Electric power 5G network electromagnetic space security situation awareness and evaluation method for novel electric power system access

Abstract

The invention provides a novel power system access-oriented electric power 5G network electromagnetic space security situation awareness and evaluation method, which is characterized by comprising the following steps of S1, analyzing an electric power 5G network electromagnetic security data acquisition method based on network measurement and an external electromagnetic interference instrument deep learning adaptability; s2, long-short-term memory electric power 5G network electromagnetic space security situation awareness based on a dynamic baseline; and S3, electric power 5G network electromagnetic space security evaluation based on geographic space visualization of local abnormal factors. The base station-user two-way interference perception and visual evaluation are realized through deep learning cooperation degree between network measurement and spectrum acquisition, long and short-term memory prediction, and user terminal RSRP and RSRQ data analysis local abnormal factors in an interference existence time period.

Description

Technical Field

[0001] The present invention relates to electromagnetic space security situation awareness, and in particular to electromagnetic space security situation awareness and an evaluation method for an electric power 5G network accessed with a new type of electric power system. Background Art

[0002] Accurately and comprehensively extracting security situation elements in the network is the basis of network security situation awareness research. However, since the network has developed into a large nonlinear complex system with strong flexibility, the extraction of network security situation elements is very difficult.

[0003] The purpose of 5G network electromagnetic space security situation awareness and assessment is to quickly determine whether there is an interference source on the wireless side in response to the access of illegal devices, and to provide a basis for further proposing wireless security control strategies. The current data acquisition methods required for perception include network measurement and instrument measurement, but there are many types and it is impossible to evaluate the efficiency and quality of data acquisition.

[0004] In terms of situational awareness, according to the frequency bands currently used by the power 5G network, the range of spectrum real-time monitoring is from 200M-2GHz. In the future, it will be necessary to scan wider frequency bands, possibly up to 3GHz, or even higher. Detecting such a wide frequency will generate a large amount of raw detection data. If the detection node does not do enough data processing, it will put a lot of pressure on the backhaul network. In addition, errors or anomalies in the original detection data itself will seriously affect the subsequent spectrum situation construction and the accuracy and effectiveness of data mining.

[0005] In terms of safety assessment, the electromagnetic environment is composed of many factors, which can be divided into natural interference electromagnetic environment and man-made interference electromagnetic environment. The former is mainly composed of factors such as lightning, static electricity and cosmic radiation; the latter has more factors. According to the various components of the electromagnetic environment effect, it is necessary to construct the current level of electromagnetic interference to determine its safety level. After obtaining the results of interference identification and risk areas, it is necessary to combine the current interference type, interference source parameters and other information to construct a graded electromagnetic space security situation indicator. Summary of the invention

[0006] Based on the above problems, the present invention will consider the following aspects: first, construct a comprehensive data acquisition method, and obtain the degree of coordination required for network measurement and instrument measurement through deep learning; second, taking into account the regularity and stability of the transmission data used in the power system (because users' electricity usage habits generally do not change much), it is possible to use dual long- and short-term memory future predictions in different bands and time periods to consider whether there is data in any band or time period that is significantly different from the prediction, thereby determining interference; third, dynamically adjust the data baseline in conjunction with long- and short-term memory; fourth, obtain information on abnormal distribution through geographic spatial visualization of local abnormal factors.

[0007] In view of the above considerations, the present invention provides a method for electromagnetic space security situation awareness and assessment of a power 5G network for access to a new power system, including the following steps: S1 Analysis of the method for acquiring electromagnetic safety data of the power 5G network based on network measurement and deep learning coordination of external electromagnetic interference instrument; S2 Long-short-term memory power 5G network electromagnetic space security situation awareness based on dynamic baseline; S3 Electromagnetic space security assessment of power 5G network based on geospatial visualization of local anomaly factors.

