is 0, and the other term is α or (1 - α), which greatly simplifies the calculation and is suitable for program operations on a microprocessor.

[0078] The updated threshold is until the loss function is minimized, where η is the learning rate, and an exemplary value is 0.01, is the loss gradient, θ (t+1) is the updated threshold, θ (t) is the original threshold. The conditions and steps for updating the threshold are as follows:

[0079] The present invention uses whether the carbon dioxide concentration is abnormal as the basis for dynamic threshold adjustment, specifically:

[0080] If in the current time window, the ratio of the short-time window to the long-time window STA / LTA satisfies where θ μ is the current threshold, then the type of electromagnetic disturbance is uncertain, and it enters the next time window;

[0081] If the ratio of the short-time window to the long-time window of the electromagnetic disturbance STA / LTA > θ μ , then the electromagnetic disturbance is abnormal. If the carbon dioxide concentration is normal, at this time, if the false alarm rate FPR = 100%, and the miss rate FNR = 0%, then the threshold θ μ is adjusted, and the adjustment amplitude is: where,

[0082] If STA / LTA > θ μ , then the electromagnetic disturbance is abnormal. If the carbon dioxide concentration is also abnormal, then the false alarm rate FPR = 0%, and the miss rate FNR = 0%, then the threshold θ μ remains unchanged;

[0083] If STA / LTA < 0.8θ μ , then the electromagnetic disturbance is normal. If the carbon dioxide concentration is normal, then the false alarm rate FPR = 0%, and the miss rate FNR = 0%, then the threshold remains unchanged;

[0084] If STA / LTA < 0.8θ μ , then the electromagnetic disturbance is normal. If the carbon dioxide concentration is abnormal, then the false alarm rate FPR = 0%, and the miss rate FNR = 100%, then the threshold θ μ is adjusted, and the adjustment amplitude is: where α is the weight coefficient and η is the learning rate, is the loss gradient;

[0085] The false positive rate FPR represents the proportion of misjudging normal events as abnormal; the false negative rate FNR represents the proportion of misjudging abnormal events as normal.

[0086] S4. Count the number of times the electromagnetic disturbance exceeds the threshold within the effective time window, and calculate the probability of the occurrence of electromagnetic abnormal events.

[0087] It should be noted that since the electromagnetic environment is relatively vulnerable to interference, it is preferred to regard the abnormal electric field and magnetic field in the same time window as one abnormal event.

[0088] The present invention also provides a processor, which is configured to execute the earthquake precursor signal abnormal detection method described in any of the above technical solutions.

[0089] The present invention also provides an earthquake precursor observation device, including the above-mentioned processor.

[0090] The electromagnetic abnormal detection method of the present invention is applied in the process of electromagnetic signal monitoring. Without manual adjustment of the electromagnetic disturbance abnormal threshold, it can automatically judge the abnormal events of earthquake electromagnetic pulses. It occupies very little microprocessing resources and can run on embedded devices. Such a device can be a disaster information verification and reporting device carried by disaster information officers. By plugging in carbon dioxide gas sensors, temperature sensors, electric field and magnetic field sensors, it constitutes an intelligent electromagnetic disturbance observer, thereby automatically judging the abnormal events of earthquake precursor electromagnetic pulses.

[0091] The above are only the preferred embodiments of the present application and are not used to limit the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An abnormal detection method for earthquake precursor signals, characterized in that, It includes the following steps: Obtain the carbon dioxide concentration observation data and electromagnetic disturbance observation data at the detection location; Preprocess and denoise the carbon dioxide concentration observation data, and then determine whether the carbon dioxide gas concentration is abnormal, Obtain the probability of abnormal carbon dioxide concentration; Preprocess and denoise the electromagnetic disturbance observation data, and determine whether the electromagnetic disturbance exceeds the threshold, including establishing a dynamic threshold model and using whether the carbon dioxide concentration is abnormal as the basis for dynamic threshold adjustment; Count the number of times the electromagnetic disturbance exceeds the threshold within the effective time window, and calculate the probability of the occurrence of an electromagnetic anomaly event.

