An intelligent data source collection method for autonomous classification identification and false data rejection

Abstract

The application relates to an intelligent number source collection method for autonomous classification identification and false data rejection, and belongs to the technical field of sensing. The method comprises the following steps: adopting an electric signal time-frequency domain characteristic value extraction method of a detection element to obtain time domain characteristics f(t')=f(t'+mT), m=0, plus or minus 1, plus or minus 2,..., and frequency domain characteristics F(omega')=F|f(t')|; adopting a multi-dimensional compound fusion perception method of a space environment and a state of a sensing device to acquire multi-dimensional real-time data, form a data model under normal conditions of environment parameters, and allocate proportional weights delta 1, delta 2, delta 3,..., delta n to each physical quantity; a proportional weight experience library correlation formula is calculated; the proportional weight and the severity level are set according to a preset segmentation function, the setting range is 0-1, and the correlation functions of different types of sensor devices are different; the application pays attention to the reliability of detection data of an intelligent sensor device, and improves the sensing accuracy of an intelligent sensing system device.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of sensing, and relates to an intelligent number source acquisition method for autonomous classification identification and elimination of false data. BACKGROUND

[0002] At present, the intelligent function of a sensor mainly focuses on human-computer interaction, so as to facilitate a user to use a sensing device more conveniently. For example, a wireless intelligent sensor and an implementation method of plug-and-play of the sensor, an intelligent sensor adaptive configuration method, device and system based on a signal model, etc. are all based on the credibility of sensing data to enrich the application scene of the sensor, and the effectiveness of the detection result of the intelligent sensor device itself is ignored.

[0003] Therefore, the application comprehensively utilizes a multi-dimensional composite fusion sensing method of a spatial environment and a state of a sensing device to obtain a health degree of the sensing device, compares sensing data with an experience data training model, autonomously identifies and eliminates false data in detection data of a sensitive element of the intelligent sensing device, and enables the intelligent sensor to autonomously and intelligently judge the effectiveness of sensing data. SUMMARY

[0004] Therefore, the application comprehensively utilizes a multi-dimensional composite fusion sensing method of a spatial environment and a state of a sensing device to obtain a health degree of the sensing device, compares sensing data with an experience data training model, autonomously identifies and eliminates false data in detection data of a sensitive element of the intelligent sensing device, and enables the intelligent sensor to autonomously and intelligently judge the effectiveness of sensing data.

[0005] To achieve the above object, the application provides the following technical scheme.

[0006] The weak electric number source time-frequency domain characteristic value extraction method of the detection element is adopted to obtain a time domain characteristic f(t') = f(t'+mT)

[0007] (m = 0, ±1, ±2,...), and the frequency domain feature F (ω') = F|f (t') |. The empirical data training parameter model L (ω, t) is the feature classifier function of the parameter classifier about the time (t) and frequency (ω) variables under normal conditions by the data source collector. The time domain and frequency domain data source values of the real-time data of the data source collector element are input into the parameter classifier extraction function L (ω', t'), compared with the empirical data training parameter model L (ω, t), and the comparison method adopts the ratio method. The tolerance of pseudo data (i.e. irregular data or abnormal data) is δ. If L (ω', t') / L (ω, t) ≤ 1-δ or L (ω', t') / L (ω, t) ≥ 1+δ, the current data is determined to be pseudo data, and the detection data of the sensitive element of the intelligent data source collector is classified, identified and removed from the detection data source which does not conform to the time domain frequency domain parameter classifier L (ω, t).

[0008] The change of the environment of the data source collector device has a great influence on the device. A multi-dimensional composite cross-sensing method of the space environment of the data source collector device and the state of the device is adopted to obtain multi-source, multi-dimensional (temperature T, humidity H, acceleration A, time t, installation position P, sensing element health degree S, device power supply health degree e and other sensing physical quantities) real-time parameters, and form a data parameter classifier E (T, H, A, t, P, S, e) under normal conditions of environmental parameters. Each physical quantity is allocated a proportional weight δ1, δ2, δ3... δn, and the proportional weight experience library correlation formula is calculated. The proportional weight and the severity level are set according to a preset segmented function, and the range is 0-1. The correlation classifier of different types of data source collector devices is different. When the proportional weight of the physical quantity is 0, the health degree of the data source collector is 0. Only when all the physical quantities of the parameter model physical quantity are 1, the health degree of the data source collector is 1. If the health degree of the data source collector is 0, it is determined that the data at this time point is pseudo data caused by the defect of the device itself, and is removed. If the health degree of the data source collector is not 0, the sensing result of the credibility of the sensing data of the intelligent data source collector is output.

[0009] The data source collector classifies and identifies and removes the pseudo data according to the above two classification and identification results, and outputs the sensing result of the intelligent data source collector.

[0010] The beneficial effects of the present application are that the present application focuses on the credibility of the detection data of the intelligent sensor device, and improves the sensing accuracy of the intelligent sensing system device.

[0011] Other advantages, objects and features of the present application will be explained in the following description, and will be apparent to those skilled in the art based on the following description, or can be learned from the practice of the present application. The objects and other advantages of the present application can be achieved and obtained by the following description. Attached Figure Description

[0012] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0013] Figure 1 This is a flowchart of the present invention;

[0014] Figure 2 This is a flowchart of the signal time-domain and frequency-domain analysis process;

[0015] Figure 3 This is a flowchart of the multidimensional perception data analysis process. Detailed Implementation

[0016] Please see Figure 1 , Figure 2 and Figure 3 This is an intelligent data source acquisition method that can autonomously classify, identify, and eliminate false data.