[0008] Wherein, S1 includes the following steps: S1-1 Select at least one network measurement data collection source and use an electromagnetic interference meter based on power port electromagnetic interference to obtain the spectrum; S1-2 uses a simulated interference source to send multiple groups of preset known simulated interference signals, and fixedly transmits simulated non-interference signals in the network (which can be arbitrarily cut off in the real historical power data for simulation). The simulated interference signals are classified into true interference and false interference. The true interference includes a set of predetermined standard interference signals and multiple groups of true fluctuation groups based on the standard interference signals. The false interference signals include multiple groups of interference signals whose differences in signal strength and signal frequency distribution from the standard interference signals and the true fluctuation groups are all outside the fluctuation threshold. S1-3 S1-1 and S1-2 collect data and spectrum respectively, build deep learning models respectively, and train the collected data and spectrum respectively to obtain the accuracy that can be achieved when the training of both is completed and , then define the degree of fit between the two .

[0009] S2 includes the following steps: S2-1 obtains the data to be analyzed through the network management system and the terminal, determines whether there is interference through signal-to-noise ratio analysis and baseline, and dynamically adjusts the power on the base station side and the baseline; S2-2 marks the frequency spectrum after power adjustment according to the dynamically adjusted baseline, giving the band of interest in the spectrum to match the degree The band of interest is divided into three parts, the middle part and the low-frequency and high-frequency parts. The former accounts for a proportion of the entire band of interest. , and the ratio of the low-frequency band range to the high-frequency band range is .

[0010] It is easy to understand that since the data obtained by network measurement and the data obtained by spectrum are different in degree of closeness to the real data, the accuracy of deep learning of the two is different. Therefore, the degree of acceptance is also different. Therefore, when analyzing the baseline and spectrum, we only use the degree of coordination to select the dynamic baseline ratio range used in the band of interest instead of the entire range.

[0011] S2-3 divides the middle part into multiple equally divided band intervals according to a preset step size, establishes a long short-term memory model with each band interval as a unit, inputs the corresponding training set spectrum of the corresponding band interval above the dynamic baseline at the input end of each unit to train the unit, and outputs a prediction of whether interference exists at the output end, and the units are connected as a transmission layer of the unit status.

[0012] S3 includes the following steps: S3-1 For the band interval predicted to be interfered, find the corresponding time period based on the timestamp of the analyzed spectrum through inverse Fourier transform, thereby obtaining multiple time periods corresponding to the band intervals predicted to be interfered according to the spectra at multiple different detection timestamps, S3-2 calls the RSRP and RSRQ data of the corresponding user side in multiple time periods, and compares them with the preset normal RSRP and RSRQ data respectively. The difference is used as the local abnormality factor. If the local abnormality factor exceeds 5-10%, it is the object of concern; if it exceeds 10%-exceeding 20%, it is suspected interference; if it exceeds 20%, it is an interference reference, and it is displayed in pseudo-color on the base station-user geographic digital map.

[0013] Optionally, for both the low-frequency and high-frequency parts, the absolute intensity of the spectrum is considered, and the time-integrated average value of multiple intensities within a synchronous long range in multiple historical data is calculated.

[0014] Optionally, the corresponding time period is 5 minutes to 3 hours from the timestamp.

[0015] Optionally, at least one network measurement includes any one or a combination of security data collection based on an LTE OAM system or a network management system, security data collection based on a drive test (DT), and measurement report (MeasureReport) data based on statistical granularity, wherein LTE OAM operates at the data link layer and reports the link status through timed interaction of OAM PDU between devices. The DT (Drive Test) drive test uses drive test equipment combined with geographic information to perform comprehensive measurements of the air interface to verify the parameters of the wireless interface, analyze, optimize and locate faults; a measurement report (MeasureReport, MR) refers to information sent every 480ms on a service channel (470ms on a signaling channel), and mainly includes two parts: uplink signal information and downlink signal information.

[0016] Optionally, deep learning uses a convolutional neural network to form the collected data into binary digital 1 black and digital 2 white pixels to form a two-dimensional image, and the frequency spectrum is used as an image and input into two convolutional neural networks for training.

[0017] Beneficial effect: Through the coordination of deep learning between multiple network measurements and spectrum acquisition, a long-short term memory prediction based on a dynamic baseline is given, and an interference existence time period based on a timestamp of an inverse Fourier transform is performed in the predicted interference band area, so that the user terminal RSRP and RSRQ data are called based on this time period to analyze the local abnormal factors, and the base station-user two-way interference perception and visual evaluation are realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Flow chart of step S1 in Example 1 of the present invention.

