A real-time detection method for geological disaster hazards of power infrastructure

Abstract

The present invention relates to a real-time detection method for geological disaster hazards of power infrastructure, comprising the following steps: S1. Obtain different types of monitoring data in the geological disaster monitoring area; separately construct initial time series data of univariate time series for any type of monitoring data; S2. Construct a residual sequence according to the difference between each data element and the average value of the pre-set window; S3. Solve two model parameters; S4. Apply the second theoretical model of extreme value theory to calculate the latest upper and lower residual thresholds to determine the normal value range; S5. Calculate the residual value of the latest data; S6. Conduct anomaly determination on the residual value. By analyzing the data patterns of each monitoring type through unsupervised learning and having the adaptive ability to adjust the detection model in real time, it can follow the data fluctuation trend to adjust the latest upper limit threshold of the data, ensuring that it can adapt to various types of monitoring data.

Description

Technical Field

[0001] The present invention relates to the technical field of real-time monitoring of geological disaster hazards, and in particular to a method for real-time detection of geological disaster hazards of power infrastructure. Background Art

[0002] In the regions of Yunnan, Guizhou, and Sichuan in China, there are rich power generation resources, especially the widespread distribution of clean energy such as hydropower and wind power; therefore, there are numerous power infrastructures (such as converter stations, step-up substations, etc.); however, due to the complex terrain features of this region, power facilities are often located in mountainous or hilly areas and are long-term faced with the threat of geological disasters, such as slope collapses, landslides, debris flows, etc. These geological disasters not only damage power equipment but may also lead to the paralysis of the power system, causing large-scale power outages and economic losses; therefore, how to quickly and effectively monitor and warn of geological disaster hazards of power infrastructure has become one of the key issues in the operation and maintenance of power systems.

[0003] Traditional geological disaster monitoring means mainly rely on fixed sensors and manual inspections, which have problems of limited monitoring range and slow response speed and cannot capture the early signals of sudden disasters in a timely manner; in addition, there are many elements of geological disaster hazards such as slope collapses, landslides, and debris flows. Although different types of sensors can be arranged according to different monitoring objects, it is still a great challenge to effectively identify potential risks from various monitoring data in real time; in recent years, machine learning and deep learning technologies have gradually been applied to the detection and warning of geological disaster hazards and can discover abnormal data beyond the normal range; although such methods can improve the detection accuracy through data-driven, they rely on a large amount of labeled data for training and need to re-adjust and train the model in different application scenarios, resulting in limitations in the adaptability and response speed of the system. Especially in a complex geological environment like Yunnan, Guizhou, and Sichuan that requires quick response, it shows certain limitations in the real-time response of geological disaster hazard monitoring. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method for real-time detection of geological disaster hazards of power infrastructure, which gets rid of the dependence on labeled data, analyzes the data patterns of each monitoring type through unsupervised learning, and has the adaptive ability to adjust the detection model in real time, and can follow the data fluctuation trend to adjust the latest upper limit threshold of the data to ensure that it can adapt to various types of monitoring data.

[0005] The technical solution adopted by the present invention to solve its technical problems is as follows:

[0006] A method for real-time detection of geological disaster hazards of power infrastructure includes the following steps:

[0007] S1. Construct initial time series data and obtain different types of monitoring data in the geological disaster monitoring area; construct initial time series data for any one type of monitoring data separately;

[0008] S2. Construct a residual sequence. Set an initial window according to the sequence size of the initial time-series data. The length L of the initial window Mi is set to 1 / 5 of the initial sequence length. The initial window Mi slides backward on the initial sequence. Starting from the position of L + 1 in the initial sequence, construct a residual sequence S based on the difference between each data element Xi and the average value of the adjacent previous window Mi. Determine the upper limit threshold of the high position and the lower limit threshold of the low position through the sorting of the residual sequence S. The elements in the residual sequence S that exceed the upper limit threshold form the upper difference overrun set, and the elements in the residual sequence that exceed the lower limit threshold form the lower residual overrun set. Each element Yi in the overrun set is the difference between the data element Xi in the initial sequence and the average value of the initial window Mi.

[0009] S3. Fit the upper difference overrun set and the lower residual overrun set to the Generalized Pareto Distribution (GPD) model, and solve the two model parameters respectively.