[0017] At time t', the sensor detects a time-domain signal f(t'). f(t') is then converted into a frequency-domain signal F(ω') using frequency-domain processing methods such as Discrete Fourier Transform or Wavelet Transform. The spectral characteristic frequency value ω' and spectral energy value W' for this time period are calculated, yielding the time-domain + frequency-domain extraction function L(ω',t') of the detected signal at this moment. This function is then compared with the empirical data-trained model L(ω,t) using a ratio method, with a tolerance for pseudo-data (i.e., non-standard or abnormal data) set. Let δ be the constant (e.g., 0.1 < δ < 0.5). If ω' / ω ≤ 1 - δ or ω' / ω ≥ 1 + δ, then the current data is considered to have a time-domain change frequency that exceeds the characteristics of empirical data and is therefore identified as pseudo data. If W' / W ≤ 1 - δ or W' / W ≥ 1 + δ, then the data signal's frequency-domain change amplitude is considered to be greater than the characteristics of empirical data and is therefore identified as pseudo data. This process identifies and eliminates pseudo data in the detection data of the sensitive elements of intelligent sensing devices where the detection signal does not conform to the time-domain frequency-domain model L(ω,t).

[0018] Changes in the environment in which sensing devices operate have a significant impact on the devices. A multi-dimensional composite fusion sensing method based on the spatial environment and the device's own state is adopted to acquire real-time data of multiple sources and multiple dimensions (temperature T, humidity H, acceleration A, time t, installation location P, sensing element health S, device power supply health e, etc.) to form a data model E(T,H,A,t,P,S,e) of environmental parameters under normal conditions. Each physical quantity is assigned a proportional weight δ1, δ2, δ3...δn, which is calculated using an empirical formula. The proportional weight and severity level are set according to a preset piecewise function, with a setting range of 0 to 1. The correlation function is different for different types of sensor devices.

[0019] Assuming that a methane sensor at the current coal mining face has a detection data of 0.8% CH4 for one frame, the environmental data is temperature 15°C, relative humidity 90% RH, acceleration 9.8g, time 19:00, installation position coal mining face, sensing element health degree 0.6, and equipment power health degree 1. According to the multi-dimensional perception data analysis flowchart and the set physical quantity weight parameters, the temperature δ1=1, humidity δ2=0.5, acceleration δ3=0, time δ4=1, installation position δ5=0.9, element health degree δ6=0.6, and power health degree δ7=1 are obtained. The calculation equipment comprehensive health degree δ=0.7. Therefore, due to the current high environmental humidity and low element life health degree, the current data is 0.7 detection data. However, according to the judgment logic that the proportion weight of physical quantity is 0, the health degree of sensor equipment is 0; only when all physical quantities of all model physical quantities are 1, the health degree of sensor is 1, it is judged that because the weight of equipment deceleration becomes 0, the detection data of the equipment has a reliability of 0, and is determined as false data. By using the multi-dimensional complex fusion perception method of space environment, equipment itself and the like, the health status of the methane sensor is obtained, the abnormal working state of the sensor in the time period is found, and the equipment health degree is judged as “severe defect level”.

Claims

1. An intelligent data source acquisition method for autonomous classification, identification, and removal of false data, characterized in that: The method includes: The time-domain and frequency-domain feature extraction methods of the electrical signal of the detection element are used to obtain the time-domain feature f(t')=f(t'+mT), m=0,±1,±2,……, and the frequency-domain feature F(ω')=F|f(t')|; The empirical data training model L(ω,t) is a feature extraction function of the dataset from the sensing device under normal conditions, with respect to the time (t) and frequency (ω) variables. The time-domain and frequency-domain signal values ​​of the real-time data of the sensor element are input into the model extraction function L(ω',t') and compared with the empirical data training model L(ω,t). The comparison method adopts the ratio method and sets the pseudo-data tolerance to δ. If L(ω',t') / L(ω,t)≤1-δ or L(ω',t') / L(ω,t)≥1+δ, the current data is identified as pseudo-data. Pseudo-data in the detection data of the sensitive element of the intelligent sensing device that does not conform to the time-domain and frequency-domain model L(ω,t) is identified and removed. A multi-dimensional composite fusion sensing method based on the spatial environment and the state of the sensing device is adopted to acquire multi-dimensional real-time data and form a data model of environmental parameters under normal conditions. Each physical quantity is assigned a proportional weight δ1, δ2, δ3...δn. The proportional weight is calculated using an empirical formula. The proportional weight and severity level are set according to a preset piecewise function, with a setting range of 0 to 1. The correlation function is different for different types of sensor devices. If the weight of the physical quantity is 0, then the health of the sensor device is 0. If all physical quantities of all model physical quantities are 1, then the sensor health is 1; If the sensor's health status is 0, the data at that time point is determined to be false data caused by its own condition defects and is therefore discarded. If the sensor health is not 0, the output will be the perception result of the credibility of the accompanying perception data of the intelligent sensing device. Based on the above two identification results, the sensing device identifies and eliminates false data and outputs the perception results of the intelligent sensing device.

2. The intelligent data source acquisition method for autonomous classification, identification, and removal of false data according to claim 1, characterized in that: The pseudo data refers to unconventional or abnormal data.

3. The intelligent data source acquisition method for autonomous classification, identification, and removal of false data according to claim 1, characterized in that: The multidimensional real-time data includes temperature T, humidity H, acceleration A, time t, installation location P, sensor health S, and device power health e. The data model is E(T,H,A,t,P,S,e).

Images