2. The earthquake precursor signal anomaly detection method according to claim 1, wherein The electromagnetic disturbance is the ratio of long and short time windows, that is where N is the length of the long time window data, M is the length of the short time window data, and x(i) is the observed data value.

3. The earthquake precursor signal anomaly detection method according to claim 1, wherein The preprocessing and denoising of the carbon dioxide concentration observation data specifically include filling in missing data, removing interference data, and eliminating the influence of temperature.

4. The earthquake precursor signal anomaly detection method according to claim 1, characterized in that The determination of whether the carbon dioxide gas concentration is abnormal specifically includes Calculate the average concentration μ of carbon dioxide over the past D1 days with a time window of h1 hours D1d , and check μ D1d for the difference between the maximum and minimum values. If it is not greater than 0.5μ D1d , then the carbon dioxide concentration is stable; Perform FFT on 24-hour data in the frequency domain and calculate the amplitude: If the amplitude is not greater than 0.2 μ 24 , Then the daily periodic carbon dioxide concentration is normal; where μ 24 is the average carbon dioxide concentration within 24 hours; If the carbon dioxide concentration exceeds 2 times the average value μ of the past D1 days D1d it is marked as abnormal; Marking the carbon dioxide concentration data that does not belong to any of the stable type, daily periodic type, and abnormal type as the uncertain type.

5. The earthquake precursor signal anomaly detection method according to claim 1, characterized in that, The preprocessing of the electromagnetic disturbance observation data specifically includes dividing the sequence of the electromagnetic disturbance observation data according to the minute window, and then normalizing each electromagnetic disturbance observation data with the global average and standard deviation of the entire sequence.

6. The earthquake precursor signal anomaly detection method according to claim 1, wherein The denoising of the electromagnetic disturbance observation data is specifically to use signal feature processing to dynamically identify interference features, specifically including that thunderstorm interference is manifested as high-frequency transient signals, vehicle vibration interference is manifested as vibrations at specific frequencies, and the subway is manifested as periodic interference; during the existence period of the interference signal, the electromagnetic disturbance is of the uncertain type.

7. The method for detecting abnormal seismic precursor signals according to claim 2, wherein The dynamic threshold model refers to the threshold of the ratio of long and short time windows, and the threshold is adjusted by gradient descent to minimize its loss function.

8. The method for detecting abnormal seismic precursor signals according to claim 7, wherein Using whether the carbon dioxide concentration is abnormal as the basis for dynamic threshold adjustment specifically means If the ratio of the short-time window to the long-time window of the electromagnetic disturbance, STA / LTA, satisfies in the current time window where θ μ is the current threshold, then the type of electromagnetic disturbance is uncertain and it enters the next time window; If the ratio of short-time window to long-time window of electromagnetic disturbance STA / LTA > θ μ , then the electromagnetic disturbance is abnormal. At this time, if the carbon dioxide concentration is normal, the false positive rate FPR = 100%, and the false negative rate FNR = 0%. The threshold θ should be adjusted μ , and the adjustment range is: where α is the weight coefficient and η is the learning rate, is the loss gradient; If STA / LTA > θ μ , then the electromagnetic disturbance is abnormal. At this time, if the carbon dioxide concentration is abnormal, the false positive rate FPR = 0%. False negative rate FNR = 0%, threshold θ μ Remain unchanged; If STA / LTA < 0.8θ μ , then the electromagnetic disturbance is normal. If the carbon dioxide concentration is also normal, the false positive rate FPR = 0%, When the false negative rate FNR = 0%, the threshold remains unchanged; If STA / LTA < 0.8θ μ , the electromagnetic disturbance is normal. If the carbon dioxide concentration is abnormal, the false positive rate FPR = 0%, the false negative rate FNR = 100%, and the threshold θ is adjusted μ , and the adjustment amplitude is as follows: where α is the weight coefficient and η is the learning rate is the loss gradient; The false positive rate FPR represents the proportion of misjudging normal events as abnormal; the false negative rate FNR represents the proportion of misjudging abnormal events as normal.

9. A processor, characterized in that, The processor is configured to execute the seismic precursor signal anomaly detection method according to any one of claims 1 to 8.

10. An earthquake precursor observation device, characterized in that, It includes the processor according to claim 9.

Images