[0010] S4. According to the model parameters and the initial time-series data, apply the second theoretical model of extreme value theory to calculate the latest upper residual threshold and lower residual threshold to determine the normal value range.

[0011] S5. When new data is collected and added to the time series, perform a smoothing operation on the latest data element Xi, that is, calculate the exponentially weighted moving average of the observed values in the past period of time through the exponentially weighted moving average method to remove the influence of normal data fluctuations. The difference between the latest data element Xi and the weighted average value forms the latest residual value.

[0012] S6. Perform an anomaly determination on the residual value. If it exceeds the upper or lower residual threshold, it is determined that the latest data deviates from the normal range and is regarded as abnormal data, and a warning message is output. Compare the residual value with the initial upper residual threshold, lower residual threshold, and the latest upper residual threshold and lower residual threshold to decide whether to update the latest upper residual threshold and lower residual threshold.

[0013] In step S1, the monitoring data includes the time-series data of rainfall, impact force, concrete pressure, soil displacement, and deep displacement.

[0014] In step S3, when fitting the upper difference overrun set and the lower residual overrun set to the Generalized Pareto Distribution (GPD) model, solve the model parameters through the composite parameter estimation method as follows:

[0015] Preliminarily calculate the distribution model parameter values through the Probability Weighted Moments (PWM) method, and then calculate the model distribution parameter values using the Method of Moments (MOM). Calculate the mean value of the two parameter values to obtain the latest distribution model parameter values.

[0016] In step S3, the steps of respectively solving the two model parameters include the following:

[0017] S3.1. According to the principle of probability weighted moment, the k-th order probability weighted moment of the GPD model is expressed as follows:

[0018]

[0019] In Equation (1), w0 is the first-order probability weighted moment; w1 is the second-order probability weighted moment; N t is the length of the overrun set; E(Y) is the variance of the overrun set; Y(i) is the i-th data after sorting the time series data;

[0020] S3.2. According to the principle of estimating the population from samples, calculate the GPD model parameters by the PWM method. The formula is as follows:

[0021]

[0022] S3.3. Calculate the GPD model parameters by the MOM method. The formula is as follows:

[0023]

[0024] In Equation (3), μ is the mean of the overrun set; S 2 is the variance of the overrun set;

[0025] S3.4. The σ and γ parameters calculated by the composite parameter estimation method are as follows:

[0026]

[0027] In Equation (4), and are the estimated values of the parameters to be solved in the GPD model.

[0028] In step S4, calculate the latest upper residual threshold and lower residual threshold. The formula is as follows:

[0029]

[0030] In Equation (5), N t is the length of the overrun set; q is the empirical value of the risk parameter for determining anomalies, that is, the probability of anomalies occurring; n is the length of the current univariate time series.

[0031] In step S5, calculate the exponentially weighted moving average of the observed values in the past period. The formula is as follows:

[0032] EMA t = α × y t + (1 - α) × EMA t-1 (6)

[0033] In formula (6), EMAt is the latest exponentially weighted moving average to be calculated, yt is the observed value at the t-th time point, EMAt-1 is the EMA value at the previous time point, and α is the smoothing factor, which represents the weight of the current observed value in the weighted average and is set according to the characteristics of data fluctuations. The EMA value of the first element yt in the time series data is equal to itself.

[0034] The beneficial effects of the present invention are as follows:

[0035] 1. It does not rely on labeled data. By unsupervised learning, it analyzes the data patterns of each monitoring type, does not require pre-training, and has the ability to adaptively adjust the detection model in real time. It can follow the data fluctuation trend to adjust the latest upper limit threshold of the data, ensuring that it can adapt to various types of monitoring data.

[0036] 2. There is no complex calculation in the detection method, enabling each detection device to have the ability of edge real-time detection. It can quickly and effectively detect abnormal data in the geological disaster monitoring data, and flexibly adjust the relevant parameters of the detection method according to the actual situation, further ensuring the practicability and adaptability of the system, improving the real-time detection accuracy and efficiency of geological disaster hazards of power infrastructure, and ensuring the stable and safe operation of power infrastructure.

[0037] 3. By accessing and maintaining each univariate time series data through device channels, different types of data are directly isolated from each other, reducing the complexity of data processing. Description of the Drawings

[0038] Figure 1 It is a schematic flow chart of the present invention. Detailed Embodiments

[0039] The present invention will be further described below with reference to the drawings and embodiments.