[0019] Figure 2 Flow chart of step S1 in Example 2 of the present invention.

[0020] Figure 3 Flow chart of step S1 in Example 3 of the present invention. DETAILED DESCRIPTION

[0021] This embodiment provides a method for electromagnetic space security situation awareness and assessment of a power 5G network for access to a new power system, including the following steps: S1 Analysis of the method for acquiring electromagnetic safety data of the power 5G network based on network measurement and deep learning coordination of an external electromagnetic interference meter.

[0022] This step is the data basis for the electromagnetic space security situation awareness, so it is necessary to specifically analyze the accuracy of the signals in the network and the interference on the power supply for the real data. Through deep learning, we can get the different recognition accuracy of data acquisition when normal signals and interference signals are mixed, so as to get the confidence level of the collected data actually used. It is characterized by the degree of coordination, that is, the credibility of the data measured by the network occupies the range of the entire signal band, and the middle part of the frequency is selected to meet the law that the interference signal with a high probability presents a normal distribution.

[0023] Since the data measured by the network is affected by the power adjustment and dynamic baseline adjustment in the subsequent step S2, the spectrum intensity is also affected. This provides a theoretical basis for using dynamic baselines to mark the spectrum and subsequent situational awareness.

[0024] S2 Long-short-term memory power 5G network electromagnetic space security situation awareness based on dynamic baseline.

[0025] This step dynamically adjusts the power and baseline of the base station side according to the signal-to-noise ratio analysis and baseline determination of the base station network management system and the user terminal. This is a general means of controlling interference that is usually used. Although it can control interference with a certain probability, it cannot be accurately perceived. When there are fluctuations caused by unknown factors, unnecessary interference control may be performed, wasting system operation resources. And it only provides an interference baseline.

[0026] However, we still believe in the baseline and mark the baseline on the adjusted spectrum, and use this baseline to obtain the long-term and short-term memory prediction of the interference above the baseline. This is just an intelligent prediction, which is completed by training multiple historical adjusted spectrum data. It is also necessary to use the inverse Fourier transform and the timestamp to find the time period of the predicted interference band, and finally combine the user-side RSRP and RSRQ data in the time period to give a more meaningful interference perception. Therefore, it belongs to the base station-user two-way perception strategy.

[0027] S3 Electromagnetic space security assessment of power 5G network based on geospatial visualization of local anomaly factors.

[0028] This step is based on step S2, that is, obtaining the time period based on the timestamp after the inverse Fourier transform of the interference band predicted by the long short-term memory, and collecting the RSRP and RSRQ data of the user terminal signal based on this time period, so as to analyze the local abnormal factors, and thus visualize the distribution of the pseudo-color interference assessment level according to the digital map.

[0029] Specifically, Figure 1The process of step S1 is given. First, three measurement methods are used, namely, network management system (mainly for signal transceiver), DT (for air interface), and measurement report based on statistical granularity (for channel), to conduct more comprehensive security data collection, and electromagnetic interference meter is used to collect the spectrum of the power port.

[0030] Another preferred implementation manner may also add a fourth type of network measurement method, namely, LTEOAM based on the link layer, so that interference data collection is also performed on the link layer. The number of types of measurements may actually be selected based on acceptable costs and analysis of situations where interference failures often occur.

[0031] Then, the collected binary data is converted into black and white two-dimensional images, and the spectrum is used as an image to build different convolutional neural network deep learning models for deep learning. The interference scene in which the learning is located is generated by signal simulation.

[0032] Specifically, a simulated interference source is used to send multiple sets of preset known simulated interference signals, and a simulated non-interference signal is fixedly transmitted in the network (which can be arbitrarily cut off from the real historical power data to simulate). The simulated interference signal is classified into true interference and false interference. The true interference includes a set of predetermined standard interference signals and multiple sets of true fluctuation groups based on the standard interference signals ( Figure 1 , which shows part of the signal time domain diagram as an example), false interference signals generally include noise caused by the natural environment or the surrounding power frequency, and include multiple groups of interference signals whose differences in signal strength and signal frequency distribution with the standard interference signals and the true fluctuation groups are all outside the fluctuation threshold.