[0040] Embodiment 1

[0041] As Figure 1 shown, a real-time detection method for geological disaster hazards of power infrastructure includes the following steps:

[0042] S1. Construct initial time series data and obtain different types of monitoring data in the geological disaster monitoring area; construct initial time series data for any one type of monitoring data separately;

[0043] S2. Construct a residual sequence. Set an initial window according to the sequence size of the initial time series data. The length L of the initial window Mi is set to 1 / 5 of the initial sequence length. The initial window Mi slides backward on the initial sequence. Starting from the position of L + 1 in the initial sequence, construct a residual sequence S based on the difference between each data element Xi and the average value of the previous adjacent window Mi. Determine the upper limit threshold at the high position and the lower limit threshold at the low position through the sorting of the residual sequence S. The elements in the residual sequence S that exceed the upper limit threshold form the upper difference overrun set, and the elements in the residual sequence that exceed the lower limit threshold form the lower residual overrun set. Each element Yi in the overrun set is the difference between the data element Xi in the initial sequence and the average value of the initial window Mi.

[0044] The initial time series data is univariate time series.

[0045] The sorting of the residual sequence S determines the upper limit threshold at the high position and the lower limit threshold at the low position according to the empirical quantile (i.e., the proportion of abnormal data determined according to prior knowledge).

[0046] S3. Fit the upper difference overrun set and the lower residual overrun set to the Generalized Pareto Distribution (GPD) model, and solve the two model parameters respectively.

[0047] S4. According to the model parameters and the initial time series data, apply the second theoretical model of extreme value theory to calculate the latest upper residual threshold and the lower residual threshold to determine the normal value range.

[0048] S5. When new data is collected and added to the time series, perform a smoothing operation on the latest data element Xi, that is, calculate the exponentially weighted moving average of the observed values in the past period of time through the exponentially weighted moving average method to remove the influence of normal data fluctuations. The difference between the latest data element Xi and the weighted average value forms the latest residual value.

[0049] S6. Perform an anomaly determination on the residual value. If it exceeds the upper or lower residual threshold, it is determined that the latest data deviates from the normal range and is regarded as abnormal data, and a warning message is output. Compare the residual value with the initial upper residual threshold, the lower residual threshold, the latest upper residual threshold, and the latest lower residual threshold to decide whether to update the latest upper residual threshold and the latest lower residual threshold.

[0050] In step S1, the monitoring data includes the time series data of rainfall, impact force, concrete pressure, soil displacement, and deep displacement.

[0051] In step S3, when fitting the upper difference overrun set and the lower residual overrun set to the Generalized Pareto Distribution (GPD) model, solve the model parameters through the composite parameter estimation method as follows:

[0052] The parameter values of the distribution model are initially calculated by the Probability Weighted Moments (PWM) method, and then the parameter values of the model distribution are calculated using the Moment Estimation Method (MOM) with extremely fast calculation speed. The mean value of the two parameter values is calculated to obtain the latest parameter values of the distribution model.

[0053] In step S3, the solving of the two model parameters respectively includes the following steps:

[0054] S3.1. According to the principle of probability weighted moments, the k-th order probability weighted moment of the GPD model is expressed as follows:

[0055]

[0056] In Equation (1), w0 is the first-order probability weighted moment; w1 is the second-order probability weighted moment; N t is the length of the exceedance set; E(Y) is the variance of the exceedance set; Y(i) is the i-th data after sorting the time series data;

[0057] S3.2. According to the principle of inferring the population from the sample, the GPD model parameters are calculated by the PWM method, and the formula is as follows:

[0058]

[0059] S3.3. The GPD model parameters are calculated by the MOM method, and the formula is as follows:

[0060]

[0061] In Equation (3), μ is the mean of the exceedance set; S 2 is the variance of the exceedance set;

[0062] S3.4. The σ and γ parameters calculated by the composite parameter estimation method, and the formula is as follows:

[0063]

[0064] In Equation (4), and are the estimated values of the parameters to be solved in the GPD model.

[0065] In step S4, the calculation of the latest upper residual threshold and lower residual threshold is as follows:

[0066]

[0067] In Equation (5), N t is the length of the exceedance set; q is the empirical value of the risk parameter for determining anomalies, that is, the probability of anomalies occurring; n is the length of the current univariate time series.

[0068] By using this method, the upper and lower limit residual thresholds can be calculated respectively to determine the normal value range.

[0069] In step S5, the exponentially weighted moving average of the observed values over a past period is calculated, and the formula is as follows:

[0070] EMA t = α × y t + (1 - α) × EMA t-1 (Six)

[0071] In formula (Six), EMAt is the latest exponentially weighted moving average to be calculated, yt is the observed value at the t-th time point, EMAt-1 is the EMA value at the previous time point, and α is the smoothing factor, which represents the weight of the current observed value in the weighted average and is set according to the characteristics of data fluctuations. The EMA value of the first element yt in the time series data is equal to itself.

[0072] In step S6, the anomaly determination does not directly use the latest data element Xi, but uses the residual value Xi' calculated in step S5 for determination. If Xi' is greater than the latest upper limit residual threshold or less than the lower limit residual threshold, it is considered that the latest data Xi exceeds the normal range and is determined as abnormal data. If the latest residual value Xi' is between the initial upper limit residual threshold and the latest upper limit residual threshold, then Xi' is added to the upper limit exceedance set. Similarly, if Xi' is between the initial lower limit residual threshold and the latest lower limit residual threshold, it is added to the lower limit exceedance set. When a new element is added to the exceedance set, the latest residual threshold can be updated through steps S3 and S4.

[0073] A real-time detection device for geological disaster hazards of power infrastructure, comprising:

[0074] A data acquisition module, configured with multiple data access channels, each channel can access a sensor terminal to collect a type of detection data.

[0075] A data conversion module, configured with a processor and a memory, for converting the analog signals collected by the sensor terminal into numerical signals. Each channel can be configured with an independent conversion rule to convert the collected raw data and store it. Each channel maintains a time series, that is, the data collected by each channel constitutes a univariate time series data.

[0076] An analysis and processing module, configured with a memory and a processor. The processor runs the detection method to detect the monitoring data collected by each channel in real time. Using the univariate time series data composed of each type of data as the processing object, when new data arrives, it detects in real time whether the data is abnormal.

[0077] A configuration module, for receiving and storing the original conversion rules from system configuration and the relevant parameter values required for the real-time detection method.

[0078] A real-time detection system for geological disaster hazards of power infrastructure, the system comprising:

[0079] A data storage module, configured to receive and store the output data from the real-time detection devices for geological disaster hazards of multiple power infrastructures.

[0080] A data display module, configured to display the alarm situation, the original data curves of each sensor, and the upper and lower limit threshold situations.

[0081] A system configuration module, configured to configure the calculation rules for converting the sensor analog signals into numerical signals, and allow users to adjust parameters such as the empirical anomaly probability and the smoothing factor according to actual detection requirements.

[0082] An output module, which outputs the collected original data, the upper and lower limit residual thresholds dynamically updated during the anomaly determination process, process data such as the exponentially weighted moving average, and the anomaly detection results.

[0083] Without relying on labeled data, analyze the data patterns of each monitoring type through unsupervised learning, without the need for prior training, and have the ability to adaptively adjust the detection model in real time, and can adjust the latest upper limit threshold of the data following the data fluctuation trend, ensuring that it can adapt to various types of monitoring data.

[0084] There is no complex calculation in the detection method, enabling each detection device to have the ability of edge real-time detection, quickly and effectively discovering abnormal data in the geological disaster monitoring data, flexibly adjusting the relevant parameters of the detection method according to the actual situation, further ensuring the practicability and adaptability of the system, and improving the accuracy and efficiency of the real-time detection of geological disaster hazards of power infrastructure to ensure the stable and safe operation of power infrastructure.

[0085] Access and maintain each univariate time series data through the device channels, so that different types of data are directly isolated from each other, reducing the complexity of data processing.