[0033] Collect data and spectrum separately, build deep learning models separately, and train the collected data and spectrum separately to obtain the accuracy that can be achieved when the training of both is completed. and , then in this embodiment, the result is that the latter is greater in historical data, so the degree of fit between the two is defined as .

[0034] Next, step S2 will be described, which includes the following steps: like Figure 2 As shown, S2-1 first obtains the data to be analyzed in the network through the network management system and the terminal, and determines whether there is interference through signal-to-noise ratio analysis and baseline, that is, empirical qualitative analysis is performed in the scenario where the transceiver is the main interference object. Thus, the power adjustment on the base station side and the dynamic adjustment of the baseline are reflected in the intensity change of the spectrum.

[0035] S2-2 Secondly, the dynamically adjusted baseline is marked in the power-adjusted spectrum, and the band of interest is given in the spectrum to match the degree The band of interest is divided into three parts, the middle part (indicated by the thick horizontal line) and the low-frequency and high-frequency parts. The former accounts for a proportion of the entire band of interest. , and the ratio of the low-frequency band range to the high-frequency band range is .

[0036] S2-3 In the middle part, this embodiment takes the frequency peak above the dynamic baseline represented by the gray dotted box as an example. In the enlarged spectrum, it will be divided into multiple equally divided band intervals (represented by rectangles in the enlarged spectrum) according to the preset step size, and a long-term and short-term memory model is established with each band interval as a unit. In the figure, unit 1, unit 2, unit n, etc. respectively represent the units corresponding to the rectangular area. After unit n, the interval can be further divided to set more intervals.

[0037] For each unit input, the corresponding training set spectrum of the corresponding band interval above the dynamic baseline is input to train the unit, and the output end outputs the prediction of whether interference exists. The units are the transmission layer of the unit status.

[0038] In this embodiment, the judgment result is that the interval where the peak tip is located is the predicted band where interference exists.

[0039] Finally, step S3 is performed, which includes the following steps: like Figure 3 As shown, S3-1 finds the corresponding time period for each band interval predicted by the LSTM as above through inverse Fourier transform based on the timestamp of the analysis spectrum, that is, the time point when the spectrum is analyzed and obtained, which is the time domain spectrum of the existence intensity in this embodiment. Thus, multiple time periods corresponding to multiple band intervals predicted to have interference are obtained based on the spectra at multiple different detection timestamps, S3-2 calls the RSRP and RSRQ data of the corresponding user side in multiple time periods, and compares them with the preset normal RSRP and RSRQ data respectively. The difference is used as the local abnormality factor. If the local abnormality factor exceeds 5-10%, it is the object of concern, and if it exceeds 10%-20%, it is suspected interference. If it exceeds 20%, it is an interference reference, and it is represented in pseudo-color on the base station-user geographic digital map. The object of concern in the figure is green, represented by √, the question mark represents suspected interference, and the exclamation mark represents interference reference, which is very likely to have interference. The triangle represents the base station, and the black dot represents the distribution of some user terminals.

[0040] In a preferred embodiment, Figure 2In the process, for the low-frequency and high-frequency parts, the absolute intensity of the spectrum is considered, and the time-integrated average of multiple intensities in a synchronous long range in multiple historical data is calculated. Thus, an additional long short-term memory unit is established, which is respectively connected to the head unit and the tail unit of the long short-term memory model, and the two units are trained as in step S2-3. Thus, the interference prediction of the high-frequency and low-frequency bands is also included.

Claims

1. The electromagnetic space security situation awareness and assessment method of the power 5G network for the access of new power systems is characterized by: The steps include: S1 Analysis of the method for acquiring electromagnetic safety data of power 5G network based on network measurement and deep learning coordination of external electromagnetic interference instrument; S2 Long-short-term memory power 5G network electromagnetic space security situation awareness based on dynamic baseline; S3 Electromagnetic space security assessment of power 5G network based on geospatial visualization of local anomaly factors.

2. The method according to claim 1, characterized in that in, S1 includes the following steps: S1-1 Select at least one network measurement data collection source and use an electromagnetic interference meter based on power port electromagnetic interference to obtain a spectrum; S1-2 uses a simulated interference source to send multiple groups of preset known simulated interference signals, and fixedly transmits simulated non-interference signals in the network, wherein the simulated interference signals are classified into true interference and false interference, wherein the true interference includes a group of predetermined standard interference signals, and multiple groups of true fluctuation groups based on the standard interference signals, and the false interference signals include multiple groups of interference signals whose differences in signal strength and signal frequency distribution from the standard interference signals and the true fluctuation groups are all outside the fluctuation threshold; S1-3 S1-1 and S1-2 collect data and spectrum respectively, build deep learning models respectively, and train the collected data and spectrum respectively to obtain the accuracy that can be achieved when the training of both is completed and , then define the degree of fit between the two .

3. The method according to claim 2, characterized in that S2 includes the following steps: S2-1 obtains the data to be analyzed through the network management system and the terminal, determines whether there is interference through signal-to-noise ratio analysis and baseline, and dynamically adjusts the power on the base station side and the baseline; S2-2 marks the frequency spectrum after power adjustment according to the dynamically adjusted baseline, giving the band of interest in the spectrum to match the degree The band of interest is divided into three parts, the middle part and the low-frequency and high-frequency parts. The former accounts for a proportion of the entire band of interest. , and the ratio of the low-frequency band range to the high-frequency band range is ; S2-3 divides the middle part into multiple equally divided band intervals according to a preset step size, establishes a long short-term memory model with each band interval as a unit, inputs the corresponding training set spectrum of the corresponding band interval above the dynamic baseline at the input end of each unit to train the unit, and outputs a prediction of whether interference exists at the output end, and the units are connected as a transmission layer of the unit status.

4. The method according to claim 3, characterized in that S3 includes the following steps: S3-1 For the band interval predicted to be interfered, find the corresponding time period based on the timestamp of the analyzed spectrum through inverse Fourier transform, thereby obtaining multiple time periods corresponding to the band intervals predicted to be interfered according to the spectra at multiple different detection timestamps, S3-2 calls the RSRP and RSRQ data of the corresponding user side in multiple time periods, and compares them with the preset normal RSRP and RSRQ data respectively. The difference is used as the local abnormality factor. If the local abnormality factor exceeds 5-10%, it is the object of concern; if it exceeds 10%-exceeding 20%, it is suspected interference; if it exceeds 20%, it is an interference reference, and it is displayed in pseudo-color on the base station-user geographic digital map.

5. The method according to claim 3 or 4, characterized in that: For the low-frequency and high-frequency parts, the absolute intensity of the spectrum is considered, and the time-integrated average of multiple intensities in a synchronized long range in multiple historical data is calculated. Thus, additional long short-term memory units are established and connected to the head unit and the tail unit of the long short-term memory model respectively, and then the two units are trained as in step S2-3.

6. The method according to claim 5, characterized in that The corresponding period is 5 minutes to 3 hours from the timestamp.

7. The method according to claim 6, characterized in that At least one network measurement includes any one or a combination of security data collection based on the LTE OAM system or network management system, security data collection based on drive test (DT), and measurement report (Measure Report) data based on statistical granularity, wherein LTE OAM works at the data link layer and reports the link status through the timed exchange of OAM PDU between devices. DT (Drive Test) drive test uses drive test equipment combined with geographic information to conduct comprehensive measurements of the air interface to verify the parameters of the wireless interface, analyze, optimize and locate faults; Measurement Report (Measure Report, MR) refers to information sent every 480ms on the service channel and 470ms on the signaling channel, and mainly includes two parts: uplink signal information and downlink signal information.

8. The method according to claim 7, characterized in that Deep learning uses convolutional neural networks to transform the collected data into binary black pixels of number 1 and white pixels of number 2 to form a two-dimensional image. The spectrum is used as an image and input into two convolutional neural networks for training.

Images