[0086] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A real-time detection method for geological disaster hazards of power infrastructure, characterized in that: It includes the following steps: S1. Construct initial time series data and obtain monitoring data of different types in the geological disaster monitoring area; construct initial time series data for any type of monitoring data separately. S2. Construct a residual sequence. Set an initial window according to the sequence size of the initial time series data. The length L of the initial window Mi is set to 1 / 5 of the initial sequence length. The initial window Mi slides backward on the initial sequence. Starting from the position of L + 1 of the data elements in the initial sequence, construct a residual sequence S according to the difference between each data element Xi and the average value of the adjacent previous window Mi. Determine the upper limit threshold of the high position and the lower limit threshold of the low position through the sorting of the residual sequence S. The elements in the residual sequence S that exceed the upper limit threshold form an upper difference overrun set, and the elements in the residual sequence that exceed the lower limit threshold form a lower residual overrun set. Each element Yi in the overrun set is the difference between the data element Xi in the initial sequence and the average value of the initial window Mi. S3. Fit the upper difference overrun set and the lower residual overrun set to the Generalized Pareto Distribution (GPD) model, and solve the two model parameters respectively. S4. According to the model parameters and the initial time series data, apply the second theoretical model of extreme value theory to calculate the latest upper residual threshold and lower residual threshold to determine the normal value range. S5. When new data is collected and added to the time series, perform a smoothing operation on the latest data element Xi, that is, calculate the exponentially weighted moving average of the observed values in the past period of time through the exponentially weighted moving average method to remove the influence of normal data fluctuations. The difference between the latest data element Xi and the weighted average value forms the latest residual value. S6. Perform an anomaly determination on the residual value. If it exceeds the upper or lower residual threshold, it is determined that the latest data deviates from the normal range and is regarded as abnormal data, and a warning message is output. Compare the residual value with the initial upper residual threshold, lower residual threshold, and the latest upper residual threshold and lower residual threshold to determine whether to update the latest upper residual threshold and lower residual threshold. In step S1, the monitoring data includes time series data of rainfall, impact force, concrete pressure, soil displacement, and deep displacement.

2. The real-time detection method for geological disaster hazards of a power infrastructure according to claim 1, characterized in that: In step S3, when fitting the upper difference overrun set and the lower residual overrun set to the Generalized Pareto Distribution (GPD) model, solve the model parameters through the composite parameter estimation method as follows: Preliminarily calculate the distribution model parameter values through the Probability Weighted Moments (PWM) method, and then calculate the model distribution parameter values using the Method of Moments (MOM). Calculate the mean value of the two parameter values to obtain the latest distribution model parameter values.

3. The real-time detection method for geological disaster hazards of a power infrastructure according to claim 1, characterized in that: In step S3, the steps for separately solving the two model parameters include the following: S3.

1. According to the principle of probability weighted moments, the k - th probability weighted moment of the GPD model is expressed as follows: Equation (1), where w0 is the first-order probability weighted moment; w1 is the second-order probability weighted moment; N t is the length of the overrun set; E(Y) is the variance of the overrun set; Y(i) is the i-th data after sorting the time series data; S3.

2. According to the principle of inferring the population from the sample, calculate the GPD model parameters through the PWM method, and the formula is as follows: S3.

3. Calculate the GPD model parameters through the MOM method, and the formula is as follows: In formula (III), μ is the mean of the transfinite set; S 2 is the variance of the transfinite set; S3.

4. Calculate the σ and γ parameters through the composite parameter estimation method, and the formula is as follows: In Equation (4), and are the estimated values of the parameters to be solved in the GPD model.

4. The real-time detection method for geological disaster hazards of a power infrastructure according to claim 1, characterized in that: In step S4, the formulas for calculating the latest upper residual threshold and lower residual threshold are as follows: In formula (V), N t is the length of the over-limit set; q is the empirical value of the risk parameter for determining anomalies, that is, the probability of anomaly occurrence; n is the length of the current univariate time series.

5. The real-time detection method for geological disaster hazards of a power infrastructure according to claim 1, characterized in that: In step S5, the calculation of the exponentially weighted moving average of the observed values over a past period is performed according to the following formula: EMA t = α × y t + (1 - α) × EMA t-1 (Six) In Equation (6), EMAt is the most recent exponentially weighted moving average to be calculated, yt is the observed value at the t-th time point, EMAt-1 is the EMA value at the previous time point, and α is the smoothing factor, which represents the weight of the current observed value in the weighted average and is set according to the characteristics of data fluctuations. The EMA value of the first element yt in the time series data is equal to itself.

Images